Original Link: http://www.anandtech.com/show/5541/amd-radeon-hd-7750-radeon-hd-7770-ghz-edition-review
AMD Radeon HD 7750 & Radeon HD 7770 GHz Edition Review: Evading The Price/Performance Curveby Ryan Smith & Ganesh T S on February 15, 2012 12:01 AM EST
With the launch of Tahiti behind them, AMD is now firing on all cylinders to get the rest of their Southern Islands lineup out the door. Typically we’d see AMD launch their GPUs in descending order of performance, but this time AMD is taking a slight detour. Rather than following up the Tahiti based 7900 series with the Pitcairn based 7800 series, AMD is instead going straight to the bottom and launching the Cape Verde based 7700 series first.
Today AMD will be launching two cards based on the Cape Verde GPU: the Radeon HD 7750, and the Radeon HD 7770 GHz Edition. As the Juniper based 5700 series never got a proper Northern Islands successor, this is the first real update for the x700 series since the launch of the 5700 series in October of 2009. Given the success of the 5700 expectations are going to be high, and to fulfill those expectations AMD will be bringing to bear their new GCN architecture along with a full node jump with TSMC’s 28nm process. But will this be enough to enable the 7700 series to replicate the success of the 5700 series? Let’s find out.
|AMD GPU Specification Comparison|
|AMD Radeon HD 7770 GHz Edition||AMD Radeon HD 7750||AMD Radeon HD 6850||AMD Radeon HD 5770||AMD Radeon HD 5750|
|Memory Clock||4.5GHz GDDR5||4.5GHz GDDR5||4.8GHz GDDR5||4.8GHz GDDR5||4.6GHz GDDR5|
|Memory Bus Width||128-bit||128-bit||256-bit||128-bit||128-bit|
|Manufacturing Process||TSMC 28nm||TSMC 28nm||TSMC 40nm||TSMC 40nm||TSMC 40nm|
With the 6000 series AMD effectively had a 5 chip stack: Caicos (6400), Turks (6600), the rebadged Juniper (6700), Barts (6800), and Cayman (6900). Since then integrated GPUs have effectively wiped out the low end of the market, and by the time Trinity launches later this year any product short of 6600 performance should be made redundant. As a result AMD has reigned in on their spread out stacks, leading to their 28nm Southern Islands product stack being just 3 products: Cape Verde (7700), Pitcairn (7800), and Tahiti (7900). Anything below the 7700 series will be rebadged Northern Islands parts, primarily Turks and Caicos.
But AMD’s product stack doesn’t tell the whole story. AMD will need to cover a wide range of products and price levels with only 3 GPUs, ranging from $109 to $550. As a result the performance levels of AMD’s various product series are being redefined somewhat, and nowhere is this more apparent right now than with the 7700 series. Why do we say that? Well let’s take a look at the specs and pricing.
Cape Verde, the GPU at the heart of the 7700 series, is AMD’s smallest 28nm GPU. With a die size of 123mm2 it’s only a hair bigger than the 118mm2 Turks GPU that powers the 6600 series. In terms of functional units we’re looking at 10 Compute Units, giving Cape Verde 640 SPs and 40 texture units. Elsewhere Cape Verde packs 16 ROPs, 1 geometry engine/rasterizer pair, and 512KB of L2 cache, with the chip coupled to a 128bit GDDR5 memory bus. Altogether compared to Tahiti this is around 31% of the CUs, 33% of the memory bus width, and half as many ROPs.
Interestingly, unlike Tahiti, Cape Verde’s CUs are organized slightly differently. GCN is designed around 4 CUs in each CU Array, with the 4 CUs sharing a read-only L1 instruction cache and a read-only L1 data cache. This is how both the 7970 and 7950 are organized, with the 7950 simply lopping off a whole CU array. However with Cape Verde 10 CUs doesn’t cleanly divide into groups of 4, so for the first time AMD has built something a bit different. In Cape Verde there are 3 CU arrays, populated in a 4/3/3 manner. With regards to performance there shouldn’t be a huge difference, but this does mean that there’s a bit less cache pressure on the CUs occupying the smaller CU arrays.
On the functionality side of things, backing up Cape Verde will be the rest of the Southern Islands family features we’ve already seen on Tahiti, such as DX10+ SSAA, PowerTune, Fast HDMI support, partially resident textures, D3D 11.1 support, and the still-AWOL Video Codec Engine (VCE). Even FP64 support is accounted for, however similar to how NVIDIA handles it on lower-end parts it’s a performance-limited implementation for compatibility and software development purposes, with FP64 performance limited to 1/16th FP32 performance.
As for the cards themselves, AMD will be releasing two Cape Verde cards: the Radeon HD 7750, and the Radeon HD 7770 GHz Edition. The 7770 will be a fully enabled Cape Verde with all 10 CUs (640 SPs) enabled, running at 1GHz for the core clock and 4.5GHz for the memory clock, with a PowerTune limit of 100W and an AMD defined typical board power of 80W. The 7750 meanwhile has 2 disabled CUs, giving it 512 SPs and 32 texture units, while the ROP count is unchanged at 16. The core clock will be 800MHz while the memory clock is the same as the 7770 at 4.5GHz, with a PowerTune limit of 75W and a typical board power of 55W. Both cards have a sub-10W idle TDP, while long-idle is sub-3W.
With those specs in mind, it’s interesting to note that the difference between the 7750 and 7770 is much wider than we’ve seen in past products. Compared to the 7750, the 7770 has 25% more CUs and a 25% core clock advantage, giving it a massive 56% shader and texture performance advantage over the 7750. With the 5700 series this gap was only 35%, and most of that was a result of core clock differences. The fact that the memory bandwidth is the same between the 7770 and 7750 equalizes things somewhat, but it’s still a huge difference for two cards that are in the same family.
This brings us back to where we started: how AMD is covering the entire $109+ market with only 3 GPUs. Between the massive performance gap between the 7700 series cards and the fact that the 7750 is a sub-75W part, it becomes increasingly obvious that the 7700 series is the de-facto replacement for both the 6600 series and the 5700 series. The 7750 will fill the 6670’s old role as AMD’s top sub-75W card, but as we’ll see its performance means it won’t be a complete replacement for the 5700 series. Instead the role of replacing the 5700 series falls to the much more powerful 7770.
As for today’s launch, AMD will be launching the 7750 at $109 and the 7770 at $159, which happens to be the same prices the 5770 and 5750 512MB launched at respectively. With midrange cards there usually aren’t any supply issues and we aren’t expecting the 7700 series launch to be any different, however as is customary AMD’s partners will be launching semi-custom cards from day one, so pricing will probably be inconsistent.
At these prices the 7700 series will be competing with a number of last-generation cards. The 7750 will be up against AMD’s 5770/6770 and the absolute cheapest of NVIDIA’s GTX 550 Ti cards. Meanwhile the 7770 will be competing with AMD’s 6850 and 6870, along with NVIDIA’s GTX 460 and their cheapest GTX 560s. And as we’ll see in our benchmarks, this is ultimately going to be more than the 7700 series can handle.
|Winter 2011/2012 GPU Pricing Comparison|
|$209||GeForce GTX 560 Ti|
|$179||GeForce GTX 560|
|Radeon HD 6870||$169|
|Radeon HD 7770||$159|
|Radeon HD 6850||$139|
|$119||GeForce GTX 550 Ti|
|Radeon HD 7750||$109|
|Radeon HD 6770||$99|
Meet the Radeon HD 7750
We’ll kick things off as always with a look at the cards themselves, starting with the Radeon HD 7750. As we alluded to before, this is the de-facto replacement for the Radeon HD 6670, and you only have to take one look at the card to understand why.
AMD’s reference design for the 7750 is virtually identical to the full-profile 6670 or the FirePro V4900, which should come as no surprise given that all of these cards are or were AMD’s top sub-75W cards in their respective markets. As a result, like those cards the reference 7750 is a full-profile card featuring a single-wide active cooler.
As the 7750 is AMD’s cheapest Southern Islands card, you won’t find much else on the card to speak of. As a sub-75W card it doesn’t need external power, and cementing its position as the replacement for the 6670 there isn’t a CrossFire connector on the card. For RAM the card uses 4 256MB Hynix GDDR5 RAM chips, which are rated for 5GHz. The card is 6.57” long overall, the same length as the 6670.
Meanwhile for display connectivity, AMD is once again using the same configuration as we’ve seen in their other full-profile mainstream cards. This means 1 DL-DVI port, 1 HDMI port, and interestingly enough 1 full size DisplayPort. The latter is particularly odd, as the rest of the Southern Islands lineup is exclusively miniDP and in the last year miniDP has become the de-facto port for source devices. AMD has told us that there’s no specific reason that they’re using a full size DisplayPort here, and we believe it’s largely being done out of maintaining consistency with previous products. With that said we’d rather see miniDP here – even if it’s just 1 port instead of 2 – so that it’s consistent with the rest of the 7000 series.
Finally, as is customary for a midrange product launch, everyone is doing semi-custom cards right off the bat. Everyone will be using AMD’s PCB for now, while none of the 7750 cards in the press materials sent to us will be using AMD’s cooler. Instead we’ll see a range of designs, from similar side-wide designs to the more common double-wide designs, and even a passively cooled design from Sapphire. Much like the 6670 the HTPC use case for the 7750 is rather obvious, so we suspect that we’ll see more passive and perhaps even some low-profile cards in the future.
Meet the Radeon HD 7770 GHz Edition
The second card launching today is the Radeon HD 7770 GHz Edition. Compared to the 7750 it’s based on a fully enabled Cape Verde GPU and features a higher core clock of 1GHz, versus 800MHz on the 7750.
Starting as always with the cooler, for the reference 7770 AMD has gone with something that is best described as half of a blower. The shroud is completely enclosed on the sides, but due to the position of the fan the card exhausts hot air out of both the front and the rear of the card, which is something that’s common for dual-GPU cards such as the 6990 but atypical for a midrange video card. Heat transfer is provided by a basic black aluminum heatsink, while the fan is embedded in the middle of the heatsink.
As for the card itself, it’s effectively an extended version of the 7750. Like the 7750 the card is equipped with 4 256MB 5GHz Hynix GDDR5 RAM chips, along with some additional VRM circuitry to handle the higher 100W power limit for this board. External power is provided by a single 6pin PCIe power socket, while at the top of the card there is a single CrossFire connector. This is actually down from 2 connectors on the 5700 series, which comes as no great surprise as virtually no one ever used tri-CF with the 5700 series in the first place. Overall the card is 8.25” long with no notable shroud overhang, making it the same length as the 5770’s PCB.
Meanwhile for display connectivity AMD is using the same configuration as we’ve seen on the 7900 series: 1 DL-DVI port, 1 HDMI port, and 2 miniDP ports. Unlike the 7900 series AMD’s not requiring their partners to include any adaptors, so buyers with 2 DVI monitors will almost always be on their own. The presence of 2 miniDP ports means that the 7770 has quite a bit of flexibility in driving various displays, however as with the 7900 series if you want to drive more than 4 displays you will need a MST hub, the release of which is still some time off.
As for partner cards, as with the 7750 all of AMD’s partners are doing their own thing. No one will be using AMD’s reference cooler, though PowerColor in particular will be using a cooler similar to it. Everyone else will be using their own double-wide coolers, most of which will be open air as we typically see in this market segment. Like the 7900 series AMD is relying on the 7770 having some degree of overclockability, so there will be a number of factory overclocked cards taking advantage of this.
Finally, I wanted to quickly touch on the naming of the 7770. The official name of the 7770 is the Radeon HD 7770 GHz Edition; with the last bit being AMD’s latest marketing push. Starting with the 7770, AMD is going to be branding every card with a reference clock at or above 1GHz as a “GHz Edition” card to capitalize on the fact that they’ve hit 1GHz. Thus in spite of what the name implies there won’t be a 7770 non-GHz Edition card, as it’s a tagline rather than a true suffix necessary to differentiate cards.
With that said, while hitting 1GHz on a GPU is a notable accomplishment for AMD and should not be ignored, it’s not going to be particularly important in the grand scheme of things. The embarrassingly parallel nature of rendering means that GPU performance isn’t nearly as tied to clockspeeds as CPU performance is, as you can always lay down more functional units to improve performance. Clockspeeds do need to go up over time due to the limited parallelization of the command processor, but otherwise GPU performance has never been heavily dependent on clockspeeds.
Meet the XFX R7770 Black Edition S Double Dissipation
Our final card of the day is XFX’s R7770 Black Edition S Double Dissipation, one of a multitude of Radeon HD 7770 cards XFX will be releasing today. XFX will be releasing two factory overclocked Black Edition cards today, with the S edition BESDD being the higher clocked of the two. It will be clocked at 1120MHz core and 5.2GHz memory, which is a 120MHz (12%) core overclock and 700MHz (15%) memory overclock compared to the reference 7770.
Like all the other 7770s being released today XFX is using the AMD reference PCB for the BESDD, so XFX is primarily relying on their factory overclock and their custom cooler to stand apart from AMD’s other partners. To that end the BESDD is very similar to the other XFX Double Dissipation cards we’ve reviewed in the last month. A copper plate sits at the base of the cooler, transferring heat to an aluminum heatsink that runs virtually the entire length of the card. Above that sits a pair of fans and then finally the metal shroud that covers the card.
Overall the design is similar to a number of other dual fan cards we’ve seen in the past, and as we’ll see in our benchmarks the resulting cooling is quite effective. In terms of quality however I feel XFX has missed the mark – the cooler is only attached to the PCB around the GPU, so at 8.25” long there’s nothing retaining the PCB at the far end to keep it from flexing. A stiffener would have been a great idea here, which is something XFX does on the 7900 series. Furthermore in the case of our sample the construction quality was also subpar, as the screws responsible for securing the metal shroud were not properly tightened and thereby leaving the shroud loose. This is nothing 30 seconds with a screwdriver couldn’t fix and I don’t have any reason to believe this is a common problem, but it’s a quality control issue none the less.
As the BESDD uses AMD’s 7770 PCB, the length of the card and display connectivity is identical. The card measures 8.25” long with no cooler overhang, while display connectivity is provided by 1 DL-DVI port, 1 HDMI port, and a pair of miniDP ports.
Rounding out the package is the same collection of extras that we saw in the 7950 BEDD. Inside you’ll find the usual driver CD and quick start guide, along with a metal XFX case badge, a mid-length CrossFire bridge, a DVI to VGA adaptor, and a passive HDMI to SL-DVI adaptor. All of this is packed in one of XFX’s pleasantly small boxes, which doesn’t use much more space than the card itself.
The MSRP on the R7770 BESDD is $184, $25 over the MSRP for a regular 7700 and pushing it well into the territory of the Radeon HD 6870 and GeForce GTX 560. XFX is offering a base 2 year warranty on the R7770 BESDD, which can be extended to a lifetime warranty by registering the card within 30 days of purchasing it.
HTPC Testbed & Software Configuration
Given its position as a sub-75W card, the Radeon HD 7750 has an obvious niche in HTPCs. In order to test out the HTPC capabilities of the 7750, we decided to reuse our existing AMD Llano testbed. 3D is an important aspect, and the AVR / display device have since been updated for this purpose. The table below lists the components in our Llano HTPC testbed.
|AMD Llano HTPC Testbed Setup|
|Processor||AMD A8-3850 - 2.90GHz, 4MB Cache (1MB/core)|
|Motherboard||ASRock A75Pro4 ATX|
|Disk Drives||OCZ Vertex 2 120 GB (OS) / 1TB Samsung HD103SJ (Media Storage)|
|Memory||G.Skill Ripjaws Series 4GB (2 x 2GB) SDRAM DDR3 1333 (PC3 10666) F3-10666CL7D-4GBRH CAS 7-7-7-21|
|Video Cards||AMD GDDR5 7750|
|Optical Drives||ASUS 8X Blu-ray Drive Model BC-08B1ST|
|Case||Antec Skeleton ATX Open Air Case|
|Power Supply||Antec VP-450 450W ATX|
|Operating System||Windows 7 Ultimate x64|
|Display / AVR||Acer H243H / Pioneer Elite VSX-32 + Sony Bravia KDL46EX720|
Hardware is only one part of the HTPC equation. The various software components used in our testing are tabulated below.
|AMD 7750 HTPC Testbed Software Configuration|
|Blu-ray Player||CyberLink PowerDVD 12|
|Standalone Media Player||MPC-HC x86 v22.214.171.12414|
|Splitter||LAV Splitter (LAV Filters 0.46)|
|Audio Decoder||LAV Audio Decoder (LAV Filters 0.46)|
|Video Decoders||LAV Video Decoder (LAV Filters 0.46)|
|MPC Video Decoder|
|Renderers||EVR Custom Presenter|
|Notes||LAV Audio Decoder was tested in both decode and passthrough modes|
|LAV Video Decoder was primarily used in the DXVA2 Copy-Back mode|
|The MPC Video Decoder and EVR-CP renderer were primarily used to ensure that legacy decoding methods were still effective|
Our first step was to put the 7750 through the HQV benchmarking process.
HQV 2.0 Benchmarking
HTPC enthusiasts are often concerned about the quality of pictures output by the system. While this is a very subjective metric, we have decided to take as much of an objective approach as possible. Over the last year or so, we have been using the HQV 2.0 benchmark for this purpose.
The HQV 2.0 test suite consists of 39 different streams divided into 4 different classes. The playback device is assigned scores for each, depending on how well it plays the stream. Each test was repeated multiple times to ensure that the correct score was assigned. The scoring details are available in the testing guide on the HQV website.
We first played back the benchmarking clips off the Blu-ray, but the scores recorded below are for standalone M2TS playback using CyberLink PowerDVD 12. Our take is that it is standalone files (such as those taken using consumer camcorders and smartphones) which require video post processing more than the already carefully transferred Blu-ray content. If certain driver post processing features are available only for Blu-ray playback, then, it is as good as not being available for consumers to take advantage.
In the table below, we indicate the maximum score possible for each test, and how much the 7750 was able to get with AMD's Catalyst 12.1-based launch drivers.
|AMD 7750 : HQV 2.0 Benchmark|
|Test Class||Chapter||Tests||Max. Score||Radeon HD 7750|
|Video Conversion||Video Resolution||Dial||5||5|
|Dial with Static Pattern||5||5|
|Film Resolution||Stadium 2:2||5||5|
|Overlay On Film||Horizontal Text Scroll||5||5|
|Vertical Text Scroll||5||5|
|Cadence Response Time||Transition to 3:2 Lock||5||5|
|Transition to 2:2 Lock||5||5|
|Multi-Cadence||2:2:2:4 24 FPS DVCam Video||5||5|
|2:3:3:2 24 FPS DVCam Video||5||5|
|3:2:3:2:2 24 FPS Vari-Speed||5||5|
|5:5 12 FPS Animation||5||5|
|6:4 12 FPS Animation||5||5|
|8:7 8 FPS Animation||5||5|
|Color Upsampling Errors||Interlace Chroma Problem (ICP)||5||2|
|Chroma Upsampling Error (CUE)||5||2|
|Noise and Artifact Reduction||Random Noise||SailBoat||5||5|
|Compression Artifacts||Scrolling Text||5||5|
|Upscaled Compression Artifacts||Text Pattern||5||3|
|Image Scaling and Enhancements||Scaling and Filtering||Luminance Frequency Bands||5||5|
|Chrominance Frequency Bands||5||5|
|Resolution Enhancement||Brook, Mountain, Flower, Hair, Wood||15||15|
|Video Conversion||Contrast Enhancement||Theme Park||5||5|
|Beach at Dusk||5||5|
|White and Black Cats||5||5|
|Skin Tone Correction||Skin Tones||10||7|
We were not able to match what AMD claimed. In all probability, the difference is in the chroma upsampling scores. We already covered this in the Llano review. To recap, the image below is a screen capture while playing the HQV benchmark clip from a Blu-ray ISO on the AMD A8-3850. Note that the light red fringes along the edgs are uniform and extend to the right and the left before tapering off.
Now, let us look at the screen capture when the M2TS is played back using PowerDVD 12 on the Radeon HD 7750.
Zoom into the red stripes on the third row (in the original screenshot) and you can see for yourself that we have some ghost samples extending above and below the actual line of pixels. The screenshot is not at the exact timestamp as that of the Llano sample reproduced above, but the reader should be able to get a general idea of the problem we are hinting at. Readers interested in finding out more about the source of this problem should read up this excellent piece to get more information.
We notified AMD about this issue during the Llano review, but action doesn't seem to have been taken even after 6 months on what is clearly a driver issue. Note that we are not using MPC-HC (which might introduce its own errors in the renderer) in this case, but a standard off-the-shelf commercial player in CyberLink PowerDVD 12. Hopefully, AMD wakes up to this issue soon.
Custom Refresh Rates with AMD 7750
One of the drawbacks of the GPUs built into the Intel CPUs was the lack of a 23.976 Hz refresh rate to match the source frame rate of many videos. Combined with the lack of reliable support for open source software, this has often pushed users to opt for a discrete HTPC GPU. Ideally, a GPU should be capable of the following refresh rates at the minimum:
- 23.976 Hz
- 24 Hz
- 25 Hz
- 29.97 Hz
- 30 Hz
- 50 Hz
- 59.94 Hz
- 60 Hz
Some users demand integral multiples of 23.976 / 24 Hz because they result in a smoother desktop experience, while also making sure that the source and display refresh rates are still matched without repeated or dropped frames.
With the Sony Bravia KDL46EX720 (the new display in our HTPC testbed) being capable of PAL refresh rates (despite not exposing it in the EDID), we were able to test out almost all the above mentioned refresh rates. The only exception was 60 Hz (which definitely works well in the usual desktop mode, but, was one for which we were unable to obtain a sample file with a matching frame rate).
Using madVR, we were able to judge how often a frame would get repeated or dropped in the renderer due to the mismatched refresh rate and frame rate. With the 23.976 fps sample, we were looking at more than 15 minutes before a dropped frame, while, for the 24 fps sample, we were looking at more than a hour before a frame repeat.
By default, modes not reported in the EDID were not available in the CCC settings. However, it was quite straightforward to display and choose the refresh rates not supported by the monitor in its Windows settings. Some of the other 'native refresh rates' we tested are reproduced below:
AMD's refresh rate matching feature works pretty well most of the time, but it could be made better by giving the user more control over the various timing aspects (maybe in an 'Advanced' hidden menu similar to what is done by NVIDIA). This could be useful for consumers who don't want to put up with frame drops / repeats even once every 20 minutes or so.
Video Post-Processing: GPU Loading
We saw in our coverage of discrete HTPC GPUs last year that noise reduction loaded up the GPU, and as such, was even disabled in the low end GPUs for want of shader resources. Starting with this review, we are planning to tabulate GPU usage under various post processing scenarios instead of running decoder benchmarks. GPU-Z gives us the necessary data for this purpose.
A 1080i60 H.264 clip (same as the one used in the discrete HTPC GPU article last year) was decoded with the LAV Video Decoder (DXVA2 Copy-Back mode) using EVR-CP as a renderer in GraphStudio Next's Decoder Performance section. Various post-processing options were turned on and off in CCC and the GPU usage recorded in each case.
|Video Post Processing GPU Usage: 1080i60 H.264|
|AMD Radeon HD 7750 (1GB GDDR5) / Catalyst 12.1|
|LAV Video Decoder DXVA2 (Copy-Back) v0.46 / EVR-CP|
|Post Processing Algorithm||GPU Load|
|No Video Post Processing||19%|
|Vector Adaptive Deinterlacing + Pulldown Detection||25%|
|Dynamic Contrast and Colour||25%|
|All Post Processing / 'Enforce Smooth Video Playback' Disabled||62%|
We also put some Full SBS / Full TAB 3D clips (which are basically 2 x HD resolution) through the same process. Those progressive clips resulted in around 70% of the GPU being loaded with all the post processing steps enabled.
It is not yet possible to use the madVR renderer from within GraphStudio Next, but, in future HTPC / HTPC GPU reviews, you can expect to find similar benchmarking with the madVR renderer (now that it is possible to use madVR along with hardware accelerated decode for AMD GPUs also). That said, we see that up to 62% of the GPU is loaded using just EVR-CP. It is not clear how much room is left for madVR processing, and we hope to address that question in future reviews.
Miscellaneous HTPC Aspects
One of the nice aspects of the Radeon HD 7750 is the fact that AMD's excellent video post processing capabilities with respect to deinterlacing, cadence detection and noise reduction are carried over from the previous efforts without the introduction of any bugs. As such, deinterlacing is of the same quality as before, and we felt that there was no necessity to repeat screenshots very similar to what we already provided in our previous Llano HTPC review.
3D works very well, and is even more seamless compared to NVIDIA's implementation. I don't play 3D games, and my only interest from a HTPC perspective is playing back 3D Blu-rays. I found that simply clicking on the 3D icon in PowerDVD shifted my VSC-32 / Sony KDL46EX720 into 3D mode. There was no need to explicitly set up the 3D display as I had to do with the NVIDIA cards. This might be a drawback for people doing 3D gaming, but for 3D media watching this is as simple as it could be.
It is not that the 7750 is without its faults. For all practical open source software purposes, MPEG-4 decode acceleration is absent even though it is a feature of UVD3. The Catalyst 10.4 release notes promised support for H.264 L5.1 stream decoding. However, consumers soon discovered that enabling DXVA decode for 4K clips often ended up in a BSOD. AMD has quietly slipped this under the radar, and now officially states that 4K decode is not officially supported for the time being, however this appears to be a matter of validation rather than hardware limitations. That said, we did see that trying to decode a 4K clip now no longer results in a hard BSOD.
The 7750 also has support for HDMI 1.4a's full specifications. This means that the GPU can drive resolutions of up to 4096x2160 at 24 fps and 3840x2160 at 30 fps over a single HDMI port! I am currently aware of only one HDMI sink supporting this over a single HDMI link, namely, the Sony VPL-VW1000ES projector. Users on AVSForum are already reporting success with driving 4K over a single HDMI link using the Radeon HD 7970, and I expect the 7750 to have no issues either. That said, if we do get access to this projector system, the 7750 will be one of the first HDMI sources to get connected to it.
I recently set up a 2x2 Eyefinity system using the 7950 to drive QFHD videos onto the displays. I was very impressed with the quality and ease of setup. Frankly, I am more excited about 4K compared to what I felt about 3D when manufacturers were trying to push that down the throat of the consumers. In my opinion, 4K (QFHD) with 2x2 23" 1080p thin bezel monitors will become a very cost effective solution for those looking at 4K for the desktop. In that respect, it is a bit disappointing that the 7750 we tested today can't drive four displays without a DisplayPort MST hub.
It is a little bit interesting to compare the GT 520 with the AMD 7750 with respect to readiness for 4K. While the GT 520 has full hardware decode acceleration for 4K videos, it is unable to push out the 4K material to the display(s). The HDMI 1.4a PHY in the GT 520 can drive only 1080p monitors and there is no way to drive four displays with it. The 7750, on the other hand, can drive 4K displays through HDMI right now (and to four monitors using an MST hub down the road), but it is unable to accelerate the decode of those videos. It will be interesting to see what NVIDIA has in store for the HTPC fans down the road. Can they deliver working cards and drivers before AMD fixes its driver issues? It is going to be a very interesting year ahead.
As a summary for our HTPC section, we have to say that the Radeon HD 7750 is an excellent addition to our HTPC testbed. It will definitely be the one to compare against when the new cards from NVIDIA and Intel's Ivy Bridge CPU come out over the next few months. We just hope that AMD will be able to get its driver act together before then.
VCE & The Test
Although it’s not unheard of for a video card to launch without all of its features enabled in its drivers, it’s rare to see a video card go for more than a couple of weeks without support for a major feature. AMD’s Video Codec Engine (VCE) – their fixed function H.264 encoder – is still not enabled as of the 7700 series launch. For the 7900 series this wasn’t such a big deal as high-end cards are typically paired with high-end CPUs, but the fact that it’s not enabled for the 7700 series launch is a bit more unsettling. These are cards that are going to be paired with slower CPUs, where having a high speed H.264 encoder is going to be all the more important. This is particularly so for the 7750, given its obvious aspirations for HTPC usage.
AMD was supposed to provide an official statement on the matter for the 7700 launch, however as of press time they have yet to do so. At this point we can only speculate what the holdup is, but regardless it’s unfortunate for AMD.
Moving on, for the launch of the 7700 series AMD provided us with a new driver, which identifies itself as 8.932.2. This driver only works with Cape Verde, and while we’re unable to test it with other cards we’re told there are no performance differences between it and the 7900 series drivers AMD released last month.
|CPU:||Intel Core i7-3960X @ 4.3GHz|
|Motherboard:||EVGA X79 SLI|
|Chipset Drivers:||Intel 126.96.36.1992|
|Power Supply:||Antec True Power Quattro 1200|
|Hard Disk:||Samsung 470 (256GB)|
|Memory:||G.Skill Ripjaws DDR3-1867 4 x 4GB (8-10-9-26)|
|Case:||Thermaltake Spedo Advance|
AMD Radeon HD 7970
AMD Radeon HD 7950
AMD Radeon HD 7770
AMD Radeon HD 7750
AMD Radeon HD 6870
AMD Radeon HD 6850
AMD Radeon HD 5770
AMD Radeon HD 5750
NVIDIA GeForce GTX 560
NVIDIA GeForce GTX 550 Ti
NVIDIA GeForce GTX 460 1GB
NVIDIA GeForce GTX 285
NVIDIA ForceWare 295.51 Beta
AMD Catalyst Beta 8.921.2-120119a
AMD Catalyst Beta 8.932.2
|OS:||Windows 7 Ultimate 64-bit|
Kicking things off as always is Crysis: Warhead. It’s no longer the toughest game in our benchmark suite, but it’s still a technically complex game that has proven to be a very consistent benchmark. Thus even four years since the release of the original Crysis, “but can it run Crysis?” is still an important question, and the answer continues to be “no.” While we’re closer than ever, full Enthusiast settings at a 60fps is still beyond the grasp of a single-GPU card.
With AMD’s GCN architecture, Crysis is finally losing some of its predictive power, but it’s still a good indicator of overall performance. And the indicators aren’t very good.
From a price/performance perspective the 7700 needs to beat at least the 6850 on the AMD side and at least the GTX 560 on the NVIDIA side in order to ensure a clear-cut victory. Unfortunately it’s falling well short of that, underperforming the 6850 by 20% never mind the GTX 560. On paper the 7770 is outgunned on everything from texture performance to ROP performance to shader performance, so take your pick on what’s holding back Crysis performance, but the end result is that it’s well behind its previous-generation peers. In fact it’s closer to the 2.5 year old 5770 than it is the 6850.
As for the 7750, things are a bit better. In spite of its massive texture and shader deficit compared to the 7770, it’s only behind by 15%, offering some strong support that the actual performance difference between the two cards won’t be nearly as big as it is on paper. That said, on a price/performance basis it should be able to beat the 5770, and it isn’t quite there. But as the only card here that’s sub -75W, it has other favorable attributes.
Elsewhere XFX’s R7770 BESDD shows us that 7770 performance scales well with clockspeeds, with a 12% core overclock netting 11% better performance. This is a good sign for overclockers, but for factory overclocked cards it’s a mixed bag, as the higher price these cards fetch only puts them more in competition with the 6870.
The story of minimum framerates is much the same. The 7700 series continues to underperform its last-generation competition by quite a bit, and unfortunately we’re still several frames per second away from being able to sustain 30fps+ in Crysis at 1680.
Paired with Crysis as our second behemoth FPS is Metro: 2033. Metro gives up Crysis’ lush tropics and frozen wastelands for an underground experience, but even underground it can be quite brutal on GPUs, which is why it’s also our new benchmark of choice for looking at power/temperature/noise during a game. If its sequel due this year is anywhere near as GPU intensive then a single GPU may not be enough to run the game with every quality feature turned up.
Things get better for AMD under Metro. Metro is extremely shader bound, which helps prop up performance in some areas and drives it farther apart in others. For the 7770, this means it still trails the 6850, but now only by 15%. Meanwhile XFX’s factory overclock helps to close that gap.
Given the shader-heavy nature of Metro, it’s interesting to note just how well the 7770 is doing versus the 5770, even though the 7770 only has 94% of the theoretical shader performance. Although the launch of the 7900 series didn’t give us a lot of data to work with regarding the efficiency of GCN shaders in games, the 7700 provides the first piece of solid data that GCN shaders are actually more efficient than VLIW5 shaders in at least some cases. Considering the die space penalty for implementing GCN functionality, this is very good news for AMD.
Of course as Metro is so shader heavy, it means it drives a wedge between the 7770 and 7750. The 7770 is now 21% ahead of the 7750, which compared to the 45% price premium is quite good, but it means the 7750 won’t be performing so close to the 7770 all of the time. And much like Crysis it’s slightly trailing the 5770.
For racing games our racer of choice continues to be DiRT, which is now in its 3rd iteration. Codemasters uses the same EGO engine between its DiRT, F1, and GRID series, so the performance of EGO has been relevant for a number of racing games over the years.
With DiRT 3 AMD finally turns the tables on themselves, leading to the 7770 taking a lead over the 6850. With an 8% lead this is the first win for the 7770, and a surprising one at that. Throw in a factory overclock like the BESDD and even the 6870 can be surpassed. In fact the only real bad news for the 7770 here is that while AMD is beating themselves, it’s a competition that NVIDIA normally wins, leading to the 7770 still trailing the 1GB GTX 460 and sitting well behind the GTX 560. The interplay of three different architectures here means that no one is necessarily going to be the winner all of the time.
As for the 7750, the 7770 gap only continues to grow. Now the 7770 is 32% ahead of the 7750, indicating just how important shader and texturing performance is in this game. The end result is that the 7750 still slightly trails the 5770.
As good as overall DiRT 3 performance is, the minimum framerates look even better. The 7750 finally edges out the 5770, meanwhile the 7770 nearly ties the 6870.
Total War: Shogun 2
Total War: Shogun 2 is the latest installment of the long-running Total War series of turn based strategy games, and alongside Civilization V is notable for just how many units it can put on a screen at once. As it also turns out, it’s the single most punishing game in our benchmark suite (on higher end hardware at least).
Although it’s not quite as strong of a showing for the 7770 series as under DiRT 3, the 7770 series again does fairly well for itself here. At 1680 the 7770, 6850, and GTX 460 1GB are effectively tied, though the XFX factory overclock isn’t enough to get the BESDD up to 6870 performance.
Meanwhile the 7770/7750 gap is now 41%, making this the most shader/texture limited benchmark yet. Unfortunately for the 7750 this means it’s now noticeably trailing the 5770.
Batman: Arkham City
Batman: Arkham City is loosely based on Unreal Engine 3, while the DirectX 11 functionality was apparently developed in-house. With the addition of these features Batman is far more a GPU demanding game than its predecessor was, particularly with tessellation cranked up to high.
With Batman the 7770 is now nearly tied with the 6850, trailing it by under 2%. Meanwhile against NVIDIA’s cards it’s hit and miss; it’s close to overtaking the GTX 460 1GB but can’t quite make it. XFX’s BESDD can do it, but then it’s trailing the GTX 560 instead. Elsewhere the 7750 finally shows some potency against the 5770, beating AMD’s previous headliner by 7%.
Portal 2 continues the long and proud tradition of Valve’s in-house Source engine. While Source continues to be a DX9 engine, Valve has continued to upgrade it over the years to improve its quality, and combined with their choice of style you’d have a hard time telling it’s over 7 years old at this point. Consequently Portal 2’s performance does get rather high on high-end cards, but we have ways of fixing that…
On the positive side, Portal 2 is easy enough to run that the 7770 easily clears 60fps at 1920, even with 4x MSAA. On the down side, the 6800 series does the same thing only better. Regardless of the resolution this is one of the worst games for the 7700 series, as even the XFX BESDD can’t crack the 6850. The 7750 somehow fares even worse, falling just behind the 5750, never mind the 5770. While this is largely academic thanks to the high framerates, it would appear we’ve found a general weakness in the 7700 series.
Editor’s Note: Earlier today DICE released a patch that among other things is supposed to improve Radeon HD 7000 series performance in the game. We’ll update our numbers to include revised benchmarks as soon as we can.
Its popularity aside, Battlefield 3 may be the most interesting game in our benchmark suite for a single reason: it’s the first AAA DX10+ game. It’s been 5 years since the launch of the first DX10 GPUs, and 3 whole process node shrinks later we’re finally to the point where games are using DX10’s functionality as a baseline rather than an addition. Not surprisingly BF3 is one of the best looking games in our suite, but as with past Battlefield games that beauty comes with a high performance cost
BF3 is an all-around GPU killer, which in the case of the 7700 series doesn’t help matters. Keeping in mind our benchmarks typically trend high, even at 1680 with Medium settings we’re not cracking 60fps with anything less than a GTX 460 1GB. In this case the 7770 should be playable, but intense firefights will definitely drop through the 30fps floor.
In any case the performance of the 7700 series is starting to show some consistency. Once again the 7770 underperforms the 6850, this time by 5%, elsewhere the 7750 noticeably trails the 5770. Nothing on the AMD side is anywhere close to the GTX 560 however.
Looking at our data, I’m a bit worried about the amount of VRAM the 7770 has. 1GB is already not quite enough for some games at 1920 with high quality settings, but BF3 is especially punishing. If we see more games like BF3, I have to wonder if 1GB will be enough for even 1680 in a year’s time.
Our next game is Starcraft II, Blizzard’s 2010 RTS megahit. Much like Portal 2 it’s a DX9 game designed to run on a wide range of hardware so performance is quite peppy with most high-end cards, but it can still challenge a GPU when it needs to.
The story is much the same under Starcraft II as it is in a number of other games, with the 7770 once more trailing the 6850. This time the 6850 leads by 8% and given AMD’s struggling SC2 performance the 7770 is well behind NVIDIA’s comparable cards. The good news here is that the 7750 has once again managed to charge past the 5770, this time enjoying a rare 11% lead. Meanwhile the 1920 numbers are a bit worrying – not that the 7700 is particularly capable of that resolution anyhow, but the 7770 just gets clobbered by the 6850 here, most likely due to MSAA.
The Elder Scrolls V: Skyrim
Prior to the launch of our new benchmark suite, we wanted to include The Elder Scrolls V: Skyrim, which is easily the most popular RPG of 2011. However as any Skyrim player can tell you, Skyrim’s performance is CPU-bound to a ridiculous degree. With the release of the 1.4 patch and the high resolution texture pack this has finally been relieved to the point where GPUs once again matter, particularly when we’re working with these slower midrange GPUs. As such, we're now including it in our test suite.
Unfortunately for the 7770, at 1680 this is another case where the 6850 is well in the lead; even XFX’s factory overclock can’t overcome the gap. At 64.3fps it’s still above the 60fps mark, but there’s clearly room for improvement. As for the 7750 it’s once again tied with the 5770.
It’s interesting to note though that at 1920 the 7700 series fares well, in fact it fares too well. 1920 is clearly hitting a memory limit for 1GB cards, and while the 5000 and 6000 series completely flounder, the 7700 series does unexpectedly well. Did AMD improve their memory allocation methods for the 7000 series? Is PCI Express 3.0 helping to alleviate the bottleneck? It’s hard to say, but clearly AMD has improved on something between the 6000 series and now.
Our final game, Civilization 5, gives us an interesting look at things that other RTSes cannot match, with a much weaker focus on shading in the game world, and a much greater focus on creating the geometry needed to bring such a world to life. In doing so it uses a slew of DirectX 11 technologies, including tessellation for said geometry, driver command lists for reducing CPU overhead, and compute shaders for on-the-fly texture decompression.
Thanks to AMD’s radically improved compute shader performance, the 7700 series fares extremely well here. Instead of stuggling to keep up with the 6850, the 7770 is well in the lead even against the 6870. Throw in XFX’s factory overclock and now a 7700 card is leading even the GTX 560 on a game NVIDIA was handily winning 2 months ago. Even the 7750 greatly benefits here, tying with the 6850 instead of the 5770 as it normally does.
It’s a shame for AMD that more games aren’t like Civ V, for if they were the 7700 series would look quite a lot better.
Moving on from our look at gaming performance, we have our customary look at compute performance. With GCN AMD significantly overhauled their architecture in order to improve compute performance, as their long-run initiatives rely on GPU compute performance becoming far more important than it is today.
With such a move however AMD has to solve the chicken and the egg problem on their own, in this case by improving compute performance before there are really a large variety of applications ready to take advantage of it. As we’ll see AMD has certainly achieved that goal, but it raises the question of what was the tradeoff for that? We have some evidence that GCN is more efficient than VLIW5 on a per-shader basis even in games, but at the same time we can’t forget that AMD has gone from 800 SPs to 640 SPs in the move from Juniper to Cape Verde, in spite of a full node jump in fabrication technology. In the long run AMD will be better off, but I suspect we’re looking at that tradeoff today with the 7700 series.
Our first compute benchmark comes from Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. Note that this is a DX11 DirectCompute benchmark.
Theoretically the 5770 has a 5% compute performance advantage over the 7770. In practice the 5770 doesn’t stand a chance. Even the much, much slower 7750 is ahead by 12%, meanwhile the 7770 is in a class of its own, competing with the likes of the 6870. The 7770 series still trails the GTX 560 to some degree, but once again we’re looking at the proof of just how much the GCN architecture has improved AMD’s compute performance.
Our next benchmark is SmallLuxGPU, the GPU ray tracing branch of the open source LuxRender renderer. We’re now using a development build from the version 2.0 branch, and we’ve moved on to a more complex scene that hopefully will provide a greater challenge to our GPUs.
SmallLuxGPU is another good showing for the GCN based 7700 series, with the 7770 once again moving well up the charts. This time it’s between the 6850 and 6870, and well, well ahead of the GTX 560 or any other NVIDIA video cards. Throwing in an overclock pushes things even farther, leading to the XFX BESDD tying the 6870 in this benchmark.
For our next benchmark we’re looking at AESEncryptDecrypt, an OpenCL AES encryption routine that AES encrypts/decrypts an 8K x 8K pixel square image file. The results of this benchmark are the average time to encrypt the image over a number of iterations of the AES cypher.
Under our AESEncryptDecrypt benchmark the 7770 does even better yet, this time taking the #2 spot and only losing to its overclocked self. PCIe 3.0 helps here, but as we’ve seen with the 7900 series there’s no replacement for a good compute architecture.
Finally, our last benchmark is once again looking at compute shader performance, this time through the Fluid simulation sample in the DirectX SDK. This program simulates the motion and interactions of a 16k particle fluid using a compute shader, with a choice of several different algorithms. In this case we’re using an (O)n^2 nearest neighbor method that is optimized by using shared memory to cache data.
It would appear we’ve saved the best for last, as in our fluid simulation benchmark the top three cards are all 7700 series cards. This benchmark strongly favors a well organized cache, leading to the 7700 series blowing past the 6800 series and never looking back. Even NVIDIA’s Fermi based video cards can’t keep up.
Before moving on from compute performance, we wanted to quickly take a look at theoretical performance. Identifying the theoretical performance of the 7700 series in relation to other cards may help explain why it’s often trailing the 5770 and 6850.
A quick look at texture fillrates gives us our answer for the 7750: it has even lower texture performance than the 5750, never mind the 5770. Thankfully very few games are heavily texture bound these days – and if they were the 7750 likely wouldn’t have enough VRAM for them anyhow – but the massive gap in theoretical texture performance between the 7750 and7700 means that the 7750 is behind virtually everything else.
Conversely if you look at the pixel fill rate it’s almost identical to the 7770, which in turn trails the 5770. However in this case the 3DMark Pixel Fill test appears to be heavily memory bandwidth bound, which is why it trails the 6870 by so much.
Moving on, looking at tessellation performance is both good and bad for the 7700 series. With a maximum of 1 triangle/clock, GCN’s tessellation improvements can only do so much. It’s enough to vault past the 5770, but the 6870 still has better tessellation performance even with its lower clockspeed. Given AMD’s use of off-die buffering, it’s entirely possible we’re looking at a memory bandwidth constraint here.
Unigine Heaven backs these findings and then some. Tessellation performance is improved relative to the 5700 series, but at best the 7700 series is only going to catch the 6850.
Power, Temperature, & Noise
As always, we wrap up our look at a new video card with a look at the physical performance attributes: power consumption, temperatures, and noise.
While it’s the second sub-series of the 7000 series, the 7700 series is actually first in a number of ways for AMD. It’s the first midrange series to feature PowerTune since its introduction in 2010, and it’s the first time the 28nm process has been used on such low power cards. As such the power, temperature, and noise characteristics of the 7700 series are virtually a blank slate waiting to be filled.
|Radeon HD 7700 Series Voltages|
|Ref 7750 Load||Ref 7770 Load||XFX R7770 BESDD|
The idle voltage for all 3 of our 7700 series cards was 0.825v. Meanwhile the load voltage for our 7750 was 1.1v, while it was 1.2v for both the reference 7770 and the XFX BESDD.
AMD’s official idle power consumption figure is <10W for both the 7750 and 7750. In practice we find that there’s a 2W difference between the two cards, with the 7750 coming in at 104W at the wall and the 7770 at 106W. AMD’s continuing efforts to reduce idle power consumption are clearly paying off, making this the lowest idle power usage figures we’ve recorded yet.
Meanwhile under the so-called “long idle” scenario with a blank monitor, the 7000 series continues to cement its lead. Officially all Southern Islands cards have a long idle power consumption level of <3W since they should all be using the same PCIe controller to keep a heartbeat going, however we’re finding there is a difference, even between the 7750 and 7770. The 7770 consumes 3W less than the 7900 series, meanwhile the 7750W consumes another 3W less, bringing it down to 97W. Since our readings are from the wall it’s tough to gage just how much these cards are still using, but at this point there doesn’t seem to be much farther to drop.
When we talk about the 7700 series bluring the line between what we’d expect out of an AMD 700 series card and an AMD 600 series card, results like this are part of the reason why. With a 75W PowerTune limit the 7750 has the lowest power consumption of any of the cards in our lineup, beating even the 5750. The 7750 is clearly in a league of its own, with only the 7770 drawing similar amounts of power as the 5700 series.
To that extent, it’s interesting to note that the XFX BESDD consumes less power than the reference 7770, in spite of its factory overclock. XFX did not increase the voltage of the card, but we’d still expect power consumption to go up at least a bit, not come down. In any case even with its 10% better performance, it’s consuming 7W less than the reference 7770 here. Otherwise at 250W the 7770 is pulling 8W less than the 5770 and 25W less than the 6850 from the wall, succinctly showcasing the power benefits of TSMC’s 28nm process.
Our results with OCCT largely mirror Metro, with the 7750 in a class of its own while the 7770 consumes less power than anything other than the 5750. Since this is our pathological test, the lack of PowerTune plays a big part here, as PowerTune keeps the 7770 capped at 100W while the 5770 and 6850 are free to go well over their TDPs.
There’s little to say about idle temperatures that hasn’t been said before. With a half-decent cooler, almost any card can reach the mid-to-low 30s. The 7700 series is no exception.
Moving on to Metro, our results are largely consistent with what we’d expect given our earlier power data. Even with the lower power consumption of the 7700 series AMD can’t quite beat the GTX 460 1GB, otherwise with temperatures in the mid to upper 60s the 7700 series looks quite good. XFX’s BESDD looks especially good thanks to the fact that it’s an open air cooler, as it only reaches 64C.
As with OCCT power consumption, OCCT temperatures are largely a story of PowerTune or the lack thereof. With PowerTune clamping down on power consumption, temperatures never rise by too much. At 71C the reference 7770 is still doing well, while XFX’s card does even better.
Based on our power and temperature data, our noise data came at somewhat of a disappointment. The 7750 has a small fan with an idle fanspeed of 39%, leading to it being quite loud relative to its competitors. Thankfully none of AMD’s partners will be using that specific cooler, so we wouldn’t expect any retail cards to be this bad. Though if silence is key, Sapphire’s passively cooled 7750 is always an option.
As for our 7770 cards there aren’t any surprises. The open air cooled XFX BESDD does the best, though at only 40.5dB the reference 7770 is doing almost as well on its own.
I had to rerun Metro a few times just to make sure our XFX BESDD numbers are right; and they were. XFX’s open air cooler combined with the 100W board limit of the 7770 means that there’s relatively little heat to dissipate, which the open air cooler does extremely well. While this isn’t technically silent it’s damn close.
Elsewhere the reference 7770 does decently, but its half blower nature hurts it when it comes to noise. The noisy 7750 only gets louder, unfortunately.
Finally, noise under OCCT is nearly the same story. The XFX BESDD finally goes above 40dB, a testament to the capabilities of XFX’s open air cooler. As for the AMD cards, there’s really not a great deal positive to say as even a traditional blower shouldn’t be quite as bad as what we’re seeing with the reference 7770. Perhaps it’s for the best that none of AMD’s partners are using their reference designs.
Wrapping things up, I once had someone comment to me that they can gauge my opinion of a product based solely on the first paragraph of the final page. If I say “there’s no such thing as a bad card, only bad prices” then it’s likely not a favorable review. That statement is once more being validated today, if only in a meta context.
To be clear, we’ve been waiting for some time to see GCN filter down to lower priced cards, and even longer to see PowerTune in particular make it down here. The fact that we now have reliable power throttling and solid compute performance is not lost on us. It’s a welcome advancement.
However our expectation with a new manufacturing process – and perhaps we’re being greedy here – is that we’ll see cards become cheaper and we’ll see power consumption come down. AMD has achieved the second item in spades, and as a result both the Radeon HD 7750 and Radeon HD 7770 are well ahead of any competing 75W and 100W cards respectively. The 7750 in particular is a standout thanks to the fact that it generally offers 5700 series performance on a sub-75W card, and even at $109 it clearly offers a great deal of value as an HTPC video card. All of this will be an even more welcome change when Cape Verde filters down to laptops in the coming months.
The problem for AMD today isn’t the power/performance curve, it’s the price/performance curve. 16 months ago AMD launched the Radeon HD 6850 at $179 amidst fierce competition from NVIDIA. Ignoring the current price of the 6850 for the moment, on average the 7770 delivers 90% of the 6850’s gaming performance for 90% of the 6850’s launch price. In other words in 16 months AMD has moved nowhere along the price/performance curve – if you go by launch prices you’re getting the same amount of performance per dollar today as you did in October of 2010. In reality the 6850 is much cheaper than that, with a number of cards selling for $159 before a rebate, while several more 6870s sell for $159 after rebate. The 7770 is so far off the price/performance curve that you have to believe that this is either a pricing error or AMD is planning on quickly halting 6800 series production.
Now to be fair there’s more to consider than just performance in existing games. The 7770 supports DX11.1, VCE, PowerTune, Fast HDMI, and other features the 6800 series doesn’t have, and it does all of this while consuming around 25W less than the 6850. But that’s just not enough. DX11.1 is a point update that’s still the better part of a year away and will only offer a tiny number of new features, while VCE is AWOL and cannot be evaluated, and Fast HDMI will be a niche feature for use with extremely expensive TVs for some time to come. This is not like the 4000/5000 series gap – today and tomorrow the 7000 series will only offer marginal feature benefits. The best argument for the 7770 is the power difference, but considering that both the 6850 and 7770 require external power anyhow that 25W difference is unlikely to matter.
The 7700 series is a fine lineup of cards, but AMD has finally shot itself in the foot with its conservative pricing. The 7750 can ride on the sub-75W niche for now, but the only way the 7770 will make any sense is if it comes down in price. Until then AMD’s worst competition for the 7700 series is not NVIDIA, it’s their 6850.
With that said, the 7700 series clearly has potential. XFX’s R7770 Black Edition S Double Dissipation does a great job demonstrating this with its virtually silent operation, while the card’s factory overclock largely closes the performance gap with the 6850. With its combination of performance and power consumption the 7700 series will be AMD’s midrange workhorse for 2012, of that there is no question. Now it’s simply up to AMD to make it so. After all there’s no such thing as a bad card, only bad prices.