...and then disaster struck.

Or at least that's how it felt. The past few weeks have been incredibly tumultuous, sleepless, and beyond interesting. It is as if AMD and NVIDIA just started pulling out hardware and throwing it at eachother while we stood in the middle getting pegged with graphics cards. And we weren't just hit with new architectures and unexpected die shrinks, but new drivers left and right.

First up was GT200, which appeared in the form of the GeForce GTX 280 and GeForce GTX 260. Of course, both of those can be paired or tri-ed (if you will), but with two cards requiring at least a 1200W PSU we're a bit worried of trying three. Then came the randomness that was the accidental launch of the Radeon HD 4850 (albeit with no architectural information) and only a couple hours later we first heard about the 9800 GTX+ which is a die shrunk higher clocked 9800 GTX that is now publicly announced and will be available in July.

And now we have the other thing we've been working on since we finished GT200: RV770 in all it's glory. This includes the 4850 whose performance we have already seen and the Radeon HD 4870: the teraflop card that falls further short of hitting its theoretical performance than NVIDIA did with GT200. But theoretical performance isn't reality, and nothing can be done if every instruction is a multiply-add or combination of a multiply-add and a multiply, so while marketing loves to trot out big numbers we quite prefer real-world testing with games people will actually play on this hardware.

But before we get to performance, and as usual, we will want to take as deep a look into this architecture as possible. We won't be able to go as deep with RV770 as we could with GT200, as we had access to a lot of information both from NVIDIA and from outside NVIDIA that allowed us to learn more about their architecture. At the same time, we still know barely anything about the real design of either NVIDIA or AMD's hardware as they prefer to hold their cards very close.

This won't work long term, however. As we push toward moving compute intensive applications to the GPU, developers will not just want -- they will need low level architectural information. It is impossible to properly optimize code for an architecture when you don't know exact details about timing, latency, cache sizes, register files, resource sharing, and the like. While, this generation, we have decidedly more information from NVIDIA on how to properly program their architecture, we still need more from both AMD and NVIDIA.

And Now, the Rest of the Story

Last week was a weird teaser - we gave you the goods, without explaining what they were.

By now you know that the Radeon HD 4850 is the best buy at $199, but today we're able to tell you much about its inner workings as well as introduce its faster, more expensive sibling: the Radeon HD 4870.

ATI Radeon HD 4870 ATI Radeon HD 4850 ATI Radeon HD 3870
Stream Processors 800 800 320
Texture Units 40 40 16
ROPs 16 16 16
Core Clock 750MHz 625MHz 775MHz+
Memory Clock 900MHz (3600MHz data rate) GDDR5 993MHz (1986MHz data rate) GDDR3 1125MHz (2250MHz data rate) GDDR4
Memory Bus Width 256-bit 256-bit 256-bit
Frame Buffer 512MB 512MB 512MB
Transistor Count 956M 956M 666M
Manufacturing Process TSMC 55nm TSMC 55nm TSMC 55nm
Price Point $299 $199 $199

Priced at $299 the Radeon HD 4870 is clocked 20% higher and has 81% more memory bandwidth than the Radeon HD 4850. The GPU clock speed improvement is simply due to better cooling as the 4870 ships with a two-slot cooler. The memory bandwidth improvement is due to the Radeon HD 4870 using GDDR5 memory instead of GDDR3 used on the 4850 (and GDDR4 for 3870); the result is a data rate equal to 4x the memory clock speed or 3.6Gbps. The Radeon HD 4870 and 4850 both use a 256-bit memory bus like the 3870 before it (as well as NVIDIA's competing GeForce 9800 GTX), but total memory bandwidth on the 4870 ends up being 115.2GB/s thanks to the use of GDDR5. Note that this is more memory bandwidth than the GeForce GTX 260 which has a much wider 448-bit memory bus, but uses GDDR3 devices.

NVIDIA GeForce GTX 280 NVIDIA GeForce GTX 260 NVIDIA GeForce 9800 GTX ATI Radeon HD 4870 ATI Radeon HD 4850 ATI Radeon HD 3870
Memory Size 1GB 896MB 512MB 512MB 512MB 512MB
Memory Technology GDDR3 GDDR3 GDDR3 GDDR5 GDDR3 GDDR4
Memory Bus Width 512-bit 448-bit 256-bit 256-bit 256-bit 256-bit
Memory Clock 1107MHz 999MHz 1100MHz 900MHz 993MHz 1125MHz
Memory Data Rate 2.2Gbps 2.0Gbps 2.22Gbps 3.6Gbps 1.99Gbps 2.25Gbps
Memory Bandwidth 141.7GB/s 111.9GB/s 70.4GB/s 115.2GB/s 63.6GB/s 72.0GB/s

The use of GDDR5 enabled AMD to deliver GeForce GTX 260 class memory bandwidth, but without the pin-count and expense of a 448-bit memory interface. GDDR5 actually implements a number of Rambus-like routing and signaling technologies while still remaining a parallel based memory technology, the result is something that appears to deliver tremendous bandwidth per pin in a reliable, high volume solution.

AMD most likely took a risk on bringing GDDR5 to market this early and we do expect NVIDIA to follow suit, AMD is simply enjoying the benefits of jumping on the GDDR5 bandwagon early and getting it right, at least it seems that way. It wouldn't be too far fetched to imagine a 55nm GT200 die shrink with a 256-bit GDDR5 memory interface, it should allow NVIDIA to drop the price down to the $300 level (at least for the GTX 260).

As we mentioned in our Radeon HD 4850 Preview, both the Radeon HD 4870 and 4850 now support 8-channel LPCM audio output over HDMI. AMD just sent over 8-channel LPCM drivers for the Radeon HD 4870 so we'll be testing this functionality shortly. As we mentioned in our 4850 preview:

"All of AMD's Radeon HD graphics cards have shipped with their own audio codec, but the Radeon HD 4800 series of cards finally adds support for 8-channel LPCM output over HDMI. This is a huge deal for HTPC enthusiasts because now you can output 8-channel audio over HDMI in a motherboard agnostic solution. We still don't have support for bitstreaming TrueHD/DTS-HD MA and most likely won't anytime this year from a GPU alone, but there are some other solutions in the works for 2008."

The Radeon HD 4870 is scheduled for widespread availability in early July, although AMD tells us that some cards are already in the channel. Given that the 4870 relies on a new memory technology, we aren't sure how confident we can be that it will be as widely available as the Radeon HD 4850 has been thus far. Keep an eye out but so far the 4850 has been shipping without any issues at $199 or below, so as long as AMD can get cards in retailers' hands we expect the 4870 to hit its $299 price point.

AMD's "Small-Die" Strategy


View All Comments

  • DerekWilson - Wednesday, June 25, 2008 - link

    i havent had any problem with any modern graphics card (dvi or hdmi) and digital hdtvs

    i haven't really played with analog for a long time and i'm not sure how either amd or nvidia handle analog issues like overscan and timing.
  • araczynski - Wednesday, June 25, 2008 - link

    interesting, what cards have you worked with? i have the 8800gts512 right now and have the same problem as with the 7900gtx previously. when i select 1080p for the resolution (which the drivers recognize the tv being capable of as it lists it as the native resolution) i get a washed out messy result where the contrast/brightness is completely maxed (sliders do little to help) as well as the whole overscan thing that forces me to shrink the displayed image down to fit the actual tv (with the nvidia driver utility). 1600x900 can usually be tolerable in XP (not in vista for some reason) and 1080p is just downright painful.

    i suppose it could by my dvi to hdmi cable? its a short run, but who knows... i just remember reading a bit on the nvidia forums that this is a known issue with the 8800 line, so was curious as to how the 9800 line or even the 4800 line handle it.

    but as the previous guy mentioned, ATI does tend to do the TV stuff much better than nvidia ever did... maybe 4850 crossfire will be in my rig soon... unless i hear more about the 4870x2 soon...
  • ChronoReverse - Wednesday, June 25, 2008 - link

    ATI cards tend to do the TV stuff properly Reply
  • FXi - Wednesday, June 25, 2008 - link

    If Nvidia doesn't release SLI to Intel chipsets (and on a $/perf ratio it might not even help if it does), the 4870 in CF is going to stop sales of the 260's into the ground.

    Releasing SLI on Intel and easing the price might help ease that problem, but of course they won't do it. Looks like ATI hasn't just come back, they've got a very, very good chip on their hands.
  • Powervano - Wednesday, June 25, 2008 - link

    Anand and Derek

    What about temperatures of HD4870 under IDLE and LOAD? page 21 only shows power comsumption.
  • iwodo - Wednesday, June 25, 2008 - link

    Given how ATI architecture greatly rely on maximizing its Shader use, wouldn't driver optimization be much more important then Nvidia in this regard?

    And is ATI going about Nvidia CUDA? Given CUDA now have a much bigger exposure then how ever ATI is offering.. CAL or CTM.. i dont even know now.
  • DerekWilson - Wednesday, June 25, 2008 - link

    getting exposure for AMD's own GPGPU solutions and tools is going to be though, especially in light of Tesla and the momentum NVIDIA is building in the higher performance areas.

    they've just got to keep at it.

    but i think their best hope is in Apple right now with OpenCL (as has been mentioned above) ...

    certainly AMD need to keep pushing their GPU compute solutions, and trying to get people to build real apps that they can point to (like folding) and say "hey look we do this well too" ...

    but in the long term i think NVIDIA's got the better marketing there (both to consumers and developers) and it's not likely going to be until a single compute language emerges as the dominant one that we see level competition.
  • Amiga500 - Wednesday, June 25, 2008 - link

    AMD are going to continue to use the open source alternative - Open CL.

    In a relatively fledgling program environment, it makes all the sense in the world for developers to use the open source option, as compatibility and interoperability can be assured, unlike older environments like graphics APIs.

    OSX v10.6 (snow lepoard) will use Open CL.
  • DerekWilson - Wednesday, June 25, 2008 - link

    OpenCL isn't "open source" ...

    Apple is trying to create an industry standard heterogeneous compute language.

    What we need is a compute language that isn't "owned" by a specific hardware maker. The problem is that NVIDIA has the power to redefine the CUDA language as it moves forward to better fit their architecture. Whether they would do this or not is irrelevant in light of the fact that it makes no sense for a competitor to adopt the solution if the possibility exists.

    If NVIDIA wants to advance the industry, eventually they'll try and get CUDA ANSI / ISO certified or try to form an industry working group to refine and standardize it. While they have the exposure and power in CUDA and Tesla they won't really be interested in doing this (at least that's our prediction).

    Apple is starting from a standards centric view and I hope they will help build a heterogeneous computing language that combines the high points of all the different solutions out there now into something that's easy to develop or and that can generate code to run well on all architectures.

    but we'll have to wait and see.

  • Amiga500 - Wednesday, June 25, 2008 - link

    Apple has passed over control of Open CL to the Khronos group, which manage open sourced coding.

    To all intentions and purposes, it is open source. :-)

Log in

Don't have an account? Sign up now