Original Link: http://www.anandtech.com/show/1225
Mobile GPU Q1/2004 Update - ATI M10, M11, and NVIDIA's NV36Mby Andrew Ku on January 28, 2004 2:30 PM EST
- Posted in
It wasn't too long ago that executives at the big system manufacturers expected the sales for notebooks to surpass their desktop line. This hasn't materialized as fast as previously predicted, but the notebook market continues to grow by leaps and bounds, particularly as the lines behind ultra-portable, thin and light, mobile workstations, and desktop replacements become blurred. Coupled with lowering prices of PC components and space saving capabilities, notebooks are finding new demographics to tap into.
So, it really comes as no surprise that more and more R&D dollars are being invested in a company's mobile division. Dothan and the mobile version of Athlon 64 are due to be out soon, and we will spend more time on that subject when the time comes. So far, we back on the subject of mobile GPUs. Gaming/rendering on the newest high-end desktop card can (for the most part) attain desirable frame rates, even with non-released games that have implemented heavily on pixel shader use.
Gaming/rendering on notebooks has never been up to par with the best of the desktop stuff (behemoth-sized desktop replacements don't count); yet, more and more people are looking for laptops that do just that. Our most recent case in point would be Half Life 2 on Mobility Radeon 9600 Pro and GeForce FX Go5650 based laptops. The laptop equipped with the Mobility Radeon 9600 Pro made Half Life 2 reasonably playable on a notebook - quite a noteworthy accomplishment. Meanwhile, the GeForce FX Go5650 (one up from the Go5600) based notebook went to a crawling halt.
Domestically, the sales for desktop replacement notebooks are uncannily high compared to ultra-portables, and vice versa for the European market. In the North American market, there is a tendency to like the biggest and baddest stuff around, and we will take it despite the price tag and size. This is why SIs and SDs are carefully watching the upcoming mobile GPU scene between ATI and NVIDIA. Unlike the desktop scene, the stage will be set not just on performance, but also on thermal characteristics. NVIDIA and ATI have an easier time on desktop cards, since they can crank up the clock speeds and slap on a bigger heatsink. Meanwhile, mobile GPUs are at the mercy of the system manufacturer (i.e. Dell, HP/Compaq, IBM, etc.), since NVIDIA and ATI need to conform their GPU's thermal characteristics to the notebook's thermal design, all the while maintaining the promised performance.
NVIDIA GeForce FX Go5700 - NVIDIA answers backAs it was with Mobility Radeon 9600 (code named M10) and GeForce FX Go56X0 (code named NV31M), NVIDIA announced their next gen mobile GPU first. Dubbed GeForce FX Go5700, it is based on the NV36 GPU used in the desktop GeForce FX 5700 Ultra and non-Ultra cards. The architecture behind the Go5700 hasn't changed much compared to its desktop brothers, aside from optimization for power consumption and thermal emissions, which is why we won't dive into this again. For a refresher course, you should hit the pages of our NVIDIA GeForce FX 5700 Ultra review.
Some key points to keep in mind: NV36M is produced on a 0.13 micron process, is a full DX9 part, and is capable of performing twice the number of floating point operations over NV31M and NV34M. This is a benefit for any post-NV35 architectures over the pre-NV35 designs. And, it is a very large factor that plays into the increase in performance that we have seen when we step up from NV30 to NV35 and from NV31 to NV36. While the 5700 Ultra is a reflection of the performance delta between NV30 and NV38 for the mid-range cards, the Go5700 will be the high-end derivative of the GeForce FX Go family for mobile products. Meanwhile, the Go56X0 (NV31M) will effectively be bumped down as a mid-range mobile solution and the Go5200 (NV34M) to the value.
Based on our conversations with NVIDIA near the Comdex 2003 launch of NV36M, the GeForce FX Go5700 should come with no increase in real estate over the GeForce FX Go56X0. According to the specs, the Go5700 should, in theory, operate a slightly lower voltage (1.1V) setting over the Go5650, but the difference in their specs is so small that we doubt there will be any battery life degrade or improvement. Power consumption should remain the same (8W) as the Go5650.
|GeForce FX Go5700||GeForce FX 5700 Ultra||GeForce FX Go5600/5650||GeForce FX 5600 Ultra (wire bond/flip chip)||GeForce FX Go5200|
|2D Core Clock||350MHz||300MHz||350MHz||350MHz/400MHz||300MHz|
|2D Memory Clock||300MHz||450MHz||350MHz||350MHz/400MHz||300MHz|
|3D Core Clock||450MHz||475MHz||350MHz||350MHz/400MHz||300MHz|
|3D Memory Clock||300MHz||450MHz||350MHz||350MHz/400MHz||300MHz|
Since it is based on post-NV35 architecture, the Go5700 operates at two different speeds, like its desktop counterpart. In 2D mode, the Go5700 operates at 350MHz core clock and 300MHz memory clock, whereas the desktop 5700 Ultra operates at 300MHz core and 450MHz (900MHz effective) memory clock. In 3D mode (gaming/graphic rendering), the Go5700 operates at 450MHz core clock, while the 5700 Ultra operates at 475MHz core clock. There is no change in memory clock while operating in between the two modes.
The highest clocks that we have seen on the Go5650 remains to be 325MHz core and 295MHz memory. The fact that NVIDIA wasn't able to get clocks up to their original announcement makes us a bit more cautious of what the final clocks will be for systems based on the Go5700. The 3D clocks for the Go5700 are actually designated for desktop replacement systems, which leads us to suspect that non-desktop replacement systems like thin and lights will have to function at lower 3D clocks or only one operating mode.
As for integrated memory package designs, NVIDIA tells us that they are not considering this at the moment because it brings no substantial improvement in performance. While this is true, integrated memory package designs are very much favored for thin and light notebooks, despite their price premium. The fact that you could have a 4-pound laptop power through the most intensive of games is no doubt very tempting.
ATI M11 - What we can say...At this point, we can't say very much, since this is an unannounced product, which no doubt brings up NDA concerns. So far, what we can say is that M11 has been branded. Additionally, ATI has no plans to use the name, "Mobility Radeon 9600 XT"... You can draw your own conclusions from that.
A while ago, we saw a system equipped with an M11 GPU, and the results were quite impressive, particularly over the Mobility Radeon 9600 (code named M10). The fact that the Mobility Radeon 9600 can attain reasonable frame rates in Half Life 2 seems to suggest that M11 will surpass that marker, perhaps making AA and AF possible in HL2 while attaining reasonable frame rates.
We will have more on this later.
ATI Mobility Radeon 9600 Pro - What does it take to go Pro?When we first took a look at the Voodoo Envy M:855 in part 1 of our coverage, we realized quickly that ATI's Mobility Radeon 9600 Pro didn't come with Overdrive and the implementation of GDDR2-M clearly was missing. We knew this to be the case for other Mobility Radeon 9600 Pro based notebooks, but we were hoping to see something new of the M:855, since its unprecedented use of a processor from the Athlon 64 family.
Back in March, we reported that the Mobility Radeon 9600 Pro would feature higher core clocks and GDDR2-M memory support for higher memory clocks. Contrary to what its name may suggest, GDDR2-M memory is not DDR2 memory. Like the "DDR2" memory used by the GeForce FX, the GDDR2-M memory that the Mobility Radeon 9600 Pro supports does not transmit four times per clock. Instead, the memory is optimized and therefore, able to run at higher frequencies than before.
The GDDR2-M memory support that the Mobility Radeon 9600 Pro offers will allow the memory chips used on the processor to operate somewhere between "DDR600 and DDR900 or above" speeds, indicating that 900+MHz effective operating frequencies are possible. Just how fast the memory will go, ATI is not yet saying, probably due to the fact that GDDR2-M memory is not readily available or in mass production quite yet. As a side note, the GDDR2-M memory offers power-saving features over both GDDR-2 and regular DDR memory chips, and it does not require external termination (saving precious space).
For those unaware, Elpida canceled GDDR2-M from their roadmap, which means that ATI's only source for this technology was removed. This only added to the confusion between Pro and non-Pro mobile graphic parts because one of the two distinct components for the Pro dubbing had been completely stripped away from the equation; the other being Overdrive. Previously, we were working with Voodoo, Dell and ATI to sort out the Pro-naming issue. During this time, we discovered that there were several system vendors that have been inquiring about the use of the Pro name, as the use of any extra derivative conjures up the sense of a stronger product. ATI admits that they could have handled this better. Granted, this issue didn't come to ATI's table of their own fault, but the cancellation of GDDR2-M was a while back and they could have addressed this issue much sooner, which they readily admit.
As we expected in March, Overdrive was to be a new feature that was intended to be implemented in Mobility Radeon 9600 Pro, but due to its early introduction, system vendors seem to have decided not to implement it until the technology matures. ATI tells us that Overdrive is actually supported by Mobility Radeon 9600 as it is on Radeon 9600 Pro and Radeon 9800 Pro, though the technology is only implemented in their respective big brothers. As we mentioned in our Voodoo Envy M:855 part 1 coverage, there are no system vendors that are currently implementing Overdrive, and we would highly caution anyone who might be expecting this feature when they buy a Mobility Radeon 9600 Pro based notebook. Based on our previous communication with ATI, we believe that there might be some notebooks popping up in future quarters, but we will have to wait to see.
This feature announcement was big news in the desktop market because it was a very rare circumstance of sanctioned overclocking. In relation to the mobile market, this is even bigger news because of the ramifications it has for any specific notebook's thermal budget. ATI next gen mobile processor code named M11 will also support Overdrive, and we are still teething to see a production notebook that implements this technology. Note that system designers have the option of turning off the hardware that enables Overdrive.
In our previous coverage, we reported that any Mobility Radeon 9600 (also known as M10) with a core clock of at least 350MHz and 128MB of video memory would be eligible for the Pro postfix, but at that time, details were still being sorted out. Thanks to the guys down at Voodoo, Dell, and ATI, the naming issue was sorted out. Any M10 chip that operates at a core speed of at least 333MHz and have 64MB of video memory will be considered Pro. The condition of the drivers will remain the same; no differentiation between Pro and non-Pro because of the shared device ID.
ATI Mobility Radeon 9600 Turbo ProJoining the confusion for prospective notebook buys is the term Turbo Pro, which added to the Mobility Radeon 9600 title sounds like some "suped-up" version of the mobile GPU. The postfix Turbo Pro is now reserved for systems with an M10 chip that meet the Mobility Radeon 9600 Pro requirements and are configured with at least 128MB operating at 240MHz memory clock.
|ATI's M10 Family Line-up|
|Mobility Radeon 9600||Mobility Radeon 9600 Pro||Mobility Radeon 9600 Turbo Pro|
|Min. Memory Configuration||-||64MB||128MB|
|Min. Core Clock||-||333MHz||333MHz|
|Min. Memory Clock||-||-||240MHz|
The new requirements for Mobility Radeon 9600 Pro have some ramifications. Previously, there were fewer distinctions between laptops with 64MB or 128MB of video memory, high and low core clocks, etc. Now, there is a three-tier naming solution for the M10 and this means that several notebooks dubbed non-Pro have suddenly and miraculously gone professional, and dare I say it, Turbo pro. If anything, these multiple naming derivatives make it more confusing, so hopefully this won't happen with M11 or future mobile GPUs.
Final WordsSo, the stage is set and we should expect to see notebooks based on both GPUs out shortly; we expect within a month. With the Go5600, NVIDIA was able to ship systems out overseas and small quantities domestically a few months after the announcement, while ATI took a lot longer. What is odd this time around is that while NVIDIA announced before ATI, shipping notebooks and design wins will be announced about the same time for both companies. This is at least what is being suggested from our talks with system manufacturers.
Ideally, we would like to see shipping notebooks out within a week or two of a product launch. ATI's Mobility Radeon 9600 and NVIDIA's GeForce FX Go56X0 both took almost 6 months before we finally saw them in some tangible retail systems. Granted, they were in other overseas markets, but the main technology market is still North America. NVIDIA has a lost opportunity to do this for the Go5700, but ATI has a chance to do this with M11. Hopefully, they will continue to follow the tradition of the Mobility Radeon 9000, which was touted as the first mobile graphics chip to be announced and shipped within a week. The benefit behind this is that all the expectations and hype can be backed up quickly based on shipping systems, instead of waiting months to verify specs like core and memory clocks.
In the past, ATI and NVIDIA officially announced clocks for their mobile products equivalent to that of or close to their desktop product line. This makes it easier to present their mobile GPU solutions to SIs and SDs at ODMs and system manufacturers, since it helps gives their clients an approximate sense of expected performance. These clocks are really at the mercy of system manufacturers, since every notebook design has a different cooling system, and therefore, different thermal budget. The official clocks are more of an internal target than anything else, and we would like to continue to see clocks that are more realistic of shipping clock speeds. Obviously, "officially announcing" clocks that are unrealistically high helps gain design wins by enticing system manufacturing clients, but in the end, notebook owners are alienated that clock speeds are lower than "announced" and thus, so is performance.
NVIDIA has a challenge ahead of themselves. They need to stay up to par with ATI from a market share perspective due to the effect of branding, which will make the typical prospective notebook buyer identify the Mobile Radeon or GeForce FX Go lineup. NVIDIA is still new to the mobile market, so technically, any new market share is good for them, but for perception sake, performance needs to be competitive. NVIDIA will be in a weaker position if the Go5700 turns out to be competitive with ATI's Mobility Radeon 9600. The desktop NV36 cards are much more promising, and we hope to see NVIDIA bring back some competition to the mobile GPU market. We have a month or so before we can finally sort out all the details, so the verdict is still out...