More Mainstream DX10: AMD's 2400 and 2600 Series
by Derek Wilson on June 28, 2007 8:35 AM EST- Posted in
- GPUs
A Closer Look at RV610 and RV630
The RV6xx parts are similar to the R600 hardware we've already covered in detail. There are a few major differences between the two classes of hardware. First and foremost, the RV6xx GPUs include full video decode acceleration for MPEG-2, VC-1, and H.264 encoded content through AMD's UVD hardware. There was some confusion over this when R600 first launched, but AMD has since confirmed that UVD hardware is not at all present in their high end part.
We also have a difference in manufacturing process. R600 uses an 80nm TSMC process aimed at high speed transistors, while their RV610 and RV630 GPU based cards are fabbed on a 65nm TSMC process aimed at lower power consumption. The end result is that these GPUs will run much cooler and require much less power than their big brother the R600.
Transistor speed between these two processes ends up being similar in spite of the focus on power over performance at 65nm. RV610 is built with 180M transistors, while RV630 contains 390M. This is certainly down from the huge transistor count of R600, but nearly 400M is nothing to sneeze at.
Aside from the obvious differences of transistor count and the number of different units (shaders, texture unit, etc.), the only other major difference is in memory bus width. All RV610 GPU based hardware will have a 64-bit memory bus, while RV630 based parts will feature a 128-bit connection to memory. Here's the layout of each GPU:
RV630 Block Diagram
RV610 Block Diagram
One of the first things that jump out is that both RV6xx based designs feature only one render back end block. This part of the chip is responsible for alpha (transparency) and fog, dealing with final z/stencil buffer operations, sending MSAA samples back up to the shader to be resolved, and ultimately blending fragments and writing out final pixel color. Maximum pixel fill rate is limited by the number of render back ends.
In the case of both current RV6xx GPUs, we can only draw out a maximum of 4 pixels per clock (or we can do 8 z/stencil-only ops per clock). While we don't expect extreme resolutions to be run on these parts (at least not in games), we could run into issues with effects that make heavy use of MRTs (multiple render targets), z/stencil buffers, and antialiasing. With the move to DX10, we expect developers to make use of the additional MRTs they have available, and lower resolutions benefit from AA more than high resolutions as well. We would really like to see higher pixel draw power here. Our performance tests will reflect the fact that AA is not kind to AMD's new parts, because of the lack of hardware resolve as well as the use of only one render back end.
Among the notable features that we will see here are tessellation, which could have an even larger impact on low end hardware for enabling detailed and realistic geometry, and CFAA filtering options. Unfortunately, we might not see that much initial use made of the tessellation hardware, and with the reduced pixel draw and shading power of the RVxx series, we are a little skeptical of the benefits of CFAA.
From here, lets move on and take a look at what we actually get in retail products.
The RV6xx parts are similar to the R600 hardware we've already covered in detail. There are a few major differences between the two classes of hardware. First and foremost, the RV6xx GPUs include full video decode acceleration for MPEG-2, VC-1, and H.264 encoded content through AMD's UVD hardware. There was some confusion over this when R600 first launched, but AMD has since confirmed that UVD hardware is not at all present in their high end part.
We also have a difference in manufacturing process. R600 uses an 80nm TSMC process aimed at high speed transistors, while their RV610 and RV630 GPU based cards are fabbed on a 65nm TSMC process aimed at lower power consumption. The end result is that these GPUs will run much cooler and require much less power than their big brother the R600.
Transistor speed between these two processes ends up being similar in spite of the focus on power over performance at 65nm. RV610 is built with 180M transistors, while RV630 contains 390M. This is certainly down from the huge transistor count of R600, but nearly 400M is nothing to sneeze at.
Aside from the obvious differences of transistor count and the number of different units (shaders, texture unit, etc.), the only other major difference is in memory bus width. All RV610 GPU based hardware will have a 64-bit memory bus, while RV630 based parts will feature a 128-bit connection to memory. Here's the layout of each GPU:
One of the first things that jump out is that both RV6xx based designs feature only one render back end block. This part of the chip is responsible for alpha (transparency) and fog, dealing with final z/stencil buffer operations, sending MSAA samples back up to the shader to be resolved, and ultimately blending fragments and writing out final pixel color. Maximum pixel fill rate is limited by the number of render back ends.
In the case of both current RV6xx GPUs, we can only draw out a maximum of 4 pixels per clock (or we can do 8 z/stencil-only ops per clock). While we don't expect extreme resolutions to be run on these parts (at least not in games), we could run into issues with effects that make heavy use of MRTs (multiple render targets), z/stencil buffers, and antialiasing. With the move to DX10, we expect developers to make use of the additional MRTs they have available, and lower resolutions benefit from AA more than high resolutions as well. We would really like to see higher pixel draw power here. Our performance tests will reflect the fact that AA is not kind to AMD's new parts, because of the lack of hardware resolve as well as the use of only one render back end.
Among the notable features that we will see here are tessellation, which could have an even larger impact on low end hardware for enabling detailed and realistic geometry, and CFAA filtering options. Unfortunately, we might not see that much initial use made of the tessellation hardware, and with the reduced pixel draw and shading power of the RVxx series, we are a little skeptical of the benefits of CFAA.
From here, lets move on and take a look at what we actually get in retail products.
Facebook
Twitter
RSS
96 Comments
View All Comments
Slaimus - Thursday, June 28, 2007 - link
Maybe that was the problem. They built their architecture for maximum theoretical performance rather than practical performance.Console hardware are very picky about the type of code they run. In the 360, the developers can organize their shaders in a way the VLIW units like.
Computer games are designed to run on a variety of hardware, so Nvidia's more flexible approach will be faster for more games.
EODetroit - Thursday, June 28, 2007 - link
All of this lacks relevance anyways... there's no Direct X 10 games out. By now we've all been burned by the "future-proofing our PCs" con so many times that we don't fall for it any more. Yes, at some point in the next 1-4 years, most new games will require DX10, and THEN maybe these video cards will be important. For now and for the next year or so, if you can't afford an 8800 you should just get something from the DX9 generation.titan7 - Saturday, June 30, 2007 - link
Company of Heroes was the world's first d3d10 game and it's been out for a month. Pretty sweet game too, Game of the Year 2006 and the highest rated RTS ever on gamerankings and metacritic.Lost Planet I think was a week behind CoH. And now nou can even download Call of Juaez from the game's website.
d3d10 is here.
yacoub - Thursday, June 28, 2007 - link
What an absolute joke the videocard industry has become in the last six months or so.
If you judge by performance in games, there is no mid-range aside from the 320MB 8800GTS and yet that has not dropped far enough in price fast enough to hit the $250 mark without needing rebates. The 640MB 8800GTS remains very near its MSRP, which is even more frustrating because it has been out since last November and yet still hovers around $400 (particularly because AMD didn't release a strong enough competitor but we'll get to that in a moment).
Then the 8600GTS came out without enough streamprocessors and bus bandwidth to perform the way a proper mid-range card would have. Who loses? Only the consumer because AMD is only just now about to get their competing product to store shelves so NVidia simply doesn't care to offer a worthwhile midrange card.
Then AMD released the HD 2900XT and it didn't match the 8800GTX so it didn't influence a price drop on the 8800GTSes by NVidia so the consumer lost out yet again.
Now AMD releases another crap mid-range card to compete with NVidia's crap mid-range card, even though this was the perfect opportunity to release a STRONG mid-range card that threatened the 320MB GTS in at least some games while offering the proper 256-bit bus and enough shaders and stream processors to handle modern games at your average 1600x and 1680x resolutions that most of us with 19" and 20" monitors run.
Really, for $250 a gamer should be able to buy a card that can run your average game at your average resolution with your average AA and AF settings.
My example would be that a modern FPS like Stalker should be able to be run full graphical settings at 1600x1200 or 1680x1050 with 4xAA and 8xAF at a solid 45fps or better.
(As opposed to 'high-end' which is a larger display at 1920x1200 or 2560x1600 with 8xAA and 16xAF.)
Thoughts?
Gary Key - Thursday, June 28, 2007 - link
I believe the purpose of these cards (2400/2600/8500/8600) are for the OEMs to meet a basic checklist of features for the upcoming back to school and holiday season. They provide excellent video playback features for media centers and "appear" on paper to have all of the features you need to play games in your spare time. For the Dell/Best Buy/HP online crowds, the marketing speak will look impressive and I bet the upcharge to go from iX3100/nv7050/atiX1250 on-board graphics will be about the same if not more as just buying the card itself.
Unfortunately, in my testing for the m-ATX roundup (which starts on Monday,finally), the 2400/8500 series did not provide any noticeable improvements in most cases over the atiX1250/nv7050 based boards while being "better" than the G965, sometimes up to 100% better but of course going from 2fps to 4fps in STALKER is not exactly setting the world on fire but those percentages look impressive. ;-)
I think the true mid-range has been or is in the process of being dropped by AMD/NV in order to improve profits. I believe they think the low end cards will suffice for the Home Office / Media Center crowds and that gamers/enthusiasts will pay up for the high end cards. Considering the dearth of competition at the high end, NV sees no point in reducing prices or rolling out new cards (thus taking current cards downstream) at this time.
I expected AMD to reduce the cost on the 2900XT 512 down to $309 or less with the 1GB cards arriving in order place some pricing pressure on NV, especially after the performance let down of the card but even that is not happening. It surprises me because they could really jump start the sales on the card and gain some market share if they had the card priced in the $269~$309 range, a point at which I think people would think the card had decent price to performance value in the market. That leads me to believe the sweet spot ($200~$300) performance market is dead for now, at least with new products.
Just some of my thoughts....
Spoelie - Thursday, June 28, 2007 - link
Maybe, maybe not. However, introduction of new technology has historically been high end and low end, with midrange only coming around the refresh.For example:
initial release
7600GT < 199$, 7900GT > 300$
x1600xt < 169$, x1900xt > 400$
~2 months later, 256mem configuration was used on the x1900xt to bring the 7900gt some competition at the 300$ pricepoint, but yet again there were no new introductions in the 200-300$ price range, leaving it open to sell yesterdays highend (x1800 7800 x800)
refresh
introduction of 7900gs, gt price dropped and replaced by the 7950gt
x1950pro introduction at 200, x1950xt 256/512 x1950xtx 512 at 250/300/350 respectively
At this point we have some decent mid range, bliss for enthousiasts.
With a little bit of luck, we'll see some nice midrange parts around autumn/winter.
SandmanWN - Thursday, June 28, 2007 - link
I agree that the Mid-Range is dead with the current generation of video cards on both sides of the fence (ATI&NV). The closest thing we have to midrange is the GTS 320 and its still over priced for that category.There is a clear lack of video cards in the $200-300 range. Its easier to find a better performer in the last generation of cards than it is 7-8 months into the current generation.
Wheres the 9800 Pro/x850 Pro of the current generation????
(And please give it a rest on the 2900XT vs GTX. You know its been said about a billion times that the 2900xt competitor is the GTS. Let the pre-production hype die already will yah?)
theprodigalrebel - Thursday, June 28, 2007 - link
The mid-range isn't dead. X1950Pro prices are awesome at the moment. :-)TA152H - Thursday, June 28, 2007 - link
Yet another weird choice for testing. You test a low-end card AMD card, with a top of the line Intel CPU? Yes, a lot of people will be going for that configuration. At least .01 %. Well, close to it anyway. And then, the most interesting part of all, you don't test, the 2400 Pro, which doesn't have a noisy fan and thus at least has something attractive about it.Granted, a high end processor exaggerates the differences in cards, but it's more academic than meaningful, since most people will not have that configuration. As you go to more mainstream processors, the delta shrinks, although in this case it's so dramatic as to be startling.
But, DX9 tests aren't particularly meaningful within the context of what these cards are. Their claim to fame is to be DX10 cards, and one is fanless, so it's going to be attractive to some people. If the DX10 performance is so miserable though, maybe it won't have appeal in that market. That's why I'm confused why you'd even bother with the DX9 article first, instead of going straight to DX10 and testing DX9 later. Clearly, DX10 is the future; it looks better, and was the focus of the product. If it performs relatively well on DX10, DX9 performance (or lack thereof), becomes almost irrelevant since it is not the focus of the card. The reverse would not be true, if DX10 performance were miserable, and DX9 performance good, it would still be a nearly useless card. The DX10 article should have come first.
strikeback03 - Thursday, June 28, 2007 - link
IIRC there were faster fanless cards last generation. I know a fanless 7600GT is still available, and the 7600GT finished ahead of the 2400 XT here, so no doubt it's better than the 2400 Pro.With the DX9 performance of these cards, AMD has certainly not given anyone a reason to upgrade. Assuming DX10 performance is better, users will wait until more DX10 games are available. Why lose performance in an upgrade on the games you currently play?