ATI Radeon HD 2900 XT: Calling a Spade a Spade
by Derek Wilson on May 14, 2007 12:04 PM EST- Posted in
- GPUs
Stream Processor Implementation
Going Deeper: Single Instruction, Multiple Data
SIMD (single instruction, multiple data) is the concept of running one instruction across lots of data. This is fundamental in the implementation of graphics hardware: multiple vertices, primitives, or pixels will need to have the same shader program run on them. Building hardware to do one operation at a time on massive amounts of data makes processing each piece of data very efficient.
In SIMD hardware, multiple processing units are tied together. The hardware issues one instruction to the SIMD hardware and all the processing units perform that operation on unique data. All graphics hardware is built on this concept at some level. Implementing hardware this way avoids the complexity of requiring each SP to manage not only the data coming through it, but the instructions it will be running as well.
Going Deeper: Very Long Instruction Word
Normally when we think about instructions on a processor, we think about a single operation, like Add or Multiply. But imagine if you wanted to run multiple instructions at once on a parallel array of hardware. You might come up with a technique similar to VLIW (Very Long Instruction Word), which allows you to take simple operations and, if they are not dependent on each other, stick them together as one instruction.
Imagine we have five processing units that operate in parallel. Utilizing this hardware would require us to issue independent instructions on each of the five units. This is hard to determine while code is running. VLIW allows us to take the determination of instruction dependence out of the hardware and put it in the complier. The compiler can then build a single instruction that consists of as much independent processing work as possible.
VLIW is a good way of exploiting parallelism without adding hardware complexity, but it can create a huge headache for compiler designers when dealing with dependencies. Luckily, graphics hardware lends itself well to this type of processing, but as shaders get more complex and interesting we might see more dependent instructions in practice.
Bringing it Back to the Hardware: AMD's R600
AMD implements their R600 shader core using four SIMD arrays. These SIMD arrays are issued 5-wide (6 with a branch) VLIW instructions. These VLIW instructions operate on 16 threads (vertices, primitives or pixels) at a time. In addition to all this, AMD interleaves two different VLIW instructions from different shaders in order to maximize pipeline utilization on the SIMD units. Our understanding is that this is in order to ensure that all the data from one VLIW instruction is available to a following dependent VLIW instruction in the same shader.
Based on this hardware, we can do a little math and see that R600 is capable of issuing up to four different VLIW instructions (up to 20 distinct shader operations), working on a total of 64 different threads. Each thread can have up to five different operations working on it as defined by the VLIW instruction running on the SIMD unit that is processing that specific thread.
For pixel processing, AMD assigns threads to SIMD units in 8x8 blocks (64 pixels) processed over multiple clocks. This is to enable a small branch granularity (each group of 64 pixels must follow the same code path), and it's large enough to exploit locality of reference in tightly packed pixels (in other words, pixels that are close together often need to load similar data/textures). There are apparently cases where branch granularity jumps to 128 pixels, but we don't have the data on when or why this happens yet.
If it seems like all this reads in a very complicated way, don't worry: it is complex. While AMD has gone to great lengths to build hardware that can efficiently handle parallel data, dependencies pose a problem to realizing peak performance. The compiler might not be able to extract five operations for every VLIW instruction. In the worst case scenario, we could effectively see only one SP per block operating with only four VLIW instructions being issued. This drops our potential operations per clock rate down from 320 at peak to only 64.
On the bright side, we will probably not see a shader program that causes R600 to run at its worst case performance. Because vertices and colors are still four components each, we will likely see utilization closer to peak in many common cases.
Facebook
Twitter
RSS
86 Comments
View All Comments
GoatMonkey - Monday, May 14, 2007 - link
That's obviously BS. This IS their high end part, it just doesn't perform as well as nVidia's high end part, so it is priced accordingly.poohbear - Monday, May 14, 2007 - link
sweet review though! thanks for including all the important and pertinent cards in your roundup (the 8800gts 320mb inparticular). also love how neutral Anand is in their reviews, unlike some other sites.:pCreig - Monday, May 14, 2007 - link
The R600 is finally here. I'm sure the overall performance is not what AMD was hoping for. Nobody ever shoots to have their newest product be the 2nd best. But pricing it at $399 and including a very nice game bundle will make the HD 2900 XT a VERY worthwhile purchase. I also have the feeling that there is a significant amount of performance increase to be realized through future driver releases ala X1800XT.shady28 - Tuesday, May 15, 2007 - link
Nvidia has gone over the cliff on pricing.
I know of no one personally who has an 88xx series card. I know one who recently picked up an 8600 of some kind, that's it. I have the best GPU of anyone I know.
It's a real shame that there is so much focus on graphics cards that virtually no one buys. These are niche products folks - yet 'who is best' seems to be totally dependent on these niche products. That's patently ridiculous.
It's like saying, since IBM makes the fastest computers in the world (they do), they're the best and you should be buying IBM (or now, lenovo) laptops and desktops.
No one ever said that sort of thing because it's patently ridiculous. Why do people say it now for graphics cards? The fact that they do says a lot about the mentality of sites like AT.
DerekWilson - Tuesday, May 15, 2007 - link
We don't say what you are implying, and we are also very upset with some of NVIDIA's pricing (specifically the 8800 ultra)the 8800 gts 320mb is one of the best values for your money anywhere and isn't crazy expensive -- it's actually the card I'd recommend to anyone who cares about graphics in games and wants good quality and performance at 1600x1200.
I would never tell anyone to buy an 8600 gts because nvidia has the fastest high end card. In fact, in this article, I hope I made it clear that AMD has the opportunity to capitalize on the huge performance gap nvidia left between the 8600 and 8800 series ... If AMD builds a part that performs in this range is priced competitively, they'll have our recommendation in a flash.
Recommending parts based on value at each price or performance segment is something we take pride in and will always do, no matter who has the absolute fastest hardware out there.
The reason our focus was on AMD's fastest part is because they haven't given us any other hardware to test. We will absolutely be talking a lot and in much depth about midrange and budget hardware when AMD makes these parts available to us.
yacoub - Monday, May 14, 2007 - link
$400 is a lot of money. Not terribly long ago the highest end GPU available didn't cost more than $400. Now they hit $750 so you start to think $400 sounds cheap. It's really not. It's a heck of a lot of money for one piece of hardware. You can put together a 650i SLI rig with 2GB of DDR2 6400 and an E4400 for that much money. I know because I just did that. I kept my 7900GT from my old rig because I wanted to see how R600 did before purchasing an 8800GTS 640MB. Now that we've seen initial results I will wait to see how R600 does with more mature drivers and also wait to see the 640MB GTS price come down even more in the meantime.vijay333 - Monday, May 14, 2007 - link
http://www.randomhouse.com/wotd/index.pperl?date=1...">http://www.randomhouse.com/wotd/index.pperl?date=1..."the expression to call a spade a spade is thousands of years old and etymologically has nothing whatsoever to do with any racial sentiment."
yacoub - Monday, May 14, 2007 - link
Yes, a spade was a shovel long before muslims enslaved europeans to do hard labor in north africa and europeans enslaved africans to do hard labor in the 'new world'.vijay333 - Monday, May 14, 2007 - link
whoops...replied to the wrong one.rADo2 - Monday, May 14, 2007 - link
It is not 2nd best (after 8800ULTRA), not 3rd best (after 8800GTX), not 4th best (after 8800GTX-640), but 5th best (after 8800GTS-320), or even worse ;)Bad performance with AA turned on (everybody turns on AA), huge power consumption, late to the market.
A definitive failure.