AMD's Radeon HD 5870: Bringing About the Next Generation Of GPUs
by Ryan Smith on September 23, 2009 9:00 AM EST- Posted in
- GPUs
A Quick Refresher on the RV770
As Cypress is a direct evolution of the RV770 design, before we talk about what’s new with Cypress we are going to go over a quick rehash of RV770’s internal workings. As it’s necessary to understand how RV770 was built to understand what Cypress changes, if you’re completely unfamiliar with RV770, please take a look at our expanded discussion of RV770 from last year. For the rest of you, let’s get started.
At the center of the RV770 is the Stream Processing Unit (SPU), a single arithmetic logic unit. The RV770 has 800 of these, and they are packaged together in groups of 5 and are what we call a Streaming Processor (SP). A SP contains a register file, a branch predictor, and the aforementioned 5 SPUs, with the 5th SPU being a more complex unit capable of transcendental functions along with the base functions of an ALU. The SP is the smallest unit that can do individual work; every SPU in an SP must execute the same instruction.
For every 16 SPs, AMD groups them together with texture units, L1 cache, shared memory, and controlling logic. This combined block is what AMD calls a SIMD, and RV770 has 10 of them. These 10 SIMDs form the core computational power of the RV770, and in the chip work with various specialized units such as ROPs, rasterizers, L2 cache, and tesselators to form a complete chip.
To utilize the computational power of the hardware, instruction threads are issued to the SPs. These threads are grouped into wavefronts, where there are 64 threads per wavefront. To maximize the utilization of the GPU, threads need to be organized so that they can feed all 5 SPUs in a SP an instruction every clock cycle. Doing this requires extracting instruction level parallelism (ILP) out of programs being passed to the GPU, which is difficult task of AMD’s compiler.
If SPUs go unused, then the performance of the chip suffers due to underutilization. This design gives AMD a great deal of theoretical computational power, but it is always a challenge to fully exploit it.
Facebook
Twitter
RSS
327 Comments
View All Comments
Xajel - Wednesday, September 23, 2009 - link
Ryan,I've send a detailed solution for this problem to you ( Aero disabled when running and video with UVD accerelation ), but the basic for all readers here is just install HydraVision & Avivo Video Convertor packages, and this should fix the problem...
Ryan Smith - Wednesday, September 23, 2009 - link
Just so we're clear, Basic mode is only being triggered when HDCP is being used to protect the content. It is not being triggered by just using the UVD with regular/unprotected content.biigfoot - Wednesday, September 23, 2009 - link
Well, I've been looking forward to reading your review of this new card for a little while now, especially since the realistic sounding specs leaked out a week or so ago. My first honest impression is that it looks like it'll be a little bit longer till the drivers mature and the game developers figure out some creative ways to bring the processor up to its full potential, but in the mean time, it looks like I'd have to agree with everyone's conclusion that even 150+GB/s isn't enough memory bandwidth for the beast, I'm sure it was a calculated compromise while the design was still on the drawing board. Unfortunately, increasing the bus isn't as easy as stapling on a couple more memory controllers, they probably would've had to resort back to a wider ring bus and they've already been down that road. (R600 anyone) Already knowing how much more die area and power would've been required to execute such a design probably made the decision rather easy to stick with the tested 4 channel GDDR5 setup that worked so well for the RV770 and RV790. As for how much a 6 or 8 channel (384/512 bit) memory controller setup would've improved performance, we'll probably never know; as awesome as it would be, I don't foresee BoFox's idea of ATI pulling a fast one on nVidia and the rest of us by releasing a 512-bit derivative in short order, but crazier things have happened.biigfoot - Wednesday, September 23, 2009 - link
Oh, I also noticed, they must of known that the new memory controller topology wasn't going to cut it all the time judging from all the cache augmentations performed. But like i said, given time, I'm sure they'll optimize the drivers to take advantage of all the new functionality, I'm betting that 90% of the low level functions are still handled identically as they were in the last generations architecture and it will take a while till all the hardware optimizations like the cache upgrades are fully realized.piroroadkill - Wednesday, September 23, 2009 - link
The single most disappointing thing about this card is the noise and heat.Sapphire have been coming out with some great coolers recently, in their VAPOR-X line. Why doesn't ATI stop using the same dustbuster cooler in a new shiny cover, and create a much, much quieter cooler, so the rest of us don't have to wait for a million OEM variations until there's one with a good cooler (or fuck about and swap the cooler ourselves, but unless you have one that covers the GPU AND the RAM chips, forget it. Those pathetic little sticky pad RAM sinks suck total donkey balls.)
Kaleid - Wednesday, September 23, 2009 - link
Already showing up...http://www.techpowerup.com/104447/Sapphire_HD_5870...">http://www.techpowerup.com/104447/Sapphire_HD_5870...
It will be possible to cool the card quietly...
Dante80 - Wednesday, September 23, 2009 - link
the answer you are looking for is simple. Another, more elaborate cooler would raise prices more, and vendors don't like that. Remember what happened to the more expensive stock cooler for the 4770? ...;)Cookie Monster - Wednesday, September 23, 2009 - link
When running dual/multi monitors with past generation cards, the cards would always run at full 3d clocks or else face instabilities, screen corruptions etc. So even if the cards are at idle, it would never ramp down to the 2d clocks to save power, rending impressive low idle power consumption numbers useless (especially on the GTX200 series cards).Now with RV870 has this problem been fixed?
Ryan Smith - Wednesday, September 23, 2009 - link
That's a good question, and something we didn't test. Unfortunately we're at IDF right now, so it's not something we can test at this moment, either.Cookie Monster - Wednesday, September 23, 2009 - link
It would be awesome if you guys do get some free time to test it out. Would be really appreciated! :)