5 - Physics | CrossFireX
What about Physics ?
PhysX - Much like NVIDIA CUDA based PhysX implementation, ATI (AMD) recently announced cooperation with Intel's HAVOC engine. Though currently far less substantial, PhysX calculations over the GPU are in the work. As it works right now (example debris/cloth) physics calculations are computed over the CPU with games that support the HAVOK API. AMD is working on moving these functions to the GPU. Thus have the stream processors (shade engine) compute these functions. It's work in progress and during a recent press-briefing we asked when we can expect driver support for GPU HAVOK physics. The answer was unfortunately a bit cold. It could be a matter of months, yet also easily be the end of the year or even later. Fact remains though that the Series 4000 do support the feature and AMD considers this "work in progress" and my general feeling is that they are certainly not hasting to get it supported.
Hover with your mouse over this photo to magnify.
CrossFireX
The product we preview today is based on the 4870, yet it has two GPUs merged together with a bridge chip utilizing crossfire technology to render your games faster. "What's Crossfire?" some of you might ask. A valid question as we take verbs like Crossfire & SLI for granted these days.
Well, just like NVIDIA's SLI, Crossfire is a situation where you add a second, third or even fourth similar generation graphics card (or in today's case GPU) to the one you already have in your PC and effectively try to double, triple, quadruple your raw rendering / gaming performance.
The idea is not new at all though .. if you are familiar with the hardware developments over the past years you'll remember that 3dfx had a very familiar concept with the Voodoo 2 graphics cards series. There are multiple ways to manage two cards rendering one frame, think of Supertiling, it's a popular form of rendering. Alternate frame Rendering, each card will render a frame (even/uneven) or Split Frame rendering, simply one GPU renders the upper or the lower part of the frame. So you see there are many methods where two or more GPUs can be utilized to bring you a gain in performance.
Today we take it even a step further. Though the X2 already has two GPUs, there's no limit to stop us combining it with yet another X2. So we will take two Radeon HD X 4870 cards, bridge them and get 4 GPUs going. So that's 4 RV770 processors running at 750 MHz. That equals 4.8 TFLOPS of performance!
Unfortunately it's also a power sucking whore in this mode, but none the less we'll look at it as well.
Well, we tried, that's for sure.
Before we dive into the photo-shoot, tests and benchmarks. Let's just compile a chart and look at the differences:
ATI Radeon |
ATI Radeon |
Radeon |
Radeon | |
# of transistors |
965 million |
965 million |
965 million x2 |
965 million x2 |
Stream Processing Units |
800 |
800 |
800x2 |
800x2 |
Clock speed |
625 MHz |
750 MHz |
650 MHz |
750 MHz |
Memory Clock |
2000 MHz GDDR3 (effective) |
3600 MHz GDDR5 (effective) |
2.0 GHz GDDR3 (effective) |
3600 MHz GDDR5 (effective) |
Math processing rate (Multiply Add) |
1000 GigaFLOPS |
1200 GigaFLOPS |
2000 GigaFLOPS |
2400 GigaFLOPS |
Texture Units |
40 |
40 |
40x2 |
40x2 |
Render back-ends |
16 |
16 |
16x2 |
16x2 |
Memory |
512MB GDDR3 |
512MB GDDR5 |
1024MB GDDR3 x2 |
1024 GDDR5 x2 |
Memory interface |
256-bit |
256-bit |
256-bit x2 |
256-bit x2 |
Fabrication process |
55nm |
55nm |
55nm |
55nm |
Power Consumption (peak) |
~110W |
~160W |
~230W |
~286W |
And that last R700 product to the right we'll look at as AMD want to reveal a bit of what they are working on. Anyway, let's have a look at the product and talk some more about installation, heat, power consumption and so on.