The ATI Radeon HD 4000 series is slated for late Q2. Information about the high-end RV770 has been leaking onto the web, from here and there, but the exact specifications has remained in the shadows. German hardware site Hartware.net has gotten their hands on what seems to be pretty authentic information. As far as we can tell, it looks believable and matches the rumors going around. We're still far away from the launch and the figures are subject to change.
The information published on hartware.de suggests that ATI will use GDDR5 to a great extent, which surprise us a little. On the other hand, ATI may know something about GDDR5 that we don't (we would be surprised otherwise). RV770 has GDDR5 support, but whether ATI will actually use it remains to be seen. GDDR4 with most of the high-end cards is a highly likely scenario. The frequencies suggested here are actually lower than what good GDDR4 chips are capable of.
RV770 will break 1TFLOPS, while the R700, with two RV770, should be capable of twice that, more than 2TFLOPS. This is higher than what he had expected, but then again the core frequencies are also higher than expected. All cores are made by TSMC and their 55nm process. We're still trying to hunt down more reliable information on the design of the "core".
When AMD/ATI, unveiled the Radeon HD 3000 series they had added a nifty feature called PowerPlay. PowerPlay dynamically controls the power consumption and will severely reduce the power consumption when idle, but also reduce the peak consumption somewhat. When we look at the figures they don't quite add up, but as stated above, these figures are all most likely going to change before the launch. The idle power consumptions of all cards are extremely low and even the dual-core Radeon HD 4870X2 will consume less than 25W when idle.
The number of TMUs have finally been increased, actually doubled, to 32 with RV770, while RV740 sports 24 TMUs, and the low-end RV710 has 8 TMUs. Don't bother getting upset with the 256-bit memory bus because it will not become any kind of bottleneck. The only time it might be even close to that is when you're pouring liquid nitrogen into the container and have the GPU running at +50% of stock frequency, which will happen though.
To end this long and speculative post, we would like to say that if not all, most of the variables here are subject to change and should be considered with a "tiny" pinch of salt. We're still 4 month away from the launch.
Bookmarks