AMD Tahiti (HD7900) GPUs Ready For Market

On January 3 we published an overview of the AMD HD7000 family of GPUs. The first series to be launched are the HD 7900-series (TAHITI) GPUs, of which the HD7970 was launched on paper on December 22. HD7950 is due in the first week of February, and HD7990 is currently unannounced. For fans of AMD graphics cards, this is very good news, as the HD7970 is faster than any current single-gpu graphics card.

The Radeon HD7970 is due to be the fastest single-GPU card available when it is released on January 9. Most graphics benchmarks that have turned up in reviews in advance of the release score HD7970 higher than other cards, but the most impressive data is in it's compute performance. The peak single precision FP performance is 3.8 TFLOPS (at the default clock sped of 925 MHz), with double precision performance coming in at a staggering 947 GFLOPS. Various reviews indicate that the cards will overclock by at least 10% without any voltage adjustments, meaning that a DP FP performance of over 1 TFLOP is easily achievable, with appropriately accelerated software.

HD7970 has 32 compute units (described in our first related article), 2048 shaders and the maximum 768 KB level 2 cache that the platform supports. This level 2 cache sits between the memory controller and the shared level 1 cache within each compute unit. Combined with up to 3 GB GDDR5 memory, the available bandwidth is about 260 GB/s. To round it off, there is a Global Data Share module which enables level 1 cache synchronization across all the compute units.

Moving to more general information, the card has 8+6 pin PCI-E power connectors, providing up to 300W, and uses a PCI-E 3 interface. The reference design has 3-way crossfire connectors, and a dual BIOS switch for those who want to experiment with a modified BIOS. A dual-link DVI connector, one HDMI port and two Mini-DP ports provide support for up to 6 displays.

Moving on to benchmarks, I'm not going to detail everything I've seen but overall most benchmarks at VR-Zone show the HD7970 to be faster than any other single-gpu card, and in some cases faster than the best current dual-gpu cards or SLI setups. What's especially interesting for anyone looking for the best compute performance they can get is that the HD7970 at stock speeds beats an HD6990 and a GTX 580 SLI configuration, when measured using ComputeMark. When we see details of the HD7990, I expect it to walk all over the competition!

Power consumption both when idle and when playing Battlefield 3 (one of the most gpu intensive games currently on the market) is lower than other top-end cards. At idle, the power usage is 30W less than a GTX 580, 40W less than Radeon HD 6970 and 90W less than Radeon HD 6990. Under load, the saving is 20W over the GTX 580, 110W compared to HD6970 and a massive 280W compared to HD6990.

Specs for the HD7990 and HD7950 are not yet released. However, the HD7950 is currently expected in the first week of February, and it is thought that the HD7990 might be launched in March. HD7990 has 2 HD7970 cores, but what speed they will run at is not yet known.

The HD7990 is expected to cost more than $700, while the HD7970 is expected to be $499. The final product in the Tahiti range is HD7950, at around $399.

More information:
Related News (newer articles):
Related News (older articles):
Comments: 0
Terms and Conditions · Privacy Policy · Contact Us (c) Copyright 2003 - 2010 Gennadiy Shvets

Search CPU-World

Search site contents:

Identify part

Identify CPU, FPU or MCU: