Review AMD Radeon HD 7970 and Nvidia GTX 680 AIB

Posted by Jon Peddie on May 2nd 2012 | Discuss
Categories: Hardware Review
Tags: nvidia amd graphics ati 3d review

The world’s two fastest graphics boards are compared—GTX 680 wins

A month or so ago AMD released their top-of-the-line Radeon HD 7079. Demos were sent out to the company’s favorite reviewers (that means NOT you JPR) and results started being reported, it looked good. Then a week or two ago Nvidia released its GeForce GTX 680 Kepler (Yes JPR, even you can have one), and the comparisons began. In the meantime, by trading some vintage wine and rare out-of-print copies of TechWatch we were able to get a black market HD 7970 (no thanks to AMD). And so we ran some tests on the two AIBs in a Sandy Bridge PC and the new Ivy Bridge PC (can’t even hint about how we got one of them).

These are state of the art high-end machines, which will get out of the way of any game, even one that is CPU limited. They are also remarkably quiet machines, even when heavily loaded.

The tests

We ran five benchmarks on the AIBs: 3D­mark 11, Unigine Heaven, Batman Arkin, Crysis 2, and Dirt. We ran single AIB, and SLI with the Nvidia boards because we actually had two GTX680s. AMD’s Catalyst 12.3 and Nvidia’s GeForce R301.1 drivers were used for the tests.

The results were very interesting, which if you have the stamina to trudge through this dribble you’ll see.

Comparison of AMD’s AIBs
  Radeon HD
7970
Radeon HD
6970
Architecture Tahiti Barts
Stream Processors 2048 1536
Core Clock MHz 925 880
Memory Clock GHz 1.375 1.375
Memory Bus bits 384 256
Memory bandwidth 264 176
Memory size GB 3 2
Memory type GDDR5 GDDR5
Texture Units 128 96
ROPs 32 32
Transistor Count (B) 4.31 2.64
Process (nm) 28 40
Wattage 195 250
Price $549 $350
Test beds used in benchmarking AMD and Nvidia’s new AIBs
Test bed Ivy Bridge Sandy Bridge
Processor Intel Core i7-3770K @ 3.5GHz 3.30 gigahertz Intel Core i7-3960X
Chipset/mobo DZ77GA-70K (Gaspar) Intel Corporation DX79SI AAG28808-500
RAM 16GB Ram 8GB RAM
Drive(s) Intel SSD SC2CW240A3, HDD 240 GB 249.68 GB HDD
Power supply AX850 Corsair Powersupply 850W Thermaletake
OS Windows 7 64-bit Windows 7 64-bit

AMD Radeon HD7970 Tahiti

AMD’s Radeon HD 7970 AIB is based on the Tahiti generation of GPUs. The Tahiti is a 4.31 billion transistor chip with 2,048 shader cores in a 365mm2 package manufactured at TSMC in their 28nm process. The AIB’s base clock is 925 MHz, and can be overclocked to 1200 MHz. We did not run any tests in over-clock mode, but if you wanted to, the easiest way to begin overclocking the Radeon HD 7970 is to use AMD’s built in Overdrive option in Catalyst Control Center.

The HD7970 has 3GB of GDDR5 running at 1375 MHz clock or 3Gbps. The overall power consumption of the chip is 195w. The Tahiti uses less power than the previous Barts-based GTX580 (GF110) which is a 520mm2 chip with 3 billion transistors (500M fewer), 512 shader cores with a clock speed of 880MHz, and a power draw of 250w. The Tahiti chip is 43 percent smaller than the Barts and has 63 percent more transistors while increasing the amount of shader cores by 33.3% and reducing the power consumption by 8 percent.

AMD Radeon HD 7970

The Radeon HD 7970 offers a dual-link DVI, an HDMI, and two mini display port connectors allowing support of four monitors or more using daisy-chaining display port monitors.

Nvidia GeForce GTX 680 Kepler

 

Nvidia’s GTX 680 AIB is based on the GK104 Kepler GPU generation. The GK104 is a 3.54 billion transistor chip with 1,536 shader cores in a 294mm2 package manufactured at TSMC in their 28nm process. The AIB’s base clock is 1006 MHz, and can be increased to 1058 MHz using Nvidia’s GPU Boost, which lets portions of the GPU overclock itself when there is a light load on the chip.

The GTX 680 has 2 GB of GDDR5 running at 1500 MHz clock or 6 Gbps. The overall power consumption of the chip is 195w. The GTX680 uses less power than the previous Fermi-based GTX 580 (GF110) which is a 520mm2 chip with 3 billion transistors (500m fewer), 512 shader cores with a clock speed of 772 MHz, and a power draw of 244w. The GK104 chip is 43 percent smaller than the GF110 and has 17 percent more transistors while tripling the amount of shader cores and reducing the power consumption by 20 percent.

The GTX 680, which is 10-inche long, is just over half an inch shorter than its closest competitor, the Radeon HD 7970.

Nvidia GeForce GTX 680

The GTX 680 offers two DVI connectors; one is dual-link, an HDMI, and one display port connector.

Test results

Nvidia’s AIBs comparison
  GTX 680 GTX 580
Architecture Kepler Fermi
Cuda Cores 1536 512
Core Clock MHz 1000 772
Memory Clock MHz 3000 2000
Memory bus bits 256 384
Memory bandwidth 192 192
Memory size GB 2 1.5
Memory type GGDR5 GDDR5
Texture units 128 64
ROPs 32 48
Transistor count (B) 3.54 3
Process (nm) 28 40
Wattage 195 244
Price $524 $400
Differences between AMD and Nvidia’s GPU specifications
  AMD HD 7970/
Nvidia GTX 680
AMD HD 6970/
Nvidia GTX 580
Stream 33.3% 200.0%
Core Clock MHz -7.5% 14.0%
Memory Clock GHz -54.2% -31.3%
Memory Bus bits 50.0% -33.3%
Memory bandwidth 37.5% -8.3%
Memory size GB 50.0% 33.3%
Texture Units 0.0% 50.0%
ROPs 0.0% -33.3%
Transistor Count (B) 21.8% -12.0%
Process (nm) 0.0% 0.0%
Wattage 0.0% 2.5%
Price 4.8% -12.5%

As mentioned we ran five tests (3Dmark 11, Heaven, Batman, Crysis 2, and Dirt 3), we ran those tests on four AIBs (GTX580, GTX680, HD6970, and HD7070), and we ran them on two machines (SNB and IVB), we also ran the tests on the Ivy Bridge (IVB) PC. With that data we calculated the Pmark for each AIB on individual benchmarks and as an average of the benchmarks. We also ran the GTX 680 in SLI mode. On the IVB machine we couldn’t run a second GTC 680 because of the limitations of the power supply in the system and a lack of time to replace the supply.

In all tests, which were run at 1920 x 1080 4XAA, Nvidia’s GeForce GTX 680 got the highest Pmark scores.

In all but one test the Nvidia GeForce GTX got the best frames per second score, as shown in the following chart.

Given the differences in specifications, with AMD having more of almost everything, you would have expected AMD to do better than Nvidia.

The differences were calculated by dividing AMD’s parameters by Nvidia’s. For example, AMD has 2,048 processors in their HD7970 and Nvidia has 1,536 so the difference is 33.3%. The first column is the comparison of the AMD HD 7970 to the Nvidia GTX680, and the second column is the comparison of the AMD HD 6970 to the Nvidia GTX580.

Although AMD has more memory bandwidth than Nvidia, which should favor results, Nvidia runs all of the clocks (memory and GPU) at a higher speed than AMD, and that seems to account for the difference in performance.

What do we think?

AMD and Nvidia are close competitors, always have been and always will be. For this round Nvidia is the champ. That can change in 90 days, and no matter who is the champ, the consumers never lose, both companies offer great products.

We think the Pmark scores tell the real story, and so the prize goes to Nvidia for having a clean sweep of the best Pmark scores across five benchmarks.

Discuss this entry