We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.
[nextpage title=”Introduction”]
AMD is launching today their latest ultra-high-end video card, the Radeon HD 6990, a dual-GPU solution that comes priced at USD 700. Let’s check its performance.
The video card we are reviewing is the reference model from AMD. When a video card is first launched, all “manufacturers” buy their video cards already assembled from AMD and only add their sticker to it. One or other manufacturer may add an overclocking, but physically all cards are absolutely identical. Only after a while manufacturers start launching customized solutions, changing the cooler and, sometimes, redesigning the printed circuit board.
The new Radeon HD 6990 arrives with an impressive USD 700 price tag, which puts it as the most expensive video card available in the market today. It has two GPUs, and this sort of explains the awful price tag. At this price tag, it doesn’t have a direct competitor – the most expensive model from NVIDIA, the GeForce GTX 580, can be found between USD 500 and USD 520 today.
So, the direct competitor for the Radeon HD 6990 would be two GeForce GTX 570 (USD 350 to USD 370 each) connected in SLI, and this will be the setup we will consider to be the Radeon HD 6990’s main competitor. Of course we are going to compete this new video card to the GeForce GTX 580 and, since we had already collected the data, we will also compare it to two Radeon HD 6870 connected in CrossFireX mode, although this setup does not compete with the Radeon HD 6990. We are also going to compare the Radeon HD 6990 to its predecessor, the Radeon HD 5970 (unfortunately we had already returned this video card and didn’t have the results for the 3DMark 11). Comparing the Radeon HD 6990 to two GeForce GTX 580 or two Radeon HD 5970 makes no sense, since we would be comparing a USD 700 system to a USD 1,000+ system.
In the table below we compare the main specs of the video cards included in our review. As mentioned, we connected two Radeon HD 6870 and two GeForce GTX 570 in parallel, and the specs below are for only one video card. The Radeon HD 5970 and the Radeon HD 6990 have two GPUs, and the specs below are for only one of the GPUs. The Radeon HD 6990 has two clock settings, selectable through a switch on the card. The number in parenthesis indicate the position of this switch.
| Video Card | Core Clock | Shader Clock | Memory Clock (Real) | Memory Clock (Effective) | Memory Interface | Memory Transfer Rate | Memory | Shaders | Price |
| GeForce GTX 570 | 732 MHz | 1,464 MHz | 950 MHz | 3.8 GHz | 320-bit | 152 GB/s | 1.28 GB GDDR5 | 480 | USD 350 – 370 |
| GeForce GTX 580 | 772 MHz | 1,544 MHz | 1,002 MHz | 4,008 MHz | 384-bit | 192.4 GB/s | 1.5 GB GDDR5 | 512 | USD 500 – 520 |
| Radeon HD 6870 | 900 MHz | 900 MHz | 1.05 GHz | 4.2 GHz | 256-bit | 134.4 GB/s | 1 GB GDDR5 | 1,120 | USD 220 – 240 |
| Radeon HD 5970 | 725 MHz | 725 MHz | 1 GHz | 4 GHz | 256-bit | 128 GB/s | 1 GB GDDR5 | 1,600 | USD 630 – 700 |
| Radeon HD 6990 | 830 MHz (2) or 880 MHz (1) | 830 MHz (2) or 880 MHz (1) | 1.25 GHz | 5 GHz | 256-bit | 160 GB/s | 2 GB GDDR5 | 1,536 | USD 700 |
Prices were researched at Newegg.com on the day we published this review, except the Radeon HD 5970, which was researched at Google Shopping, and the Radeon HD 6990, which we are using the manufacturer suggested retail price. All graphics chip listed above are DirectX 11 parts.
You can compare the specs of these video cards with other video cards by taking a look at our AMD ATI Chips Comparison Table and NVIDIA Chips Comparison Table tutorials.
This card is rated to pull up to 375 W from the power supply. It is important to understand that some people are making the mistake of saying that this is a “450 W” video card. What happens is that the cooler used by this video card can dissipate up to 450 W, but that doesn’t mean that the video card pulls 450 W from the power supply.
Now let’s take an in-depth look at the Radeon HD 6990 reference model.
[nextpage title=”The AMD Radeon HD 6990″]
Below we have an overall look at the AMD Radeon HD 6990 reference model.
Figure 1: AMD Radeon HD 6990
Figure 2: AMD Radeon HD 6990
This video card has one DVI-D and four mini DisplayPort connectors. All Radeon HD 6990 video cards will ship with three adapters: one mini DisplayPort to single-link DVI-D (passive adapter), one mini DisplayPort to single-link DVI-D (active adapter) and one mini DisplayPort to HDMI (passive adapter). This will allow you to use the mini DisplayPort connectors even if your video monitor doesn’t support this kind of connection, which is probably the case.
Figure 3: Video connectors
The Radeon HD 6990 requires two eight-pin power connectors, shown in Figure 4.
Figure 4: Power connectors
[nextpage title=”The AMD Radeon HD 6990 (Cont’d)”]
The Radeon HD 6990 comes with two ROM chips, selectable through a switch located near the CrossFireX connector, see Figure 5. The card comes with this switch on the position “2,” which loads the default clocks and voltages of the video card (830 MHz, 1.12 V). When moved to the position “1,” the video card is overclocked to 880 MHz and voltage is increased to 1.175 V. We will test this card with its switch in both positions.
Figure 5: Dual-BIOS switch
In Figure 6, you can see the video card with its cooler removed.
Figure 6: Video card with the cooler removed
Figure 7: Voltage regulator
The GPU cooler can be seen in Figure 8. It uses two separate heatsinks, one for each GPU. The heatsinks use vapor chamber technology, which is a technology similar to the one used on heatpipes. According to the manufacturer, this GPU cooler can dissipate up to 450 W.
Figure 8: The GPU cooler
Each GPU is connected to eight 2 Gbit GDDR5 chips, for a total of 2 GB per GPU (2 Gbit x 8 = 2 GB) or 4 GB total. Each chip is connected to the GPU using a 32-bit data lane, making the video card’s 256-bit memory interface (32 bits x 8 = 256) per GPU.
The chips used are H5GQ2H24MFR-T2C parts from Hynix, which support up to 1.25 GHz (5 GHz QDR) and since on this video card memory is accessed at 1.25 GHz (5 GHz QDR), there is no margin for you to increase the memory clock rate while keeping the chips inside the maximum they support. Of course you can always try to overclock the memory chips above their specs.
Figure 9: Memory chips
Before seeing the performance results, let’s recap the main features of this video card.
[nextpage title=”Main Specifications”]
The main specifications for the AMD Radeon HD 6990 video card include:
- Graphics chip: AMD Radeon HD 6990 running at 830 MHz (position “2”) or 880 MHz (position “1”)
- Memory: 2 GB GDDR5 memory (256-bit interface) from Hynix (H5GQ2H24MFR-T2C), running at 1.25 GHz (5 GHz, QDR), per GPU
- Bus type: PCI Express x16 2.0
- Video Connectors: One DVI-D and four mini DisplayPort
- Video Capture (VIVO): No
- Cables and adapters that come with this board: One mini DisplayPort to DVI-D (passive), one mini DisplayPort to DVI-D (active), and one mini DisplayPort to HDMI (passive)
- Number of CDs/DVDs that come with this board: NA
- Games included: NA
- Programs included: NA
- More information: https://www.amd.com
- MSRP in the US: USD 700
[nextpage title=”How We Tested”]
During our benchmarking sessions, we used the configuration listed below. Between our benchmarking sessions the only variable was the video card being tested.
Hardware Configuration
- CPU: Core i7 Extreme 965 (3.2 GHz, 8 MB L2 memory cache)
- Motherboard: MSI Big Bang XPower (1.3 BIOS)
- Memories: 3x 2 GB G.Skill F3-10666CL7T-6GBPK (DDR3-1333/PC3-10666, CL7-7-7-18)
- Hard disk drive: Western Digital VelociRaptor WD3000GLFS (300 GB, SATA-300, 10,000 rpm, 16 MB cache)
- Video monitor: Samsung SyncMaster 305T (30” LCD, 2560×1600)
- Power Supply: SilverStone Element ST75EF
- CPU Cooler: Intel stock
- Optical Drive: LG GSA-H54N
Software Configuration
- Windows 7 Ultimate 64-bit
- Video resolution: 2560×1600 @ 60 Hz
Driver Versions
- AMD video driver version: 8.84.3 Beta (Radeon HD 6990)
- AMD video driver version: Catalyst 10.10d
- NVIDIA video driver version: 266.58
- Intel Inf driver version: 9.1.2.1008
Software Used
- 3DMark 11 Professional 1.0.0
- Aliens vs. Predator + Benchmark Tool
- Call of Duty 4 – Patch 1.7
- Crysis Warhead – Patch 1.1 + HOC Bench Crysis Warhead Benchmark Tool 1.1.1
- Far Cry 2 – Patch 1.03
- Lost Planet 2
Error Margin
We adopted a 3% error margin. Thus, differences below 3% cannot be considered relevant. In other words, products with a performance difference below 3% should be considered as having similar performance.
[nextpage title=”Call of Duty 4″]
Call of Duty 4 is a DirectX 9 game implementing high-dynamic range (HDR) and its own physics engine, which is used to calculate how objects interact. For example, if you shoot, exactly what will happen to the object when the bullet hits it? Will it break? Will it move? Will the bullet bounce back? It gives a more realistic experience to the user.
To get accurate results, we had to disable the 80 FPS limit in the game. To do this, input the command, “/seta com_maxfps 1000” (minus the quotes) into the console (` key). It can be set to any number greater than 200.
We ran this program at three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, maxing out all image quality controls (i.e., everything was set to the maximum values in the Graphics and Texture menus). We used the internal game benchmarking feature, running a demo provided by NVIDIA called “wetwork.” We are putting this demo here for downloading if you want to run your own benchmarks. We ran the demo five times, and the results below are the average number of frames per second (FPS) achieved by each video card.
| Call of Duty 4 – Maximum | 1680×1050 | Difference |
| GeForce G TX 570 x2 – SLI |
210.4 | 19% |
| Radeon HD 6990 (1) | 177.4 | |
| Radeon HD 6870 x2 – CrossFireX | 176.4 | 1% |
| Radeon HD 6990 (2) | 172.4 | 3% |
| GeForce GTX 580 | 172.3 | 3% |
| Radeon HD 5970 | 168.2 | 5% |
| Call of Duty 4 – Maximum | 1920×1200 | Difference |
| GeForce GTX 570 x2 – SLI | 203.4 | 19% |
| Radeon HD 6990 (1) | 171.4 | |
| Radeon HD 6870 x2 – CrossFireX | 170.1 | 1% |
| Radeon HD 6990 (2) | 167.4 | 2% |
| Radeon HD 5970 | 164.4 | 4% |
| GeForce GTX 580 | 158.6 | 8% |
| Call of Duty 4 – Maximum | 2560×1600 | Difference |
| GeForce GTX 570 x2 – SLI | 176.6 | 19% |
| Radeon HD 6990 (1) | 148.4 | |
| Radeon HD 6990 (2) | 144.4 | 3% |
| Radeon HD 6870 x2 – CrossFireX | 139.4 | 6% |
| Radeon HD 5970 | 135.2 | 10% |
| GeForce GTX 580 | 113.7 | 31% |
[nextpage title=”Crysis Warhead”]
Crysis Warhead is a DirectX 10 game based on the same engine as the original Crysis, but optimized (it runs under DirectX 9.0c when installed on Windows XP).
We used the HardwareOC Crysis Warhead Benchmark Tool to collect the data for this test.We ran this program at three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, all at very high image quality (but with no anti-aliasing and no anisotropic filtering) and using the Airfield demo. The results below are the number of frames per second achieved by each video card.
| Crysis Warhead – Very High | 1680×1050 | Difference |
| Radeon HD 6990 (1) | 56 | |
| Radeon HD 6990 (2) | 52 | 8% |
| Radeon HD 6870 x2 – CrossFireX | 51 | 10% |
| GeForce GTX 570 x2 – SLI | 50 | 12% |
| Radeon HD 5970 | 48 | 17% |
| GeForce GTX 580 | 45 | 24% |
| Crysis Warhead – Very High | 1920×1200 | Difference |
| Radeon HD 6990 (1) | 52 | |
| Radeon HD 6990 (2) | 50 | 4% |
| GeForce GTX 570 x2 – SLI | 49 | 6% |
| Radeon HD 6870 x2 – CrossFireX | 48 | 8% |
| Radeon HD 5970 | 44 | 18% |
| GeForce GTX 580 | 39 | 33% |
| Crysis Warhead – Very High | 2560×1600 | Difference |
| GeForce GTX 570 x2 – SLI | 45 | 2% |
| Radeon HD 6990 (1) | 44 | |
| Radeon HD 6990 (2) | 43 | 2% |
| Radeon HD 6870 x2 – CrossFireX | 34 | 29% |
| Radeon HD 5970 | 30 | 47% |
| GeForce GTX 580 | 26 | 69% |
[nextpage title=”Far Cry 2″]
Far Cry 2 is based on an entirely new game engine called Dunia, which is DirectX 10 when played under Windows Vista with a DirectX 10 compatible video card.
We used the benchmarking utility that comes with this game, setting image quality to Ultra High (with x8 anti-aliasing) and running the “Ranch Long” demo three times. The results below are expressed in frames per second and are an arithmetic average of the three results collected.
| FarCry 2 – Ultra | 1680×1050 | Difference |
| Radeon HD 6990 (1) | 149.4 | |
| Radeon HD 6990 (2) | 142.5 | 5% |
| GeForce GTX 570 x2 – SLI | 136.3 | 10% |
| Radeon HD 6870 x2 – CrossFireX | 129.0 | 16% |
| GeForce GTX 580 | 116.7 | 28% |
| Radeon HD 5970 | 109.2 | 37% |
| FarCry 2 – Ultra | 1920×1200 | Difference |
| Radeon HD 6990 (1) | 137.7 | |
| Radeon HD 6990 (2) | 130.4 | 6% |
| GeForce GTX 570 x2 – SLI | 128.7 | 7% |
| Radeon HD 6870 x2 – CrossFireX | 118.2 | 16% |
| GeForce GTX 580 | 101.1 | 36% |
| Radeon HD 5970 | 96.2 | 43% |
| FarCry 2 – Ultra | 2560×1600 | Difference |
| Radeon HD 6990 (1) | 106.2 | |
| Radeon HD 6990 (2) | 103.3 | 3% |
| GeForce GTX 570 x2 – SLI | 103.2 | 3% |
| Radeon HD 6870 x2 – CrossFireX | 80.5 | 32% |
| GeForce GTX 580 | 69.0 | 54% |
| Radeon HD 5970 | 64.9 | 64% |
[nextpage title=”Aliens vs. Predator”]
Aliens vs. Predator is a DirectX 11 game that makes full use of tessellation and advanced shadow rendering. We used the Aliens vs. Predator Benchmark Tool developed by Rebellion. This program reads its configuration from a text file (our configuration files can be found here). We ran this program at 1680×1050, 1920×1200, and 2560×1600 resolutions, with max texture settings, 32x anisotropic filtering and 8x anti-aliasing.
| Aliens vs. Predator | 1680×1050 | Difference |
| Radeon HD 6990 (1) | 96.0 | |
| Radeon HD 6990 (2) | 92.0 | 4% |
| GeForce GTX 570 x2 – SLI | 83.4 | 15% |
| Radeon HD 6870 x2 – CrossFireX | 62.8 | 53% |
| Radeon HD 5970 | 61.3 | 57% |
| GeForce GTX 580 | 52.0 | 85% |
| Aliens vs. Predator | 1920×1200 | Difference |
| Radeon HD 6990 (1) | 79.0 | |
| Radeon HD 6990 (2) | 76.0 | 4% |
| GeForce GTX 570 x2 – SLI | 67.7 | 17% |
| Radeon HD 6870 x2 – CrossFireX | 51.4 | 54% |
| Radeon HD 5970 | 50.4 | 57% |
| GeForce GTX 580 | 42.6 | 85% |
| Aliens vs. Predator | 2560×1600 | Difference |
| Radeon HD 6990 (1) | 50.1 | |
| Radeon HD 6990 (2) | 48.0 | 4% |
| GeForce GTX 570 x2 – SLI | 43.3 | 16% |
| Radeon HD 6870 x2 – CrossFireX | 31.6 | 59% |
| Radeon HD 5970 | 30.8 | 63% |
| GeForce GTX 580 | 27.1 | 85% |
[nextpage title=”Lost Planet 2″]
Lost Planet 2 is a game that uses a lot of DirectX 11 features, like tessellation (to round out the edges of polygonal models), displacement maps (added to the tessellated mesh to add fine grain details), DirectCompute soft body simulation (to introduce more realism in the “boss” monsters), and DirectCompute wave simulation (to introduce more realism in the physics calculations in water surfaces; when you move or when gunshots and explosions hit the water, it moves accordingly). We reviewed the video cards using Lost Planet 2 internal benchmarking features, choosing the “Benchmark A” (we know that “Benchmark B” is the one recommended for reviewing video cards, however, at least with us, results were inconsistent). We set graphics at “high,” anti-aliasing at “MSAA8x” and DX11 at “full.” The results below are the number of frames per second generated by each video card.
| Lost Planet 2 | 1680×1050 | Difference |
| Radeon HD 6990 (1) | 72.4 | |
| Radeon HD 6990 (2) | 70.4 | 3% |
| GeForce GTX 570 x2 – SLI | 69.4 | 4% |
| GeForce GTX 580 | 50.9 | 42% |
| Radeon HD 6870 x2 – CrossFireX | 49.5 | 46% |
| Radeon HD 5970 | 46.7 | 55% |
| Lost Planet 2 | 1920×1200 | Difference |
| Radeon HD 6990 (1) | 66.4 | |
| Radeon HD 6990 (2) | 64.2 | 3% |
| GeForce GTX 570 x2 – SLI | 62.6 | 6% |
| GeForce GTX 580 | 44.7 | 49% |
| Radeon HD 6870 x2 – CrossFireX | 41.2 | 61% |
| Radeon HD 5970 | 40.2 | 65% |
| Lost Planet 2 | 2560×1600 | Difference |
| Radeon HD 6990 (1) | 56.3 | |
| Radeon HD 6990 (2) | 55.4 | 2% |
| GeForce GTX 570 x2 – SLI | 53.1 | 6% |
| GeForce GTX 580 | 31.7 | 78% |
| Radeon HD 6870 x2 – CrossFireX | 31.5 | 79% |
| Radeon HD 5970 | 30.1 | 87% |
[nextpage title=”3DMark 11 Professional”]
3DMark 11 measures Shader 5.0 (i.e., DirectX 11) performance. We ran this program at three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, selecting the four graphics tests available and deselecting the other tests available. We used two image quality settings for each resolution, “Performance” and “Extreme,” both at their default settings. The results being compared are the “GPU Score” achieved by each video
card.
| 3DMark 11 – Performance | 2650×1600 | Difference |
| Radeon HD 6990 (1) | 3087 | |
| Radeon HD 6990 (2) | 2990 | 3% |
| GeForce GTX 570 x2 – SLI | 2847 | 8% |
| Radeon HD 6870 x2 – CrossFireX | 1844 | 67% |
| GeForce GTX 580 | 1675 | 84% |
| 3DMark 11 – Performance | 1920×1200 | Difference |
| Radeon HD 6990 (1) | 5128 | |
| Radeon HD 6990 (2) | 5054 | 1% |
| GeForce GTX 570 x2 – SLI | 4971 | 3% |
| Radeon HD 6870 x2 – CrossFireX | 3101 | 65% |
| GeForce GTX 580 | 2901 | 77% |
| 3DMark 11 – Performance | 1650×1050 | Difference |
| Radeon HD 6990 (1) | 6695 | |
| GeForce GTX 570 x2 – SLI | 6617 | 1% |
| Radeon HD 6990 (2) | 6585 | 2% |
| Radeon HD 6870 x2 – CrossFireX | 4144 | 62% |
| GeForce GTX 580 | 3831 | 75% |
| 3DMark 11 – Extreme | 2560×1600 | Difference |
| Radeon HD 6990 (1) | 1965 | |
| Radeon HD 6990 (2) | 1884 | 4% |
| GeForce GTX 570 x2 – SLI | 1825 | 8% |
| Radeon HD 6870 x2 – CrossFireX | 1241 | 58% |
| GeForce GTX 580 | 1075 | 83% |
| 3DMark 11 – Extreme | 1920×1200 | Difference |
| Radeon HD 6990 (1) | 3141 | |
| GeForce GTX 570 x2 – SLI | 3047 | 3% |
| Radeon HD 6990 (2) | 3037 | 3% |
| Radeon HD 6870 x2 – CrossFireX | 2014 | 56% |
| GeForce GTX 580 | 1780 | 76% |
| 3DMark 11 – Extreme | 1650×1080 | Difference |
| Radeon HD 6990 (1) | 3911 | |
| GeForce GTX 570 x2 – SLI | 3909 | 0% |
| Radeon HD 6990 (2) | 3815 | 3% |
| Radeon HD 6870 x2 – CrossFireX | 2501 | 56% |
| GeForce GTX 580 | 2265 | 73% |
[nextpage title=”Conclusions”]
In our tests, the performance gain by having the switch set at “1” (overclocked BIOS) instead of “2” (default BIOS) was up to 8%. We don’t know why users buying this video card would leave video card at its default BIOS since you can have a bit higher performance by simply flipping a switch. Because of that, the comparisons below will be based on the results achieved with the switch positioned at “1.”
Compared to the fastest card from NVIDIA, the GeForce GTX 580, the Radeon HD 6990 was up to 85% faster, depending on the game and resolution: up to 31% faster on Call of Duty 4, between 24% and 69% faster on Crysis Warhead, between 28% and 54% faster on FarCry 2, 85% faster on Aliens vs. Predator, between 42% and 78% faster on Lost Planet 2, and between 73% and 84% faster on 3DMark 11.
The GeForce GTX 580, however, does not compete with the Radeon HD 6990. A competing system would be two GeForce GTX 570 video cards installed in SLI. Under this scenario, the Radeon HD 6990 was up to 17% faster, with the exception of Call of Duty 4, where the GeForce GTX 570 setup was 19% faster than the new Radeon HD 6990. The Radeon HD 6990 was between 6% and 10% faster on Crysis Warhead, between 3% and 10% faster on FarCry 2, between 15% and 17% faster on Aliens vs. Predator, between 4% and 6% faster on Lost Planet 2, and up to 8% faster on 3DMark 11.
The Radeon HD 6990 is truly the fastest video card ever built, targeted to the ultra (and rich) enthusiast who wants to have the fastest gaming PC in town. Add two of them in CrossFireX and the sky is the limit.