ASUS HD4850 - first tests of the latest GPU AMD RV770
Specifications, efficiency of the cooling system
The GPU RV770 offers a lot of improvements as compared to the predecessors R600 and RV670. This is what AMD itself states:
| |
ATI Radeon™ HD 4850 |
ATI Radeon™ HD 3870 |
| Q-ty of transistors, mln |
965 |
666 |
| Process technology, nm |
55 |
55 |
| Streaming processors |
800 |
320 |
| Texture units |
40 |
16 |
| Rastering units |
16 |
16 |
| GPU clock speed, MHz |
625 |
775 |
| Type and effective frequency of the video memory |
GDDR3, 2000 MHz |
GDDR4, 2250 MHz |
| Computational power of the GPU, TFlops |
1,0 |
0,497 |
| Bus type |
PCI Express 2.0, x16 |
PCI Express 2.0, x16 |
| Support for DirectX |
10.1 |
10.1 |
| Tessellation unit |
yes |
yes |
| Integrated video decoder |
ver. 2.0 |
ver. 1.0 |
| Support for ATI PowerPlayTM |
yes |
yes |
The number of streaming processors has increased from 320 to 800, and the number of texture units has gone up from 16 to 40. It's just the small number of texture units was the bottleneck for R600/RV670, and now this flaw has been fixed. The number of mathematical operations (Multiply-Add) on the GPU has gone up by more than 2 times, which became possible due to the increased number of streaming processors. The integrated video decoder (UVD) has also been improved and acquired improved support for decoding the HD content and the feature for dynamic contrast adjustment. We will review these improvements in more detail in our forthcoming materials, but now let's get round to practical tests.
First, let's check the efficiency of the cooling system. The Riva Tuner utility does not yet support the new family of AMD video cards, so we'd better make use of the Catalyst Control Center panel. As before, we'll be testing the efficiency of the cooling system using the Firefly Forest test from the 3DMark 06 suite. The test conditions: the resolution 1600x1200, 4-X FSAA, and 16-X AF. After nine runs of the test, the GPU heating sensor indicated 86 C, whereas at rest the temperature was merely 6 degrees lower. This is very easy to explain – when the video card runs in the 2D mode, its fan rotates at very low speeds. Once it switches to the 3D mode, the fan of the cooling system gains speed and does not let the temperature of the VPU rise sharply. The clock speeds of the video card are identical to the recommended and are 625 MHz for the GPU and 2000 "true" MHz for the video memory. Below you can see a screenshot taken from the GPU-Z utility:
Benchmarking and conclusions
| Test configuration |
| CPU |
Intel Core 2 Quad X6850 3.0 GHz |
| Motherboard |
ASUS P5E3 (Intel X38) |
| RAM |
Corsair Dominator DDR3 2 x 2 GB 1333 MHz 6-6-6-18 1T |
| Operating system |
Windows XP+SP2+DirectX 9.0c / Windows Vista for DX10 tests |
| Power supply unit |
Thermaltake Toughpower 750 W |
To test ASUS EAH4850 and Radeon HD3870, we used Catalyst 8.6, for Leadtek GTX 280 - ForceWare 177.34, for GeForce 9800GTX - ForceWare 175.19. OK, off we go. Traditionally, we start our introduction to the results with tests of the 3DMark suite.
At 3DMark’05, ASUS EAH4850 contrived to lose to the predecessor Radeon HD3870. The cause of that was the low usage of the GPU by the 3DMark test, so the chances of different GPUs equaled, but because of the higher frequency of the latter the weak one beat the strong one. On the other hand, 3DMark’06 put all the dots on i's, and ASUS EAH4850 took its honorable 3rd place thus slightly yielding to GeForce 9800GTX. But in games the situation appears to be different as we can see. We'll be running all the further tests with 4X FSAA and 16X AF.
At Call of Duty 4, ASUS EAH4850 yields a bit to GeForce 9800GTX. But in practice such a minor difference in FPS will be hardly noticeable.
At Crysis, ASUS EAH4850 starts taking a lead with the rise in resolutions. Interestingly, we'll see such a pattern several times.
The cards demonstrate similar results in Windows Vista as well. That is, in Windows Vista the new product works equally well as it does in Windows XP.
At Call of Juarez, the new product takes a sure lead over GeForce 9800GTX, albeit the leadership is not so pronounced. That is already enough to state that the new Radeon HD4850 is no worse than GeForce 9800GTX. Although formally the new product is aimed at competition versus merely GeForce 8800GTS 512 and GeForce 8800GT.
Once we switched to Windows Vista, we see that the new product loses at low resolutions, but takes a revenge in higher resolutions. That is really nice to see because the future is just with demanding modes, however hackneyed that may sound.
At Need for Speed Pro Street Racing, the new product is on par with GeForce 9800GTX.
But at Need for Speed Carbon, ASUS EAH4850 leaves GeForce 9800GTX well behind.
At Prey, ASUS EAH4850 beat GeForce 9800GTX in two modes of three, and just at high resolutions. Perhaps the developers did spend more efforts to the demanding graphic modes.
Final Words
AMD has done a serious error-correction work, which resulted in the new GPU RV770 with the Radeon HD4850 video card created on its base. Certainly, the strict performance diagrams do not reflect all the beauty of the image shown by Radeon HD4850. We'll come back to the investigation of the matter in our forthcoming materials. Moreover, in the nearest future, we'll be testing both the new products - GeForce GTX 280 and Radeon HD4850, at very high resolutions and hard graphic modes.
Delightful is the fact that the recommended price for the new product is merely $199. Needless to say, NVIDIA is holding a leadership in terms of the absolute performance level for single-processor video cards. But it is very well known that even the most fierce competition for the buyers' preference (and wallets) is going on just in the mid-end pricing sector at which Radeon HD4850 is aimed. So it's up to the buyer to choose what is really sensible.
 |
Top Stories: |
 |
 |
|
MoBo:
| |
|
|
VGA Card:
|
|
|
|
CPU & Memory:
|
|
