Digit-Life Hardware News
17.05.2008
[01:09] Razer Launches Goliathus Speed and Control Edition Soft Mats
[00:53] Plextor Shrinks New MediaX PC-to-TV Media Player
[00:42] Daily Mailbox
[00:30] Transcend Unveils New V90P USB Flash Drive
[00:23] Microsoft Windows Now Available on One Laptop per Child XO Laptops
[00:18] IBM Research Unveils Breakthrough In Solar Farm Technology
[00:10] Creative Unleashes Sound Blaster X-Fi Titanium Fatal1ty Professional and Champion Cards
15.05.2008
[16:16] Sharp Achieves the Highest Power Density for Direct Methanol Fuel Cells (DMFC)
[16:10] Elpida Offers First DDR2 SDRAM With x32-bit I/O Configuration
14.05.2008
[21:13] Daily Mailbox
Your link here

Home Home
Latest News | Platform | Coolers | HDD/DVD | Video | Sound | Network | Imaging | Mobile
Monthly | Rightmark Tools | Search | Forum | Mailing | Links | Advertise | About Us
Digit-Life Articles Feed    Digit-Life News Feed

Latest Articles:

Gigabyte MA78GM-S2H Motherboard on AMD 780G Chipset (Socket AM2+)

NVIDIA GeForce 8800 GT Performance Scaling in Modern Games

Sapphire PURE CrossFireX 790FX Motherboard on AMD 790FX Chipset (Socket AM2+)

Passively Cooled Gigabyte GeForce 9600 GT 512MB

i3DSpeed, April 2008






ATI RADEON 3850/3870 (RV670)

320 Shader Processors and 256-bit Memory Bus

Direct3D 10: Vertex texture fetch rate

Vertex Texture Fetch tests measure the speed of many vertex texture fetches. These tests are similar, and the correlation of their results in Earth and Waves tests must be also similar. Both tests use displacement mapping based on texture lookups. The only major difference is that the Waves test uses conditional branches, while the Earth test does not.

Let's analyze the first test (Earth) in Effect detail Low mode:

This time the three modes demonstrate different performance of graphics cards. The simpler the mode, the higher the effect of memory bandwidth on performance. It corroborates to the conclusion in the article devoted to the GeForce 8800 GT, where we discovered a significant performance difference between the GeForce 8800 GT and the 8800 GTX.

You can see it well in the comparison of the HD 2900 XT and the HD 3850. The old solution easily wins in the Low mode, both cards perform on a par in the Med mode, and the HD 3850 shoots forward in the High mode. The same applies to the HD 3870. But it outperforms the HD 2900 XT even in the Med mode. The HD 3870 competes with the GeForce 8800 GT with varied success. The NVIDIA card wins in two modes, and the new card from AMD comes a little forward in the most complex mode.

The ratio between performance of the new Mid-End cards from AMD and the old RADEON HD 2600 XT card has changed, the latter is outperformed threefold in the most complex mode. Let's have a look at results of this test with more texture lookups:

The situation hasn't changed much. The difference between the HD 3800 and the HD 2900 has become smaller, the effect of memory bandwidth remains, although the more expensive card from the new family is actually faster than the old card. The previous Mid-End card (RADEON HD 2600 XT) is still much slower, and the GeForce 8800 GT now wins in all tests.

Let's have a look at results of the second vertex texture fetch test. The Waves test executes fewer texture lookups, but it uses conditional branches. The number of bilinear texture lookups in this case reaches 14 (Effect detail Low) or 24 (Effect detail High) per each vertex. Geometry complexity changes just like in the previous test.

The Waves test shows similar results to those in the previous test, we can see the effect of memory bandwidth even in NVIDIA graphics cards. Both new cards of the HD 3800 family look good, the cheaper card performs on a par with the HD 2900 XT. The more expensive card is faster. Both cards are two-three times as fast as the HD 2600 XT, but they still cannot outperform the GeForce 8800 GT that wins in complex modes. It's defeated in the Low mode because of the lower memory bandwidth. Let's analyze the second mode:

As test complexity grows, both HD 3800 cards outperform the HD 2900 XT. There must have been some optimizations made. All the other conclusions hold good - performance in the Low mode is limited by memory bandwidth, TMU and ROP units play more important roles in High modes. The GeForce 8800 GT copes with vertex texture fetch tests better than the new cards from AMD. When geometry complexity is low, performance is limited by lower memory bandwidth. As the amount of fetched data grows, this card becomes the fastest product.

Conclusions on the synthetic tests

  • The RV670 is based on the R6xx architecture. To all appearances, the new GPU hasn't changed much in its 3D part (the R600 included almost all features required by DirectX 10.1). Like a forefather, it's notable for high computing performance, especially in modern and future applications with lots of complex shaders. The RV670 offers high efficiency of unified processors and higher clock rates (compared to the R600), which help this GPU demonstrate very good results in most synthetic tests. The RADEON HD 3850 and the HD 3870 are generally no slower than the more expensive HD 2900 XT in our tests, and they are even faster sometimes!


  • The weakest spot of the RV670 hasn't changed since the R600 - the new GPU still has relatively few texture units. So video cards based on this GPU cannot demonstrate higher performance in texturing-intensive tests. Another little problem is a narrower memory bus (compared to the R600) and lower bandwidth, which is insufficient to perform on a par with the RADEON HD 2900 XT in all tests. On the other hand, the HD 3850 and the HD 3870 perform very well even with lower bandwidth, they fare on a par with the top solution. So this memory bandwidth is quite enough for Mid-End solutions.


  • Even though the RADEON HD 3870 is slightly outperformed by the GeForce 8800 GT in most tests, performance in games will play the pivotal role. We'll analyze performance in games and retail prices in the next part of the article. The RV670 has an advantage over the G92 - it's less complex and has a smaller die surface. So its manufacturing costs must be lower, all other things being equal. You should keep tabs on our reviews and retail prices, and make your choice not on the ground of synthetic tests and recommended prices, but on game tests and real retail prices in your town.

Our analysis of synthetic tests of the RADEON HD 3850 and the HD 3870 shows that the new Mid-End cards from AMD offer high performance, similar to older top solutions. Like the NVIDIA GeForce 8800 GT, they compete with much more expensive cards! What concerns energy efficiency, such monsters as the HD 2900 XT cannot compare with them, because the new GPUs are manufactured by the 55 nm fabrication process, which gives important advantages in power consumption and heat release.

We should also mention the potential advantage of the RV670 and its cards - DirectX 10.1 support. Even though this version does not exist yet (except for beta versions for developers, of course), this feature is an important competitive advantage, because many users evaluate graphics cards by their support for this API. We cannot evaluate the 10.1 advantages so far, but we'll pay attention to this issue in future articles.

The next part of the article will contain tests of the new Mid-End cards from AMD in games. These results should prove that our synthetic conclusions are true.


PSU kindly provided by TAGAN

The Dell 3007WFP monitor kindly provided by NVIDIA



Andrey Vorobiev (anvakams@ixbt.com)
Alexei Berillo (sbe@ixbt.com)
December 10, 2007



Latest News | Platform | Coolers | HDD/DVD | Video | Sound | Network | Imaging | Mobile
Monthly | Rightmark Tools | Search | Forum | Mailing | Links | Advertise | About Us

Copyright © by Digit-Life.com, 1997-2008. Produced by iXBT.com
Design by Explosion