ABIT RX600XT-PCIE

Introduction

That was an early live-demo engineering sample of what is now known as the Alderwood chipset. That was most likely the very first working system in Taiwan that time, and the "risky show" was not aimed at visitors, but at the competitors.

It is quite understandable for ASUS, of course. Overcoming all the compatibility issues in the early engineering samples of the platform made of entirely new components and making it work all right is like launching a spacecraft. Even nowadays, 10 months past that remarkable event, after all the announcements of the project participants, while assembling already the third LGA775 system, the only concern is "will it really start up?.." All the three did start up.

The project for the new standard of the PCI-E graphics bus started a couple of years ago, and then a year ago the PCI-E versions of semi-operative cards existed physically at labs. In February 2004, ATI and NVIDIA already had their finished solutions and were awaiting the launch from Intel. ATI was the first to get nerves, and on the third day of Computex`2003, on 3rd June, Dave Orton, ATI's vice-president presented the whole PCI-E line of ATI to the public - from X600/X300 and Radeon X600 (used in notebooks) up to the complete line of FireGL chips (V7100, V5100, etc...) with the native PCI-E interface.

No one cared about the lack of space to fit the cards - all were deadly tired of waiting. In any case, the press, the manufacturers and sales community were aware that in two weeks an official launch of CPU LGA775 and i915P/i925X chipsets would follow. The announcement took a load off the minds of all, and then .. it rolled along..

The very same day, ATI's partners presented their own versions of X600/X300. All in one day :).

ASUS AX600XT *758x590	Palit X800 *900x675

Sapphire R600XT/Pro *900x675	Sapphire R300 *900x602

Gigabyte R800XT/R600Pro *900x675	HIS X800XT/R600XT *900x675

Powercolor X800XT *900x675	Powercolor X600XT *900x675

The PCI-E Family for July 2004

To date, not so many PCI-E solutions have been announced. NVIDIA presented its own series of cards on the base of the latest GeForce 6800 (it's still unknown how many cards on the base of the chip the PCI-E product line will make up) and three cards of the old arsenal - the "old hi-end" is presented by GeForce PCX 5900, the mid-end will be filled by GeForce PCX5750 (on the FX5700 chip plus the PCI-E bridge), and the low-end sector is presented by PCX5300 (on the base of FX5200 chip plus the PCI-E bridge).

For ATI, the situation is simpler - for the high-end two X800XT/Pro cards are to be presented, with two more X600XT/Pro for the mid-end, and X300SE for the low-end. All the chips by ATI for cards of the PCI-E standard offer integrated support for the PCI-E. On the one hand, there is no need to complicate the card through installing a bridge (the chip is inexpensive, but there are too many pins, which results in cumbersome and problematic wiring, higher defect ratio during assembly and formidable sizes of the card themselves). On the other hand, this is just another order to TSMC for as many as three more chips with the demand still uncertain.

Both approaches have a right to exist and poorly correlate with the "performance" concept. That's because ATI cards seem to be easier and cheaper to produce, whereas for NVIDIA cards there is no need to develop and order new chips. All in all, this is about "pure marketing" that has nothing to do with us. For now, there aren't games which will require double AGP 16x bandwidth! But they will appear eventually.

The following table shows a performance correlation between all the new video cards:

PCI-Express x16 Roadmap `2004
Hi-End	GeForce 6 Series	Radeon X800XT/Pro
Old Hi-End	GeForce PCX5900	-
Middle	GeForce PCX5750	Radeon X600XT/Pro
Low-End	GeForce PCX5300	Radeon X300SE

I can say straight off, the table does not reflect the reality a bit. At the end of today's tests, we'll publish its corrected version.

ABIT RX600XT-PCIE was the first video card that arrived at our test lab in the PCI-E make, so we'll start reviewing all what is to come to the retail shelves soon.

ABIT RX600XT-PCIE Video Card Features

1. Video card: ABIT RX600XT;
2. Adapters: 15-pin D-sub/ DVI and S-Video/RCA;
3. Cables: S-Video and RCA-Composite
4. Software CD: Catalyst PCI-E, Acrobat Reader, DirectX 9.0b, PowerDVD 5
5. Highly detailed manual on the produce of ABIT made on RADEON RX Series PCI -E chips (in .pdf) plus two printed manuals of similar contents.

Now we are moving on to the most exciting part, - description of the board's features.

Design and layout

The video card is made by ABIT on the orange PCB, traditional for boards built on ATI chips. The power supply scheme and the PCB are in many ways similar to those for RADEON 9600XT.

*1200x805; 251kb

*1200x883; 290kb

The board offers 128 Mb DDR memory with the 128-bit data transfer bus all made as eight Hynix memory modules marked HY5DU283222 AF-25 made in the BGA form factor, and 2.5 ns access time, as you can see from the marking.

Half the memory chips are positioned over the front side of the card, with the other half on the reverse, which makes it resemble most 9600XT boards built on the RV360 chip where the positioning of memory chips is similar to what we see in this case.

Now a few words on the cooling system. On this board, it is made as a radiator whose base and heat-spreading fins are common for both the GPU and the four memory chips located near the GPU on the front side of the board, with an aluminum cap of the cooler.

In fact, we see that the radiators on the memory are more likely decorative items since the cover only the memory chips located on the front side of the board. The memory chips located on the reverse side are not equipped with whatever heat-spreaders.

That is an essential shortcoming, since the installation of additional radiators on the video memory has never been an issue.

The board offers a standard set of outputs: analogous, digital, and TV-Out.

The rated clock speed of the ATI RV380 core is 500 MHz. At first glance, it was difficult to get to it. The cooler is firmly fitted on the thermal paste applied over all the four front memory chips and the core itself. That is not an obstacle for true enthusiasts:

The general specifications for ABIT RX600XT-PCIE versus ATI Radeon 9600 XT are gathered in the following table:

Now we are getting round to assembling a test configuration:

LGA Test Configuration

To test the PCI-E video cards, we assembled an absolutely canonical system made up of Intel's i925X motherboard, 3.4Mhz Extreme Edition processor, a Matrix RAID mounted with two Matrox MaxlineIII SATA disks. Even the MSI's SATA CD-Rom was taken just in case - for the purity of experiment. The DDR2 533MHz memory made up of two 512MB bars completes this configuration of top-end total price. The system had to be powered from a new-format PSU (24pin + power cable SATA) made by HiPRO - HP-W460GC31 (460W). Not so much of a "budget" system :), but with it you can forget about upgrading the video card test facility for the coming year and a half.

CPU	P4 3.4MHz Extreme Edition 800FSB LGA P4 3.6Mhz 800FSB LGA P4 3.2Mhz 800FSB LGA
Mb	Intel D925XCV (i925X)
Memory	PC2-4300 (533MHz DDR2) 2x256Mb in the dual-channel-mode Memory latency timings - 4:4:4
HDD	Matrox MaxlineIII SATA, 2x250Gb Matrix RAID
CD-ROM	MSI XA52P COMBO Writer SATA
OS	WinXP + SP1 + DirectX 9.0b
Drivers	ForceWare 61.11 Catalyst 4,7

Frankly, we had to take a lot of trouble over the test facility. The thing is, the first PCI-E video card that arrived proved to be broken, and the test configuration wouldn't start up for two days. It was impossible to figure out what was wrong having new components in singletons - the board does not offer external diagnostics. In the end, after two days of unsuccessful experiments there came ABIT RX600XT which started up at last. Two days afterwards, there came two more PCX5900 cards by Gigabyte, and two MSI's cards - RX600XT and PCX5750. With none of those cards there were compatibility and stability issues. Moreover, for two weeks of merciless runs the test stand neither reset nor hung, which was a real surprise. Even now we can safely assert that Intel did a thorough job preparing the transition and worked a lot with the partners developing standards, drivers, overcoming compatibility issues of components.

By the way, the corporate vocabulary of Intel has no word like "partnership", - instead, they say "customer". So, the whole Taiwan for Intel is a "customer" :-)

A few words on the drivers and BIOS. The existing version of WinXP + SP1 is absolutely unaware of where it is put. All the necessary drivers for correct system operation take up as much as two disks, and the installation is not an issue but for one - the lack of time. The number of drivers is too great, nor they are all certified. As rumors have it, WinXP SP2 will contain the whole original set of necessary drivers for the LGA775 platform.

The new BIOS is two big - over 2 Mb. Remember that the BIOS for i865/875 was much smaller in size - about 400 kb. Since new versions are released almost once every week (the platform is continuously polished), the first happy owners of the LGA will have to frequently do the troublesome re-flashing procedure. By September, all this fever should go down, and the system will take its proper form.

For NVIDIA PCX cards, we used ForceWare 61.11, taken from NVIDIA's FTP-resource for developers and the press.

Test software:

Synthetic benchmarks:

1. Sandra 2004 + Service Pack 2 (SP2);
2. PCMark2004 Patch 120;
3. 3DMark2003 Patch 340;
4. 3DMark2001SE Patch 330;
5. Codecreatures v1.0.0 (a DirectX 8.1 application, shaders on, Hardware T&L);

   Gaming benchmarks:
   

6. Unreal Tournament 2003 (Direct3D, Hardware T&L, vertex shaders, Dot3, cube texturing.);
7. Unreal Tournament 2004
8. AquaMark 3 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L, AquaMark3 Triscore mode);
9. HALO: Combat Evolved 1.2 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L, quality set to the maximum possible);
10. Gun Metal Benchmark 2 v1.20s (a DirectX 9.0 benchmark, Vertex Shaders 2.0, Pixel Shaders 1.1, Hardware T&L);
11. X2: The Threat Demo (Direct3D, multitexturing, Dot3, running in the benchmark mode embedded in the demo version);
12. Final Fantasy XI Official Benchmark 2 (a benchmark for assessing the performance in Final Fantasy XI. The developers haven't presented any data on the gaming engine);
13. Tomb Raider: Angel of Darkness v49 (DirectX 9.0, Vertex Shaders 2.0, Pixel Shaders 2.0);
14. Half-life 2 leaked beta (DirectX 9.0, Vertex Shaders 2.0, Pixel Shaders 2.0);
15. FireStarter (DirectX 8.1/DirectX 9.0, pixel and vertex shaders, particle system, dynamic lights, projected textures);
16. FarCry ver1.1 (DirectX 9.0, Pixel Shaders 2.0, our own demo "3Dnews001" was used).

System Tests: CPU

To get some general idea of the new system, we'll first demonstrate the results of testing the new line of CPU built on the Prescott core made in the LGA775 form factor. Three processors took part in the tests - P4 3.4XE, P4 3.6 and P4 3.2. To some tests, we added results produced with Socket 478 P4 3.2.

Comanche 4 is one of the most processor dependent applications. The rule is simple - the older the game, the better:

As we can see, the "processor-dependence" for Comanche 4 with X600XT is over at the 1920x1440 resolution, and the graph that's been produced quite vividly demonstrates the performance difference for new processors. The P4 3.4 Extreme Edition LGA775 leaves P4 3.6 LGA775 well behind, which was somehow of a surprise. But that makes it a bit clear why they ask $999 for P4 3.4 Extreme Edition LGA775.

"Far Cry" is the other extreme. At "Low quality", (in the High-mode the difference is minimum, of course) we get the following:

Well, what to say to that? ... The conclusion is evident: a card of ATI X600XT level is weak for this system - the gap is disastrous as the resolution goes up. The CPU is ready to feed up to 150 finished scenes to the graphic rendering pipelines of the video card, but at the "output" we get merely 66 fps at 1600x1200. In the ideal case (if a X800XT or GeForce6800 are used), at low quality settings these lines should turn to ideal parallel straight lines which characterize only the processor performance. No need to take these comments and graphs as a joke :-))). You would rather say "Who on earth ever plays on such a system in the low-quality mode?". But on that page we primarily estimate the CPU first...

For completeness and order, we'll show a graph for "Far Cry" in the high-quality mode with three new processors:

Another interesting option for demonstrating the difference in the processor clock speeds - the "UT2004":

The picture is absolutely standard and expected. Moreover, if we set the quality to the maximum, the difference remains unchanged:

That again proves the fact that the engine of the game is outdated, but we have been charged extra $60 only for the gameplay in a new package. The engine is old - it offers no version 2.0 shaders, nor anything of the arsenal of most recent graphic technologies. The game still remained at the DX8.1 level.

We tested the previous UT2003 in the Hi-Quality without any doubts of seeing a difference in any case (it is anyway more correct to reset the image quality settings when testing CPU). But nevertheless:

Isn't that a staggering graph? First, we added results produced on the old P4 3.2 Socket478 system with a Radeon 9600XT video card which competes with the three new processors + ATI X600XT. The 9600XT and X600XT is the same core, except the two things - the I/O block of the AGP standard has been revamped to the PCI-E (the native support itself). Secondly, the memory latency timings have been raised (600Mhz -> 760Mhz).

What do we see now? At 640x480 where the video card performance is of no importance, and the CPU clock speed is the same, - both readings are equal to 159 fps. But the difference in video cards performance is seen on the next step of the resolution ladder. The theory fitted ideally to the values taken. Basically, any reviewer would draw such graphs from scratch for any configuration, without any specific values, but the lines will be positioned in an absolutely correct manner. That is, when you start testing, you have an absolutely clear idea of what should be bent where and above what positioned... This allows instantly "seeing" erroneous results and irrelevant figures. It's always pleasant to see such canonical figures shape up in the end :)

And, a few more of "pure" processor load tests:

System Tests: DDR2 Memory

SiSoftware made us all happy - on 29 June, there was released the long-awaited Sandra 2004 Service Pack 2 (SP2) with included support for the DDR2 and Intel 91X, 925X chipsets (as well as for Athlon 64 939). Authors, hardware-related web-sites and test labs can freely make the commercial version of the program available - for details, read the press release.

In the configuration with two channels of DDR2 533 memory (512 MB for each channel), we got the following:

At the same time, we verified the theoretical difference in the DDR2 memory bandwidth when switching the system from the dual-channel mode of memory operation to the single-channel mode. By the way, after reboot the system displayed a message saying it was "not the most successful choice". The creators of BIOS had a riotous time :-)

Dual-channel scheme:	Single-channel scheme:

As we see, the drop amounts to 25% when switching to the single-channel mode. Now let's see what it is like in games:

With gaming applications, the difference is minimum. We tested a dozen more games - no difference in results were found with them. That is, if you have only a single DDR2 module and are not into video encoding, this is not a reason for worry. All in all, in modern games it is the speed of the video card which is decisive.

Testing ABIT RX600XT-PCIE

The system has been assembled, debugged, all the games, patches, benchmarks and tools have been installed. We took the readings of system tests, and from the results it is clear that all is OK - the test configuration demonstrates the expected performance, shows no conflicts, and all the system devices run in their rated modes. That is, the software and hardware proved to be an ideal fit. We can now start testing the video cards. What to compare, actually? Where is the the point of departure which will allow comparing AGP and PCI-E video cards on two absolutely different systems? It is like comparing Apple and PC, which some used to like in be past to make the public laugh, but they no longer do so.

All seems to be very simple - you get a Radeon X600XT-PCIE video card (on the RV380 core), acquired by the courtesy of ABIT, there is a Radeon 9600XT AGP (RV360), with the only difference in them is in some differing parts of the GUI. But all is not as simple as it seems.

First, for X600XT video cards ATI has changed the memory latency timings through raising them essentially as compared to Radeon 9600XT (730 versus 600Mhz), and ABIT raised them even more - up to 760Mhz. That is, the X600XT will be in any case faster than its AGP counterpart.

Secondly, the system itself built on the i915P/i925X chipsets is very different from the previous i865P/i875. If we look at the factors of the new system memory application, use of the PCI-Express X16 graphic bus, changed processor socket, and Matrix RAID separately, we can hardly notice any difference, but merging all these technologies we get not merely a "noise level", but an entirely new system which should be regarded as a whole, a set of new functionalities.

That is, it is very difficult to perform a "direct and pure" comparison of AGP and PCI-E video cards made on kindred cores on different platforms, so in this transition period we have to compare systems as "old versus new". And the results of comparison of PCI-E versions of cards produced on the same system can be regarded absolutely relevant already.

The situation was resolved through using other PCI-E video cards from the most popular sector, middle-end, that arrived during tests, which will help us position the first mid-end PCI-E cards made by NVIDIA and ATI. Nevertheless, results for Radeon 9600XT AGP produced on the i865P system were included into most tests.

First, the synthetics:

The very first test did not bring any surprises. All is predictable enough. We can very well see the anticipated leap of ABIT RX600XT from the seemingly similar Radeon 9600XT, but the increased memory frequencies plus the more powerful system played their part.

All the synthetic results are well predictable and ideally fit the theoretical forecasts. In real games, all is different...

Gaming benchmarks

With the first engine of a real game, GeForce PCX5900 shows results lower than ABIT RX600XT-PCIE, which are essentially lower.

History repeats in Aquamark 3 as well, albeit not in that fatal manner.

For UT2003, we traditionally give results for the three integrated demos - normally, results are "astray", but this time the "picture" was evident - GeForce PCX5900 takes a small lead over ABIT X600XT, with the old AGP-version Radeon 9600XT going closely behind, and PCX5750 closing the list.

In UT2004, we used our own demo "Demo-3Dnews003.demo4" (put it into the "demos" directory) where a serious action was written, worth showing ;-)

*1200x885	*1200x885

Again a similar picture - today's mid-end card by ATI is well on par with the yesterday's hi-end. It's high time to change the grid of reference for partitioning the markets :-).

At GunMetal (Pixel Shaders 1.1), NVIDIA cards demonstrate quite a good showing.

Like GunMetal, the faithful HALO defended the desecrated honor of PCX5900 a bit.

"Half-Life 2" is not optimized stolen demo. Its results are no more than optional.

The "Tomb Rider: Although "Angel of Darkness" is part of the honorable list of games financed by NVIDIA, the game proved to be a nightmare for the company.

FireStarter also degrades the recent Hi-End (PCX59000) to the level of common Middle of today.

Both in the "quality" mode and without it, but at the "worship" game FarCry the situation is the same - X600XT wins in the middle-end sector for PCI-E solutions.

Page 6 - Final Words

	Contents:
	Findings on the system In the testing practice, there are two absolutely different concepts - "engineering sample" and "end product". In the former case, these are hardware singletons assembled in a semi-handicraft manner in subordinate production facilities at company labs. If they happen to appear on our test facilities, the purpose is the only one - to make oneself aware of the consumer properties of future products. Their task is to start and work a bit :-). Requirements to the end products manufactured in batches and for which the consumer is to pay quite an amount are different. While assembling the fist test facilities on the LGA platform, you can't help thinking that you deal with a big "engineering sample" whose only purpose is at least to start up. Only after two weeks of successful practice in re-installing processors, memory kits from various manufacturers, video cards etc.. you arrive at the understanding that you have a complete and well-polished solution. What is most strange is that a system using no WHQL driver didn't hang up at lest once and behaved always correctly. All the implemented novelties allow to regard the i915P/i925X platform as a most serious change of the PC since the times of the first Pentium processors. But - has it all added much to the performance? On the primitive level - yes, it has, and more likely due to the use of the RAID technology which improved the "system response" to a serious extent. But it will take a while when we feel the role of the PCI-Express bus with the doubled bandwidth (up to 4 GB/s) to the full. Its purpose, like the task of many introduced technologies is to solve the system bottlenecks. Findings on the ABIT RX600XT-PCIE Video Card Our tests of the first mid-end PCI-E video cards brought the first surprise - ABIT RX600XT (235$) on the average offers greater or equal performance as compared to GeForce PCX5900 (255$). This is confirmed by the posted price for these products (see pricewatch.com). Here is the amended table: PCI-Express x16 Roadmap `2004 Hi-End GeForce 6 Series Radeon X800XT/Pro Hi-Middle GeForce PCX5900 Radeon X600XT Middle GeForce PCX5750 Radeon X600Pro Low-End GeForce PCX5300 Radeon X300SE Of the PCI-E video cards which are already available on sales, of most interest is the "GeForce PCX5900 versus Radeon X600XT" rivalry, because no Hi-End cards with the PCI-E bus will be available on sales until September. Having decided to assemble a computer on a new platform, you would have to choose just between the two. ABIT RX600XT to date is the fastest of all available PCI-E video cards. In our next test session, we'll compare the canonical GeForce PCX5750 and Radeon X600Pro, and reveal "the best of the best" among the participants in the Ultra/GT/NonUltra group and the XP/Pro group. The low-end is unlikely to be of interest to anyone, at least at the first stage of launching the i915P/i925X - these are quite strange and cheap cards for system of $1500 price range. Read more on this topic: Intel Pentium4 LGA775 Processors Abit AA8 : i925X Alderwood Asus P5AD2 Premium : i925X Alderwood Gigabyte 8GPNXP-Duo : i915P Grantsdale

Copyright © 2005 Digital-Daily. All Rights Reserved. contact - info@digital-daily.com