Digital-Daily : Video : fx5600ultra-vs-radeon9600pro

FX5600Ultra vs Radeon9600Pro

Author: Andrey Kuzin
Date: 28.08.2003

II. Tests

Test configuration:

CPU	P4 2,4Mhz 800FSB (Northwood D1)
Mb	Epox 4PDA2+ (i865PE)
Memory	PC2700 2x256Mb 400Mhz in the dual-channel mode Timings - 3:8:3:3
Video cards	ABIT Siluro FX 5600Ultra DT 128Mb ABIT Siluro FX 5600Ultra OTES 128Mb Sapphire Radeon9600 Pro 128Mb Ultimate Edition Gainward Ti4200 128Mb
OS	WinXP + SP1
Drivers	Detonator 44.03 Catalyst 3.6

Benchmarking software used:

As compared to the previous tests, we added one more program - "GunMetal Benchmark" (DX9). Below, we'll tell about it in more detail.

To compare the current "middle range", we took the Gainward Ti4200 128Mb; of course we could add some result on a dozen more video cards into the graph, but that hinders effective perception of the data. So, we'd better restrict ourselves to four video cards, and in some cases we'll add results on other contenders.

3DMark 2003

The results for Ti4200 in the 3DMark 2003 benchmark can't be regarded as valid. In reality, this card can't show a twofold lag from those tested today - it simply skips the "Mother Nature" test since it lacks support for shaders. So, we look only at the FX5600Ultra and 9600Pro - the view is marvelous! From that it's absolutely clear why the slightly overclocked revision to FX5600Ultra was released. The 10% performance boost allowed to win the crown back to NVIDIA at the overall scores.

While in the "Battle Of Proxycon", "Troll`s Lair" and "Wings Of Fury" benchmarks the performance of 9600Pro was right in the middle between the new and old core of the FX5600Ultra, ATI takes a sure lead back as soon as shader processing is needed.

At multitexturing, none of the cards has been comparable to the Ti4200... curious, isn't it?

At shader programs, ATI solutions are head and shoulders above NVIDIA produce. By the way, the company has succeeded to improve the situation with the release of Detonator 45.23, albeit not that significantly.

3DMark 2001SE build 330

Readers sometimes ask "why are testers still presenting results measured with the outdated 3DMark 2001SE benchmark?"

The answer is quite straightforward - to put the cards which do not support DX9 into equal conditions with most recent up-to-date cards and estimate the real performance boost in the "old games", i.e. those played by most :)

As we see, there is some boost albeit not so critical. The regular overclocking of Ti4200 allows raising its score from 12456 points to 13812 achieved by the R9600Pro. The idea behind the new generation of video cards is in the advanced potentials of DX9.

Comanche 4 Bench

Without the AA enabled, the last-year Ti4200 easily overtakes the FX5600Ultra, be it the old or the new core. But that isn't a synthetic benchmark, but a real gaming application. With the AA enables, the situation changes sharply and all take their rightful places. This confirms the words of Alan Tike on the optimization of modern video cards solely for the image quality. His words regarding the "sufficiency of 30 fps" are disputable but the truth is there's no disastrous performance drop as new image quality improvement functions come into play. In fact, the ATI9600Pro performed with much more impressive results.

Codecreatures Benchmark Pro

Again the good old Ti4200 takes a lead over the FX5600Ultra old core (green) in which the performance was "dragged up" from its former disgraceful level to that of the whole group.

Village Mark ver1.17 PowerVR

This benchmark is old enough and offers some smart "feature" ... It seems like it hadn't come into the heads of the creators there might be times when the fps exceeds 200. That is, 200 fps is the limit above which the results are not displayed on the screen and the "0" is put instead. The Ti4200 was unable to issue results at 640x480 - its fps score exceeded 200 points, while the FX5600Ultra new core did a good job of that :-)))

By the way, the days of this benchmark are counted... the top-level cards (FX5900 and R9800Pro) show merely two figures - at the 1280x1024 and 1600x1200 resolution, and the line that joins the two points is in the upper right corner. A little bit more and the only thing that is left of the benchmark is the 1600x1200 point, at which the lifetime is over.

RightMark Video Analyzer rev0.4

Recently, the 0.5 version of the benchmark has been released, but on that very day we returned the FX 5600Ultra OTES card to ABIT, just the card based on the new chip, so we were unable to re-test it - that's still the only sample of the card all around Moscow. We kept only one old-core version in our test collection for good - and it will represent the FX 5600Ultra in all further tests :-)

UT2K3 Benchmark

GunMetal Benchmark 1.2S

We'd been looking forward to new games with support for DX9 and shuddered at their release. GunMetal is just the very "new-generation" game whose release has been so impatiently anticipated. If you remember the 10-year-old 2-dimensional Raptor, then the GM is just the Raptor but 3-dimensional, - a "fly-and-shoot" game. Just a regular arcade made in the Japanese anime style by Yeti studio. The unrealistic bright colors, flying transformers, coiling clouds, rocking grass, volumetric trees etc., etc... It's absolutely impossible to play that dreadful stuff due to these reasons:

1. The game is aimed at oligophrenic young teenagers...
2. The game is extremely sluggish at its default settings even at ATI 9800Pro 256Mb!!! What on earth are those poor 40 frames in the 640x480 resolution, while all are already used to 500 fps at QuakeIII? :-)

But those wise developers uploaded a separate program called "Gun Metal Benchmark" on their website and now are winning laurels of high citing rates. For lack of anything better the reviewers would simply have to test the DX9 implementation on that real gaming application.

Two recorded demos - Bench1 and Bench2 - come bundled. The rate at which the results are reproduced is very high, which is nice.

As we see, we won't be able to play GunMetal around on middle-level cards at the default gaming settings. 24 fps for for the ATI card and 15-17 for the NVIDIA card is absolutely useless. With the realism of clouds and grass disabled, things get better a little, but our goal is not to play to one's heart content, but to measure the DX9 implementation. The result is distressing for the NVIDIA card.

In the benchmark, there is the option for setting the "validity/invalidity" of results. As we got it, with the DX9 hardware acceleration enabled the results are always marked as "valid". The results produced through software rendering are regarded as "invalid". But things are not so simple as they seem...

For example, the Radeon9200 coped with the test quite easily (i.e. it was possible to play at the "slide-show" rate) and deserved its rightful "invalid" estimate, so didn't fall within the table, but the Ti4200 simply wouldn't start the game at all! Remember that both these cards do not support DX9. That is why we had to use only FX5200 as the low-end alternative.

For reference, here is a list of available video cards which were unable to start the game.

Video card Status

NVIDIA MX440-8x 64Mb ASUS not supported

NVIDIA Ti4200-8x 128 Mb Gainward not supported

NVIDIA Ti4800-8x 128Mb MSI not supported

SiS Xabre 600 64Mb PowerColor not supported

ATI 9200 128Mb Gigabyte It did start (although at 19 fps with 640x480 resolution)!
Benchmark Valid: No - some materials unsupported

Let's make the readers happy in the second demo embedded in the game - we add a few more video cards to make the view complete, because the test is absolutely new and fresh.

As we see, even the previous hi-end model, ATI 9700Pro (vinous color), takes a lead of the favorite of today, NVIDIA FX5900 (blue) at low resolutions.

In the middle level, the situation is the same - the 9600Pro (green) seriously overtakes both the 5600Ultra cards (crimson and yellow).

The findings regarding this test are simple - the requirements imposed upon DX9 cards are deeply staggering. A very well-timed solution from MS :). The video industry has much space to grow up.

Findings

Even with the slower (and thus cheaper) memory, the RV350-based video card (600 MHz versus 800 MHz) demonstrates some superiority over the NV31 new core. By the way, the power supply voltage of the Samsung 2.8 ns chips on the "Sapphire Radeon 9600 Pro Ultimate Edition" video card is definitely understated in order to reduce the heat emission - the rated frequency for these chips equals 700 MHz. It's absolutely evident that with the standard cooling the results will be even more impressive. At the same time, the card does not require additional power supply.

A couple of things more to note. The RV350 has around a thousand pins, which makes the wiring of the boards an awfully complicated and costly entertainment. Hardly any of the video cards manufacturers will do the job of developing a 8-layered board that differs from the reference design developed by ATI. That is, all Radeon9600Pro cards (like Radeon9800Pro) will look like twins. Only the cooling systems, textolyte color and package bundles will differ.

Things at NVIDIA are worse at that - various revisions to the core and floating memory frequencies guarantee a huge variety of FX5600-based video cards. Things will definitely not end up with Ultra and non Ultra, and instead of the clear-cut distinction of frequencies we'll get a variety of makes. As the saying goes, "frequencies have gone zooming". On the one hand, this is not bad... testers will be employed for years. On the other hand, manufacturers somehow hate marking the box with MERELY THREE DIGITS! These are the core frequency, memory frequency and access times, which has long been demanded by the consumer.