Sapphire Radeon 9700 Pro Review

By: Dmitry Zinovyev

I remember when in 1997 at ComTech we were anxiously watching Quake with its fantastic graphics played at Diamond Multimedia's expo stand. There was the flaunting caption 'DiamondMonster3D' proudly waving above the three or four monitors. The sight was so enchanting and touching that I decided I would definitely buy a 3D accelerator as the manager advised me in order to get such a beauty. No sooner had a week passed than I was walking around the radio market in Moscow hunting out for the long-awaited rig... But that time I was unable to get it because it hadn't been released by then. Anyway the vendors were unanimously claiming the future was already there and it was no use buying some monsters while IT was already in.

The 'it' was indeed impressive that time. The card was then called ATI Rage Pro, had 8 MB onboard and along with that supported 3D graphics acceleration. The price was equally as impressive, around $300. That time I was so obsessed with it that easily forked out for it without compunction. In about one and a half hour my joy turned into disappointment, because that time was enough to get home and try the game on my computer. ATI Rage Pro wouldn't run with Quake that time! It would run with almost no rigs except for a small DOS utility bearing the dear caption 'Test3D'. That's how my acquaintance with 3D gaming graphics and ATI started.

Years passed. It's sad to see the prices for top-end 3D accelerators almost haven't changed for the past five years, and if you want to get a best model you'll anyway fork out $300-$400. For these years, ATI has released chips and video cards of numerous makes, which helped ATi produce hold second to fourth positions. The cause of that was poor marketing, worthless programming team and the vicious practice of delaying chips shipments after the announce. The release of Rage 128 with the awful drivers seriously undermined the reputation of that fairly good chip and created a biased opinion of the company's products. The hasty release of first Radeon-based cards with raw drivers repeated the old bad mistake. The times were hard, and every month the lab was facing a market failure.

In no earlier than a year did ATI succeed in fixing the driver flaws, but the winning time for Radeon elapsed. And when the competition was shaping up to look more like NVidia's victory over the other contenders, ATi made a 'move of the knight' and released the new R200 chip. In fact, that was the last straw for ATI to take its niche and it did make use of it to her own advantage. Although released six months later than NV20(GeForce3), the R200-based cards (Radeon 8500) had several novelties appealing to potential buyers and looked like a breakthrough as compared with NVidia products. The N-patch technology dubbed by ATI with the nice term "TrueForm" and the adaptive antialiasing have not yet found adequate response from nVidia to date, and support for these functions is expected only in the new NV30 chip.

To visualize the current situation on the graphics chips market, I am bringing in a comparative table for the GeForce4 Ti4600 chips versus the recent ATI's developments:

After the release of NV25 (GeForce 4), R200-based video cards grew less impressive and gave in the crown to NVidia who had been alert all that time and through masterful pricing strategies for the GeForce line succeeded in taking hold of all the market niches. Traditionally, ATi responded belatedly and released the R300 chip which we are looking into. It is being positioned as a high-powered solution, that is why cards based on the chip cost around $300-$400. Recently, it finally came to ATI that one is as good as none, so the company has given up the idea of exclusive production and started delegating the manufacture of graphics accelerators to third-party manufacturers.

That decision is absolutely right since the competition of manufacturers will give greater dispersion of prices for cards based on the same chip as well as add more flexibility in choice for the end users. In fact, ATI switched to a distribution scheme similar to NVidia's. Of course, there is some specifics, but ... let's talk about it next time...

In the previous review, we described the theoretical part of the R300 chip. In this review, we'll try to explain how the new ATI's product behaves in practice. For that, we took the Sapphire card for testing, which can be viewed as the closest to ATI's reference boards since the line is produced at the same factories. On the whole, the current situation with ATI's R300-based cards is quite amusing. ATI ships to its manufacturing partners ready-made cards, and the partners merely fit coolers of their own designs on top of the chip and pack the finished product into their package bundled with software kits and cables. So it's almost impossible to come across a card made in a way other than the reference design, although as ATI assures things with it are going to change in the nearest time. But.. let's come back to Sapphire 9700 Pro.

Sapphire 9700 Pro

The card is made on a red PCB so appealing lately to all the manufacturers of Hi-End accelerators. The color doesn't play the leading part - what matters is that the PCB is six-layered and thus costly to manufacture. Another distinction of Radeon 9700 cards is the external power socket.

The socket is identical to that used for applying power to the FDD, but anyway every card comes bundled with a separate power cables to give enough flexibility in plugging the card to the PC's power supply unit. On the reverse side of the PCB, an additional radiator is installed for cooling the elements near the monitor output.

The number of intensely heating component parts in modern cards is increasingly going up. The Samsung (BGA) memory is installed on both sides of the PCB and marked as K4D26323RA-GC2A. Totally, there are eight memory chips on the card. This quantity is determined by the specific ways the 256-bit bus is arranged.

As per the manufacturer's online documentation on the website, the maximum speed of the memory is 350 MHz (700 MHz DDR), with the fetch time being 2.86 ns. The chip has stepping A13 and was produced on the 30th week of the year 2000.

Cooling system

The Sapphire 9700 uses a standard cooling system and is of pretty small bulk. This must have bewildered you slightly since the lead for additional power implies it greater consumption and thus more heat dissipated. The cooler is fastened with standard plastic pins but at the same time is fitted to the chip tightly enough. As for the layer between the chip and the cooler, it merits a special mention. Its touch resembles double-sided sticker, so the best thing to do with it is scratch off completely and replace with high-quality thermopaste. Otherwise, the quality of cooling may prove ineffective. By the way, - the memory. No cooling is provided for it, the way it was not in the reference design. What is good, the cooling of the graphics core won't hinder installation of a third-party memory cooling kit. If we carry through all the above actions, the operational stability of the card will be much better, and the overclocking potentials enhanced essentially, provided suitable utilities are used.

As per unconfirmed evidence, some cards feature an amusing bug in the way the cooling of chips is arranged. As a matter of fact, the side frame is by a third of millimeter higher above the core and prevents the cooler from fitting tightly to the chip. On the cooler there is a heat-conductive plate lubricate with thermopaste on both sides, so the efficiency of the system is out of the question - it is definitely ugly since the plate doesn't do the right heat-conductive job. You've got to make use of all your imagination to invent a new system for cooling the card after removal of the thermopaste. But don't be hasty at scratching it off - the simplest trick of adding five grammes of thermopaste into there won't work. With the gap as wide as that it will dry up in a couple of weeks, and the chip will have no cooling at all. So you won't do without a copper plate or a correctly milled cooler. Luckily, our card came without this exotics, and no problems arose; nevertheless, take care ... warned means guarded :).

2D

As with all ATI produce, the 2D quality has nothing to complain about. There are some fine points deserving a special mention in here. Firstly, our evaluation specimen displayed an annoying artifact in the form of horizontal lines on the dark background (I presume they are also there on the lighter background, although invisible). Those stripes appeared and vanished in a random manner, anyway were clearly seen. The artifact didn't depend on either the PC where we installed the card or the monitor. BUT, let me stress this out - I regard this artifact as a flaw detected in a specific specimen which can't be there in a whole line. We'll find that out in the nearest future. Another problem came up. After installing fresh beta drivers for DirectX9 it became no more possible to use a monitor plugged in to the DVI-out. The card simply wouldn't detect it; on the other hand, no problems with the TV-Out were found.

D-Sub DVI & TV-Out

When I connected all the three devices and powered on the PC, the monitor connected to the D-Sub as well as the NTSC-operated TV-set got activated. However, the monitor connected to the DVI wouldn't switched on earlier than Windows finished loading with the monitor enabled in the Control Panel. The TV-Out supports all the TV standards and allows displaying all the lower resolutions up to 1024x768 on the TV-set. The display quality can be regarded satisfactory since horizontal stripes (seemed to be of different nature, they were static) were again seen on the dark background. Another thing to note - since the card has only one DVI-connector, you can't plug in two DVI panels simultaneously.

Benchmarks

We tested the Sapphire ATI 9700Pro video card on a platform of the following configuration:

(Hi-End system):

• Processor Intel Pentium 4 2.8 GHz;
• Motherboard - Intel 850EMV (i850E, 512 Mb RDRAM PC800, AGP 4x);
• Sound card Creative Sound Blaster Audigy;
• HDD Seagate Barracuda ATA IV 7200 rpm

We tested the performance of AsusV9280S with the following suites and applications:

• Return to Castle Wolfenstein  - Checkpoint demo (OpenGL, Quake 3 core);
• Unreal Tournament 2003 Demo (Direct3D stress test benchmark);
• DooM III E3 Demo - extremely resource-hungry overkill benchmark with shaders, bump-mapping, OpenGL and lots of other stuff to be topical by the end of 2003;
• Codecreatures demo - complex DirectX 8 benchmark;
• 3DMark 2001SE - synthetic Direct3D benchmark.

3DMark 2001SE

On the one hand, this benchmark is the most impartial since the accelerator is tested throughout the whole range of 3D applications, and the results are made up of data received from an aggregate of various sub-benchmarks. Unfortunately, you can't test the cutting-edge features of the new accelerators with this benchmark, but this is not yet topical since other accelerators don't support them either yet, and to date there are still no games made especially for DirectX9 and ATi Radeon 9700 features. Those cool new benchmarks won't appear earlier than new state-of-the-art accelerators matching these functionalities hit the market. We are expecting the release of Unreal 2 and 3DMark 2003 especially for testing the performance of both state-of-the-art chips and Microsoft's new API DirectX 9. Only these benchmarks will be able to test the performance difference between the NV30 and R300. As for now, what we can do is to compare the R300 versus the past-generation video cards:

As is seen from this benchmark, the difference between the current NVidia's top-end chip and the card in question becomes visible only at higher resolutions, and the R300 leadership is by no means outstanding. That the ATI's product is taking the lead over nVidia's at performance is already a good showing. At the low resolutions, no difference is seen at all, and the Ti4600 is leaving the contender behind. It is absolutely evident the low resolutions are strongly affected by the processor resources. Keep in mind that in our tests we used the next to last Pentium4 model. It's very likely that on systems of lower clock speeds the difference at test results will be within the error.

The key distinction is not in the FPS, but in the functionality. As for now, the software is still unable to make much of the maximum benefit exhibited by Radeon. But Unreal 2 and DooM 3 as well as some other games are already coming and will be able to make use if not all but most of the R300 potentials.

Unreal Tournament 2003 Demo

A very hard and resource-hungry benchmark producing a torture of a load both on the processor and the video subsystem.

That's where R300 is showing all its muscles. Be the processor a bit weaker, the results for high and low resolutions would have differed in the minimum. Only at the maximum resolutions the load was enough to free the result from processor dependence and take all the contenders to town. In the two low resolutions, ALL the cards showed strong processor dependence, which means a processor much faster than Pentium4 2.8 GHz would be needed to increase performance, but the higher the resolutions the more vivid the difference is.

Return to Castle Wolfenstein

This benchmark can be freely regarded as a processor dependence test. This benchmark is most frequently used for testing weak video cards since its requirements to the graphics accelerator can't be regarded as high enough, because the Quake3 engine is quite an easy task for modern accelerators.

The trend is absolutely predictable, and the result was the same for all the resolutions. Only at the upper resolution a 10 fps departure from the main results was seen. This result gives one good opportunity - to use enhanced anisotropic filtering and antialiasing in the gameplay since the addition of these functions does not push the performance down.

Codecreatures demo

The benchmark is extremely hard to pass. Seems like the guys at Codecult were determined to make a killer benchmark able throwing any accelerator. The number of polygons in the frame sometimes exceeded 500000, which is approximately equivalent to a level shown by declared albeit un-released games or even better than that.

Judging by the results, we can see that it's not the processor, but the accelerator that is the main limiting factor in the Codecreatures benchmark. The NV30 will probably not change the situation. The trend is as before, and again the ATI 9700Pro looks like a higher stage in the accelerators evolution. A very indicative benchmark. That's how the ATI 9700 Pro's raw performance ranks versus NVidia's cards currently on sale.

DooM III E3 Demo

Sometimes you anxiously feel like peering into the future trying to see how well things will be then. Some our readers think the final DooM will be optimized after the release and will run faster than it does now. Others (among them myself) believe as soon as all the currently missing options are introduced into DooM (soft shades, complete physics regardless of location etc.) the game won't run faster than now at least.

On seeing the graph I could help exclaiming "Wow! That's cool!". What we can assert for sure is that Radeon 9700Pro will put up a good show at playing DooM3 and other D3-based games with worthy FPS. Of course, it's still hard to play at resolutions higher than 800x600, the FPS is low, but the higher resolutions are meant for further generations of chipsets - at least two generations will pass until the core is able playing at full capacity. However, the Radeon 9700 Pro performed better at 1024x768 than the GeForce 3 at 640x480.

The results for Sapphire card by the example of the demo are simply fantastic considering that ID Software hasn't optimized Radeon - DooM3 is optimized for NV25, NV30 and .... R100 - that is the regular Radeon. We can expect optimization for ATI R300 by the final version, which can't be said of NV17 and other value chips. Let's wait and play...

Findings

It's still a mystery what ATI has done this time. Either it has finally made a hit product that although released later than Ti4600 is able taking away quite a large portion of the market. Or ATi has simply outdistanced nVidia and released a card featuring all the functions of a new-generation accelerator and is able competing with the still inexistent nVidia's NV30. Let's wait for NV30 and see, it's still premature to make conclusions. But ... For now there is one evident thing - to make the most of the Radeon 9700 Pro performance, a powerful, if not the top-end, CPU is a must.

To make the card show its full capacity, at least a 2.5 GHz processor is required, otherwise the system will be ill-balanced. An autocade normally moves at the speed of the slowest chain, which is a rule, so it's useless to expect super-high FPS in Q3-driven games and good performance in state-of-the-art resource-hungry games with weak processors - the 3D accelerator alone doesn't count.

The new ATI's chip is the first sample of the state-of-the-art accelerators which make a big difference just at the functional level, not the performance level. As we have already noted, the performance of the chip is merely a next generation in the accelerators evolution albeit quite high. The functionality of R300 is currently an overkill, nevertheless in the near future more products where the functionality might be of use will be shipped in sufficient quantities.

Recommendation: it doesn't make sense rushing over to the nearest computer shop to buy a 9700 Pro to upgrade GeForce 3. Its current price is overly high, but to date there is still a lack of games or other products making use of DirectX 9, but if you are not planning to upgrade you computer in the coming six months or going to assemble a new system this accelerator is the best choice.

Pros:

• Unmatched 3D performance, especially at high resolutions;
• Hardware support for DirectX 9.

Cons:

• Poor cooling system;
• Artifacts in 2D;
• High price.