nVidias NV38 is here. NordicHardware brought a board back from the latest visit at Gainward. This means that time has come yet again for another world première here at NordicHardware. (Please tell us if you tire and we’ll try to settle down.) Aside from the Gainward NV38, we have two other FX cards from Gainward and another one from Creative…
nVidias NV38 is here. NordicHardware brought a board back from the latest visit at Gainward. This means that time has come yet again for another world première here at NordicHardware. (Please tell us if you tire and we’ll try to settle down.) Aside from the Gainward NV38, we have two other FX cards from Gainward and another one from Creative.
Finally, it’s time for NordicHardware to take a look at Gainwards CoolFX and we didn’t settle with one package – we have looked at two of them: CoolFX 1600XP and 1800XP. And what’s so special about this solution? Well, the fact that Gainward’s GeForce FX 5900, 5900 Ultra, 5950 Ultra and even 5700 Ultra is avaliable from retailers with complete watercooling systems make these products completely unique. No other manufacturer of today offers this. The closest we’we come to seeing this with other manufacturers is prototypes with pre-mounted waterblocks.
Recent development in this area tend to thrill us computer enthusiasts. It wasn’t that long ago the first watercooled OEM computer was introduced to the Asian market and now the time has come for graphics cards. As you might have heard, there are an escalating conflict brewing between ATi and nVidia. The main reason for this conflict is that for the first time in ages we have two manufacturers that can compete more or less on equal terms. Equal from a technical viewpoint and equal when it comes to sales opportunities and last but not least pretty much equal in terms of PR reign.
The situation as it is today will force more manufacturers to attempt to create unique and innovative solutions. You need to make your product stand out in the crowd so to speak. This is in ways a paradox since we currently have an unusually homogeneous high-end market where almost every Radeon 9800 Pro is practically the same if not identical to one and other, and the same goes for the GeForce FX 5900 Ultra cards.
The most intruiging part of this review will of course be to see how the NV38 (FX 5950 Ultra) will hold up against the R360 (Radeon 9800 XT). But let us not get ahead of ourselves. Let’s start with a look at the specifications of the four cards.
Unfortunately, we did not have the actual package available when this review was made, but at least we managed to get our hands on some of the printed material.
Chip: |
NV35
|
Manufacturing process : |
0.13-micron
|
Transistors: |
~125 mil.
|
GPU: |
500 MHz
|
Memory speed : |
900 MHz / 28,1 GB/s
|
Pixel Shader: |
2.0
|
Vertex Shader: |
2.0
|
Pixel Pipelines/Pixel Fillrate: |
4 / 2000 MP/s
|
TMUs/Texel Fillrate: |
2 / 4000 MT/s
|
RAMDAC: |
(2) 400 MHz
|
Memory size: |
256 MB
|
Memory type and interface: |
256-bit, DDR-SDRAM
|
In and outputs: |
VGA, DVI-I, S-Video (S-Video In/Ut, Composite In/Ut)
|
Accessories: |
ViVo-dongle, DVI-VGA-adapter, ATX-Y-cable, water cooling kit
|
Software: |
Expert Tool
|
Fullversion software : |
WinCinema
|
Estimated price: |
7500 SEK (exchange rates)
|
There are a lot of things among the specifications to look closer at. First off, the clock frequencies are higher than the ones on nVidia’s reference card. 500/900 versus 450/850 to be exact. Personally, I did hope for a little more than that as there already is a 5900 Ultra from eVGA with the same frequencies, only using ordinary air cooling that is. Last but not the least is the price, and you can say it really hurts your wallet badly. No less than 7500 SEK is nicked from your pockets if you want a CoolFX 1600/XP, which, despite all the accessories and watercooling gadgets, in the world is an insane price for a graphics card.
|
|
A complete description of the CoolFX cooling is presented on a later page.
Regarding in and outputs, it is very nice to find a Video In on this card. If you are not satisfied with S-Video and composite. Obviously, there is one VGA and one DVI, and if you want two VGA there is also a DVI-VGA adapter.
The accessories are the same for both 1600XP and 1800XP and more details concerning them can be found on page six.
We’ve gotten our hands on a early sample of Gainwards NV38-based CoolFX 1800XP. The one we got is built on nVidias reference PCB but final retail version will ship on Gainwards red PCB with some few added tweaks for more overclocking headroom.
Chip: |
NV38
|
Manufacturing process : |
0.13-micron
|
Transistors: |
~125 mil.
|
GPU: |
550 MHz
|
Memory speed : |
1050 MHz / 32,8 GB/s |
Pixel Shader: |
2.0
|
Vertex Shader: |
2.0
|
Pixel Pipelines/Pixel Fillrate: |
4 / 2200 MP/s
|
TMUs/Texel Fillrate: |
2 / 4400 MT/s
|
RAMDAC: |
(2) 400 MHz
|
Memory size: |
256 MB
|
Memory type and interface: |
256-bit, DDR-SDRAM
|
In and outputs: |
VGA, DVI-I, S-Video (S-Video In/Ut, Komposit In/Ut)
|
Accessories: |
ViVo-dongle, DVI-VGA-adapter, ATX-Y-kabel, vattenkylningskit |
Software: |
Expert Tool
|
Fullversion software : |
WinCinema
|
Estimated price: |
8500 SEK
|
The official NV38 specifications set the clock frequencies to 475/950 MHz. Gainward decided this was not enough and went for 550/1050MHz instead. Thanks to these amazing clock frequencies Gainward breaks the 30 GB/s bandwidth barrier which bodes for great FSAA performance. Without a doubt Gainward has made sure this is the fastest retail nVidia board ever made. How it holds up to the 9800 XT is examined later on. The price tag is intimidating but actually it’s not much more expensive than the CoolFX 1600XP.
|
|
As mentioned : we’ll take a closed look at the cooling later on.
Cooling, features and bundle is identical to the 1600XP.
Creative helps us out with their latest 5900 board.
Chip: |
NV35
|
Manufacturing process : |
0.13-micron
|
Transistors: |
~125 mil.
|
GPU: |
400 MHz
|
Memory speed : |
700 MHz / 21,9 GB/s |
Pixel Shader: |
2.0
|
Vertex Shader: |
2.0
|
Pixel Pipelines/Pixel Fillrate: |
4 / 1600 MP/s
|
TMUs/Texel Fillrate: |
2 / 3200 MT/s
|
RAMDAC: |
(2) 400 MHz
|
Memory size: |
128 MB
|
Memory type and interface: |
256-bit, DDR-SDRAM
|
In and outputs: |
VGA, DVI-I, S-Video
|
Accessories: |
ATX-Y-cable |
Software: |
nVidia Demos
|
Fullversion software : |
Gun Metal
|
Estimated price: |
3000 SEK
|
Even though this Creative board is built on the NV35 chip just like the CoolFX 1600XP is much cheaper. It’s a non-Ultra board and it sports slow memory, 700 MHz to be exact. These new 5900 boards running at 400/700 are pretty fresh on the market and represents the “low budget” section of the 5900’s. Apart from the bundled game GunMetal (which is an awesome game by the way) there’s not much to look at here in terms of the bundle.
|
|
Creative buying MSI-boards is no secret. A quick comparison of this cooling compared to the one MSI sports speaks for itself. I did like the fact that the fan is very silent, almost as quite as the Inno3D 5900 boards (the most silent 5900 board we’ve ever tried).
Unlike the two CoolFX boards from Gainward Creative does not support Video In on this videocard .
As mention the bundle is not the greatest ever to say the least. Not even all needed cables/adapters are included. Though as mentioned the fast paced space shooter GunMetal is included and bodes for hours of fun.
The fourth and last board is once again a Gainward product.
Chip: |
NV35
|
Manufacturing process : |
0.13-micron
|
Transistors: |
~125 mil.
|
GPU: |
400 MHz
|
Memory speed : |
700 MHz / 21,9 GB/s |
Pixel Shader: |
2.0
|
Vertex Shader: |
2.0
|
Pixel Pipelines/Pixel Fillrate: |
4 / 1600 MP/s
|
TMUs/Texel Fillrate: |
2 / 3200 MT/s
|
RAMDAC: |
(2) 400 MHz
|
Memory size: |
128 MB
|
Memory type and interface: |
256-bit, DDR-SDRAM
|
In and outputs: |
VGA, DVI-I, S-Video
|
Accessories: |
DVI-VGA-adapter, S-Video-Komposit-adapter, ATX-Y-kabel |
Software: |
Expert Tool
|
Fullversion software : |
WinDVD 4.0
|
Estimated price: |
3000 SEK
|
Gainward 5900 “low budget edition” is equipped with the same clock frequencies and even the same price tag as the Creative 5900. The board does seem to have a bit better cooling with memory heatsinks and all. Since this isn’t a “Golden Sample” there is no Enhanced Mode and the standard clocks will have to suffice.
|
|
Hopefully the memory heatsinks will add some overclocking headroom. The fan itself was pretty silent though not as good as Creatives board.
The situation is pretty similar to Creatives though instead of a full version game Gainward bundles a full version DVD player. Unlike Creative Gainward do bundle all the necessary cables and adapters.
Shall we take a closer look at the CoolFX package then?
|
|
The box’s contents are divided into two layers. On the first, we find the software, the cables, tubes and manuals etc and on the other, we find the true beauty: the card, the radiator with fan, the pump, the water tank and the anti-corrosion fluid.
|
The pump is from Eheim; a name that watercooling enthusiasts should know about. The model which is used is the 1046 – a light-weight in the assortment. However, it is more than enough for the purpose and it has also got an acceptable level of noise. As you can see, the water tank is small. It should be, as we’ve only one component to cool so anything larger would have been overkill.
|
The radiator is from Innovatek, which also provides the GPU block and the tubes, and is approximately 145×135 mm in its measurements. The radiator and the cooling block are both made in aluminium. And for the radiator to be of any use, we also need a…
|
A 120 mm fan from, the to me, unknown Y.S. TECH manufacturer. Despite running on 12 volts, the fan is quiet, although if you do find the noise a problem you could always lower the voltage without a major problem. Variable fan RPM as standard wouldn’t make us complain though.
|
If you look at the image, you can easily think that the card is really small, but as most of you know a 5900 Ultra, and now 5950 Ultra, is a pretty space requiring card. Space requiring is also CoolFX and to mount it in your case (i.e. you have the radiator and the pump standing on the bottom of the case) it practically requires you to have a very big case or no PCI cards installed on the mainboard. An example of how you can mount the whole thing can be seen below:
|
As is shown below, it’s pretty tight in the case here. To take the air from hot harddrives, it is not the optimal solution either. Best is if you can manage to mount the radiator on the rear side of the case so that it blows out the hot air in an orderly manner. If you want to fiddle around some, you can probably find some good solution on the whole thing but it is not painless. On the other hand, it is actually the cooling’s effectiveness in practice which is to be judged and it works very good to say the least:
|
Above you have an example from 1600XP at maximum overclock. The idle temperature is a constant 30 degrees celsius on both the cards and at full load the temperature increases to about 50. It doesn’t take many seconds, maybe 5, for the card to be down at 30 c after you have finished a 3D application. In other words, the cooling is very effective.
|
Among the accessories, we find the ViVo dongle which tells us that the cards has Video In. The WinCinema package from Intervideo has software for DVD playback, video capture, CD ripping etc. The blue joints in the right corner are for mouning the tubes in strange directions and are important to use if there is risk for the tubes to be folded. Gainward has even been nice enough to include an ATX contact which makes it possible to start the watercooling without starting the computer – valuable if you want to look after leakage before you turn on everything.
What the thermal paste is good for, I have yet to figure out, but then again it doesn’t hurt to have itthere.
What I have forgotten to document is the anti-corrosion fluid, which is hardly anything exciting. For the interested, I can tell that it is made by Innovatek and it is called innovaProtect.
Not much has changed in our test system except for changing the mainboard to an ABITs NF7S 2.0. We had some minor stability issues with the Shuttle board we used earlier. We would like to take the chance and thank ABIT for helping us with the mainboard.
Test system
|
|
Hardware
|
|
CPU: |
AMD Athlon XP 3200+ (400) Mhz
|
Cooling: |
Corsair HydroCool 200
|
Mainboard: |
ABIT NF7S 2.0 (nForce2 400 Ultra)
|
RAM: |
768 MB DDR333 @ 2-5-2-2 Timings:
3x 256 MB Corsair TWINX512-3200LL DDR-SDRAM |
Graphics cards:
|
Gainward GeForce FX 5900 (128 MB, 400/700) |
HDD:
|
80 GB Western Digital Caviar 7200 RPM Special Edition (8 MB cache)
|
Sound card:
|
Creative Soundblaster Audigy 2 ZS Platinum Pro
|
PSU: |
Nexus NX-3000 300W
|
Ethernet:
|
D-Link DFE-530TX 10/100 |
Software
|
|
Operating system: |
Windows XP Professional (Service Pack 1 + uppdateringar)
|
Video drivers: |
nVidia: Detonator 50 52.16
ATi: Catalyst 3.8 |
Other drivers: |
nVidia UDA Chipset Drivers v2.45
|
Benchmark applications:
|
Unreal Tournament 2003 (v2225) |
Just like in the Radeon 9800 XT previwe we have used 12 benchmarks. The difference this time is that we swapped out SPECViewperf 7.1 with Halo: Combat Evolved. This is yet another game using DirectX9 effects, and that should make it interesting to test.
Regarding the drivers: we used Detonator 52.16 and Catalyst 3.8 for this review. Gainwards CoolFX 1800XP (5950 Ultra) is sold with the 52.16 on the installation CS, and they may therefore be regarded as “official”. nVidia confirmed that the 512.16 will be the official drivers for NV36/38.
In the review you will find quite a lot of expressions and abbreviations, if you are not familiar with them we suggest you take a look in the list below. In the same list we have some other expressions used specificly in this review (they are not always used).
- AA, FSAA (Full Scene Anti Aliasing) – Makes edges appear smoother (reduces “saw tooth graphics” in 3D-graphics), all modern graphics cards should be able to handle some kind of AA. AA can be used with different “effect” 2, 4, 6 x osv. The more the merrier.
- AF, Aniso (Anisotropic Filtering) – Increases sharpness/focus on textures/surfaces in 3D-graphics), all modern graphics cards should be able to handle AF as well. Just like with AA you can use it with different effects, the more the merrier in this case too.
- App (Application Preference) – If something is marked with this suffix it means that we have let the game decide how much AF to use instead of manually setting it in the control panel.
- Trilinear and bilinear filtering – Two different kinds of texture filtering used separately or combined with Anisotropic filtering. The highest quality setting is Trilinear and Anisotropic filtering. Right now, Trilinear filtering is standard for newly released games and must be considered a minimum for acceptable quality using the measures of today.
- Aliasing – Graphic errors due to insufficient information/precision. Most common are the “dancing” pixels on edges or in textures. By using AA and AF you can reduce aliasing.
If you find anything you don’t understand you are more than welcome to contact me with your questions using the adress found in the upmost part of this page. (Please remember that we are not an idealistic support service, for support questions we suggest that you use our forums or the retailer where you bought your products.)
What we didn’t have time to test is the image quality of the new Detonator 52.16. We might update this review later or perhaps write a separate article on that topic. We can say as much as that we don’t find many differences compared to the older drivers. If you look at our preview of 9800 XT you’ll get a good grasp of the differences. We can add that the problems we had with image quality using the Detonator 51.75 are now somehow gone.
To make the diagrams viewable, as we’ve got results from 11 cards in 10 bars (Gainward’s 5900 and Creative’s 5900 share bar, since the greatest measured difference was less than 1%), we’ve marked the cards based on nVidia with green bars and those based on ATi with red.
In all tests we used the resolution 1280×1024 with 4x AA (Full Scene Anti Aliasing – in other words edge smoothing) and 8x AF (Anisotropic filtering – in other words advanced texture filtering) if nothing else is written. We chose the 1280×1024 resolution (or 1280×960 i the games that didn’t offer 1280×1024) since this is a reasonable setting regarding performance, but also because most of our readers have got monitors that supports this resolution.
In the tests where 1280×1024-4xAA/8xAF has proven too demanding we have primarily chosen to lower the resolution to 1024×768 and then lowered and finally disabled AA/AF until a playable level of performance is reached.
After doing the actual performance tests we spent approximately 30 minutes (in some cases a lot longer than that) sitting down playing the game properly to feel how it actually works in real life.
We do not have any comparative screenshots in this review, since we’ve been in an awful hurry. We’re hoping to being able to return with screenshots and also some performance results from an additional 3 tests. Another thing about the image quality that hasn’t changed with NV38 is nVidia’s FSAA-implemention. So concerning the edge smoothing in all of today’s tests ATi is still the King upon the throne of benchmarking and testing. On the other hand I can say that we haven’t found any direct differences in image quality from any of the two competitors. The cards are more or less identical in most tests. What differs under the current circumstances are the differences in AA and AF. ATi has a quite large lead concerning AA, while nVidia has a pretty small (everything based on my own personal opinions) lead concerning AF.
|
Quake 3: Arena
|
The Open GL based game Quake 3 is used in these tests for evaluating the performance in the somewhat older titles. A large amount of titles are based on the “Q3”-motor. We use the test demo four.dm_67 in the testing utility Q3Bench.
|
|
The nVidia-based cards crushes ATi in Quake 3 and CoolFX is the final word as it gives almost 100 fps more than a Radeon 9800 Pro 256 Mb.
Subjectiv analisys: Quake 3, the grandfather of all of today’s 3D-tests, works equally good on basically all the cards we’ve tested. Sure, there is a slight noticeable difference between the fastest and the slowest card, but that is very small.
|
Unreal Tournament 2003
|
UT2003 is a DirectX 8-based game that puts massive pressure on the graphics cards with large textures and high numbers of polygons etc. A number of games are based on this motor. We use the more graphically demanding flyby test. We’re testing the game on the map Inferno.
|
|
The Radeons are unmatched winners in UT2003. The only card that succeeded in sneaking past even the slowest Radeon is CoolFX 1800XP. The two 5900s from Gainward and Creative are quite slow and has got a lot worse performance than the “regular” 5900.
Subjective analisys: No matter if looking at figures or just “feeling the flow” the ATi-based cards are better for UT2003. Even though the ATi-based cards’ lead here is percentually less than that which nVidia has over ATi in Quake 3, it is in the lower performance environments that those percents really counts (for example no one notices the difference between 250 and 500 fps, but the difference between 25 and 50 fps is quite substantial).
|
WarCraft 3: Reign of Chaos
|
WarCraft 3 is own of this year’s bestsellers, which makes it a good game for testing. Despite the fact that the graphics lack extravagant techniques it is still a pretty demanding game. The performance tests are done on the first level in the WC3 demo with the help of FRAPS.
|
|
ATi’s cards take the lead once more, the only card that overtook the Radeon 9800 “non-pro” was the extreme-cooled and overclocked (and three times as expensive) CoolFX 1800XP.
Subjectiv analisys: Despite some differences in the numeric results I can’t say that you feel any differences in performance even when comparing the fastest and the slowest card. In the matter of this occurrance the image quality is what settle it, and here ATi is in the top with high quality FSAA. The graphics (in better words the points of view) are not of the type where Anisotropic filtering is visibly different between the cards.
Quake 3 is so far the only game where nVidia still has the advantage, but we have only done a third of all of our tests yet. Next game is Mafia, a personal favorite with which I spend a lot of hours.
|
Mafia: The City of lost heaven
|
Mafia is based on a Direct3D engine (developed by the Mafia team) which uses large amounts of low-quality objects to get a high detail level. Similar 3D engines can be found in i.e. GTA. We have used Free Rides first level with FRAPS to measure the performance.
|
|
The Radeon cards doubtlessly have a vast advantage in Mafia. Even ATi’s “non-Pro” kicks som serious CoolFX 1800XP butt.
Subjective analysis: Mafia runs smoother and looks better with ATi; it is sometimes that simple. However, the difference is far from that big as our tests imply. The game runs quite nicely even with the lower clocked 5900 cards.
|
Comanche 4
|
Comanche 4 is based on a self-developed Direct3D engine which uses Pixel and Vertex Shaders together with high resolution textures. The game is one of few which really “needs” a graphic card with 256 MB. We’re using the downloadable demo’s benchmark tool.
|
|
Detonator 52.16 increases the performance slightly and keeps nVidia on the top. The margins are small, though.
Subjective analysis: My personal experience is that Comanche works very good with all cards except Radeon 9800 “non-Pro”, which makes the game lag from time to time.
|
Counter-Strike
|
Counter-Strike does not need any presentation. The new 1.6 version is claimed to be more demanding than before. Therefore, we chose to test it. The benchmark consists of a demo from the map cs_havana and a total of 18 players. |
|
The difference here is insignificant. Works just as well with 5900 as 9800 XT.
Subjektive analysis: In my opinion, I think that ATi and nVidia’s cards are on par in Counter-Strike. The image quality is a little bit higher with ATi, but for the most people CS is like breathing, and you won’t actually see the small things. Though if I would have to choose, the XT is my choice.
Six tests are done and four remain.
|
Battlefield 1942:
Secret Weapons of WWII |
The Swedish-developed Battlefield 1942 is still a very popular online game and therefore it feels important to test. Yet again, we use the tool FRAPS and we test the downloadable demo. |
|
The Radeon cards are prominently faster in Battlefield – there is no doubt about it.
Subjective analysis: All BF fanatics should look for something else than nVidia based graphic cards. The game flows and looks better on ATi’s cards. Mainly you escape the frustrating mouse lag which occur in some games with the FX card when using a high screen resolution and FSAA activated.
|
Tomb Raider:
Angel of Darkness |
Tomb Raider is the first commercial game which completely utilises DirectX 9.0 for its graphics and therefore it is an interesting test object. The test is performed with the latest patch’s build-in tool. The level Prague3a was chosen for purpose. |
|
Those of you who’ve seen our 9800 XT preview remembers that ATi completely crushed nVidia in Tomb Raider. Detonator 52.16 saves the day in both image quality and performance, but still ATi holds a very tight grip around the four top positions.
Subjective analysis: Despite a good job with Detonator 52.16, Tomb Raider is hopelessly slow on nVidia’s cards with the settings we use. What our average FPS in the diagrams does not reveal is that ATi’s cards has a relativly even FPS all the time, while nVidia’s cards have higher peaks but mainly significantly lower dips.
|
Star Wars Jedi Knight:
Jedi Academy |
Jedi Academy is the sequel to the popular Jedi Knight II. It is built around the Q3 engine but has among other things very high resolution textures and more light effects. We have tested our own demo recording from the map Traspir where we met 7 opponents. |
|
nVidia rules in Jedi Academy. Especially CoolFX 1800XP impresses with a whole 22.5% better than a 9800 XT!
Subjective analysis: Detonator 52.16 corrects some problems we had with FX and the game runs very good on the nVidia based cards; clearly better than on ATi’s. All you Jedi wannabies out there should make sure the sticker on the box says nVidia the next time you buy a graphics card.
New this time around is the game Halo which has been developed by Bungie/Gearbox. The game was released for Xbox and “converted” for PC just recently. The PC version uses DirectX 9 functionality in certain aspects and is therefore another game of the so called “new generation”, so to speak.
|
Halo: Combat Evolved
|
You could call Halo the first real DirectX9 game, which of course makes it very interesting. We test the performance by adding -timedemo in the command line. That measures the performance in the game’s cutscenes and gives a pretty decent picture of the card’s performance in the game.
|
|
In Halo the cards are very close. CoolFX 1800XP does not manage to express itself in an extra ordinary way compared to the other cards and in principle the performance is the same with the seven fastest cards. The test’s looser it Gainward and Creative’s 5900 card with 700 MHz memory and Radeon 9800 “non Pro” since the performance with these are a bit short.
Note: The nice fellows at Beyond3D has informed us of some slight problems with the shader quality on nVidias boards in Halo and other games that use pixel shaders when in conjunction with anisotropic filtering. We’ll examine this in a follow up article.
Concluding word about the game tests |
In the last review, we were sorry for not having Halo in the test suite. Now, it’s here, and we see that ATi and nVidia’s cards are neck-to-neck in the test. What still is missing are the two delayed giants Half-Life 2 and Doom 3.
The review’s newcomer, GeForce FX 5950 Ultra, shows to be a very minimal upgrade of FX 5900 Ultra. (Just like Radeon 9800 XT is a pretty minimal upgrade of 9800 Pro.) Those of you that have already have a 5900 Ultra or an equivalent card from ATi don’t have to worry. 5.5% faster core and 11.8 % faster memory can be translated to about 3-8 % better performance, which is actually somewhat less than what we gained on the leap from 9800 Pro to the 9800 XT. But that is to be quite honest hairsplitting; regardless if we’re talking about 5 or 10 % difference, it’s terribly hard to tell, if not impossible.
The spectacular CoolFX cards perform nicely and actually accommodate an essential difference. CoolFX 5950 Ultra is, calculated as a percentage, “more” faster than 5950 Ultra “standard” than what 5950 Ultra is faster than 5900 Ultra. (What should I say? The sentence was difficult to formulate.) Which is an honorable exploit. Furthermore, we can also see that a CoolFX 1600XP principally offers the same performance as a 5950 Ultra. Though the question is: why would anyone want to buy a CoolFX 1600XP right now when 5950 Ultra and CoolFX 1800XP are released? On that question, I don’t have a suitable answer and I don’t think Gainward has either.
Creative’s 3D Blaster FX5900 and Gainward’s FX PowerPack! Ultra/1100 TV/DVI belongs to the new wave of cheap 5900 cards with slower memory. The problem with the two cards is that they are both in the same price range as a whole bunch of “regular” 5900 (i.e. those that have got 850 MHz memory) which renders these two competitors as pretty superfluous products. Besides, are you willing to pay another ten or twenty euros, you can actually get a Radeon 9800 Pro 128 MB.
Speaking of raw performance, we can quickly state that ATi pulls off the prize in 5 of 10 tests while nVidia does so in only 3. The other two cards are about so equal that we have to count the differences in promille instead of per cent. This despite that we have included Gainward’s extreme-cooled CoolFX 1800XP speaks for itself. If we shall count “grand slams”, i.e. tests where one competitor totally crushes the other, we also see that it is of ATi’s advantage where they wrestle down nVidia’s cards in 3-4 tests while the nVidia cards only manage to get one real neck swing on ATi in the now outdated Quake 3. The nVidia cards’ major benefit in the review is Jedi Academy, but before we include that in our “final grade”, I suggest you read the text about the results at the top bottom on the previous page.
What is interesting to see when we now have this many tests is how different the cards perform in different tests. For example, if you are a die-hard BF1942 fan, there is no doubt that you should buy an ATi card, while the Quake 3 veterans rather should look in nVidia’s direction. For this reason, we will continue to increase our test suit until we have about 15-20 games so you readers can make such a well-informed decision as possible.
I begin to feel that perhaps I’m sounding tedious when, I in review after review, come to the same conclusion but the fact is: ATi has both the better performance and the better image quality in the majority of the tests we expose the cards to. The Detonator 50 series levels out the differences to some extent, but still doesn’t reach all the way. CoolFX, despite its cooling, can’t reach all the way through either, even if it helps to keep on track.
Watercooling and overclocking, two words which usually reside in symbiosis. When it comes to Gainward’s boards they got a bit too cocky when they said that their card not only had the highest standard clock frequencies, but also overclocked the best. If you look at the results below, mainly for the 1800XP it becomes awkwardly obvious that it is far from the truth that their CoolFX overclocks well. In short Gainward has already maxed out the card. 1600XP is a little better thanks to the memory overclocking on the other hand the core overclocking is even worse.
But on the other hand, the same time that Gainward stands for the worst overclockers they also stand for the best when their 5900 reaches nice clock frequencies with an increase of on the memory frequencie of a fully 18 %. Creative’s 5900 also manages fine, 10 % is not very good but it is far from bad.
Product
|
Standard
|
Overclocked
|
Per cent
|
CoolFX 1800XP
|
550/1050
|
570/1070
|
3,6/1,9
|
CoolFX 1600XP
|
500/900
|
510/1010
|
2,0/12,2
|
Creative 5900
|
400/700
|
440/780
|
10,0/11,4
|
Gainward 5900
|
400/700
|
460/825
|
15,0/17,9
|
When it comes to CoolFX it is simply a question of both eating the cake and still keeping it. The reason that it does not overclock any good is simply that Gainward has already done it for you. Therefore the percentual gain is very small despite that the card might get some higher clock frequencies than other 5900 Ultra and 5950 Ultra. Though you really pay for the overclocking that Gainward “does for you” since the cards are so much more expensive than similar products.
Well, although Gainward willingly name their Golden Sample as special chips, memory and other components and price them for good overclocking it is often their cheaper products that really has overclocking potential and can deliver a very competitive price/performance ratio when they are overclocked. The most impressive card we wave seen from this angle is with no doubt (and now I am not only talking about Gainward) Gainward’s 5900 256 MB aka Ultra/1300 TV/DVI. If it comes down to it it is really that what is the point with overclocking: getting more free performance. And as long as Gainward charges for overclocking for you the overclocking looses it point.
If you look at it from the “regular person’s” point of view the situation is somewhat different. If you don’t want to bother with “manual” overclocking or if you simply don’t want to risk destroying your products Gainward’s Golden Samples are very nice products for those who are looking for performance that other manufacturers can’t deliver.
Unfortunaley you have to please yourselves with the raw numbers on how the cards are clocked. The time pressure we had when we wrote the review was too tough for us to manage to test and make graphs of the results. Hopefully we can deal with that in the following article which we mentioned earlier.
The tale begins to reach its end. There has been a whole lot of information and expressions in the review today so we have actually got two pages of conclusions. Turn the page for the final.
That was all of it and with that we’ve given you an insight of how CoolFX and maybe above all else how the GeForce FX 5950 Ultra performs in practice.
5950 is built on the 5900 series but just like the Radeon 9800XT we hardly see any great improvements. Apart from the extra MHz we haven’t found any differences and the fact is that the card actually is built on the same PCB as the ordinary 5900 Ultra. From this point of view the Radeon 9800XT actually has improved more since they, apart from raised the clock frequencies calculated as a percentage more than nVidia, actually has rebuilt their PCB, added a temperature guard and and thus their OverDrive technique.
Eventually 5950 has less architectual tweaks. For example we have heard that they have modified their core somewhat to gain higher yields (which means something like number of succeeded tries per try to manufacture a chip, i.e. if we try to manufacture 10 chip and only 8 works we have a yield of 80 %). As far as we can see, this hasn’t affected the performance in any way, so principaly we are only looking at a small raise of the clock frequencies.
What NV38 has is a new cooling, but since we only have tested Gainward’s own version of NV38 with watercooling we haven’t had any chance to evaluate the new reference cooling from nVidia.
If we are going to speak about the specific cards for a while instead we can begin with the CoolFX. Gainward has begun with what you may call a pretty reckless project. The new CoolFX packages are extremely expensive. They cost as much as an OK PC (2 GHz, 512 MB RAM, monitor etc.) so the question is: how will you justifice this price.
My first thought, and I presume that this is what most of other people also might think, was that the compensation was in pure performance. But that is not the case. With the CoolFX series Gainward aim more against the gadget maniac more than the computer enthusiast or the crazy overclocker. The card is twice as expensive as one without watercooling but offers far from double the performance. The fact is that you pay almost twice as much for an average of 7.5 % extra performance (5950 Ultra compared to CoolFX 1800XP).
The target group is what is of weight in the coherence and Gainward’s hope is that this will succeed among the people who always have to have the latest cell phone, 60 inch plastma tv and so on.
If they succeed or not remains to be seen on the sale numbers. Personally I would actually want to say that we don’t really have a market for “gadget maniacs” in the computer world yet, but I hope that maybe, Gainward will be there to open up such a market. We have the extreme overclockers, we have the crazy modders but might not have the correspondence to a Bang Olufsson yet. I.e. the TV that might not be that extremely better in a technical point of view but has that little extra which makes the gadget maniac within you come to life. That rumour, that design, even that picture of a special type of person. In other words despite that we already have expensive products within the computer world as of today and I think, and really hope, that there also will be a market for “luxury products”, because even of you can’t afford it yourself, it is damn fun to try, look at and wonder.
Now when the computer is becoming an equally ordinary property like the TV, the stereo etc. Gainward’s timing might be right, who knows.
What I like with the idea is also that they have simply got the guts to try something new. Watercooling is according to many the way of the future and even if I personally hope that the manufacturers instead manages to develop the techniques in such a way that the heat dissipation will be lower (Low K, SOI, smaller manufacturing process, better transistor economy and so on, are examples on how this may be done) but the prospectives are probably right when they say that we will need extreme cooling pretty soon.
Gainward has also thought even further than just graphics cards cooling. Their upgrade solution, CoolPC, gives the opportunity to connect cooling for both northbridge and CPU in your CoolFX system. With everything in a nice red color (and perhaps red colored water) it looks damn nice and gives better cooling than most of the existing air cooling solutions. (We hope to evaluate exactly how good at a later occasion.)
Watercooling is often followed by silence. And the level of noise on the CoolFX cards is low, if not as low as on the real quiet graphics cards which we have seen, for example Inno3D’s 5900 “non ultra”, but in any case it is much quieter than nVidia’s reference cooling for the 5900 Ultra. (How it measures against the reference cooling for the 5950 Ultra remains for us to find out.)
The previous page was dedicated to CoolFX and NV38 in general. Lastly, before we end the review it is time to summarize the four cards which we have reviewed today. For you with bad memory, it looks like following:
- Gainward CoolFX PowerPack! 1600XP – NV35/FX5900 Ultra, watercooled, 256 MB, 500/900 MHz
- Gainward CoolFX PowerPack! 1800XP – NV38/FX5950 Ultra, watercooled, 256 MB, 550/1050 MHz
- Gainward FX PowerPack! Ultra/1100 TV/DVI – NV35/FX5900, 128 MB, 400/700 MHz
- Creative 3D Blaster 5 FX5900 – NV35/FX5900, 128 MB, 400/700 MHz
As we have seen in the performance results, ATi still has the trump cards on hand with its Radeon 9800 XT; this despite Gainward’s two extreme CoolFX cards. ATi also sports, what I and many with me consider to be, the best image quality. If you are quite frankly out to get the fastest card on the market, with great image quality, the situation hasn’t changed since the Autumn of 2002: ATi still rules.
If we instead look at the two cheaper cards from Gainward and Creative, we have some difficulties to see how they will manage in the offered resistance. Both among ATi’s cards but also in the own paths. The fact is that a “regular” 5900 with 150 MHz faster memory doesn’t cost more than approximately €10-20 more than these two cards and under these circumstances, neither 3D Blaster FX5900 nor Gainward 1100 TV/DVi is something to prefer.
Regardless of how nVidia’s cards stand up against ATi’s, they are still doubtlessly very fast. Someone that buys a NV35/38 based graphics card today will hardly have any problems making the games flow like they should. But in the end, we still don’t consider any of today’s cards as a given buy. If we disregard what many consider to be tiny performance differences and finicky image comparisons, the fact still is that CoolFX is a damn cool product serie sand I really hope that others will follow Gainward’s example.
|
|
||||
|
|
If you are looking for something different, something with an exclusive touch, something that will make your friends green of jealousy, CoolFX 1800XP is definitly something to look closer at. We also think it’s very positive that Gainward hasn’t rationalised away the Video In support on these cards since it actually provides more value. For you who only want a damn fast graphics card, there are smarter choices for us who don’t light the cigar with burning dollar bills. CoolFX is not aimed at us deadly people if I understand Gainward correctly, but the company has certainly given us something to drool about. Thanks to its watercooling in particular but also its courage to hunt new consumers in new categories, we hand over our Innovation distinction to Gainward’s CoolFX card:
![]() Gainward CoolFX |
The two slower cards from Creative and Gainward are more boring and fall headlong in my opinion. None of them has anything special speaking for them, but on the other hand, there are many products in the same price range which speaks against them. An ordinary 5900 “non-ultra”, or even better – a Radeon 9800 Pro 128 MB – is by far a better choice and only costs about €10-30 more which in this price range is pretty insignificant.
A tip on the hottest Christmas gift this year when it comes to graphics cards seems to be either a CoolFX 1800XP or an ASUS Radeon 9800 XT depending on what you’re after. Personally, I have written to Santa and asked for a package with an ASUS logo on it. We still have a little over two months left but the question is if it really is time for something to happen in that time. Of course, you can trust in NordicHardware keeping you posted.
Finally, I wish to remind that we, hopyfully, will launch what you might call “part two” of this review. With more test results, image comparisons etc. although I don’t want to promise anything yet since the schedule is “pretty” tight here an NH.
Thanks to Gainward and Creative for supplying the boards for this review