VOGONS


Nvidia 5800 Ultra vs Radeon 9800 Pro

Topic actions

Reply 20 of 35, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

I think we can all agree a 9800pro is a good replacement: as fast or faster in most scenario's and a more common card so easy to replace fairly cheap.

If you want Something faster you can use the 6800 card. That cpu is not fast enough to get the full potential of the 6800 but it is more than fast enough to show a difference and the 6800.

But 9800pro is a fine choice and will be a bit more period correct.

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 21 of 35, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie
meljor wrote:
I was playing around a couple of weeks ago and started testing the fastest portion of my Nvidia agp cards in 3dmark 2003 (standa […]
Show full quote

I was playing around a couple of weeks ago and started testing the fastest portion of my Nvidia agp cards in 3dmark 2003 (standard settings) and 2 Ati cards.

Arock conroe865pe, core2duo 3ghz, 1gbddr1, windows xp32 I do not remember which driver but it was one that could do geforce4 to geforce7 so there might be a better one for some cards (as always)

I don't have the FX5800 ultra but i do have cards that come very close. Here are some 3dmark '03 scores:

FX5600 ultra 128mb 3533 3dmarks
FX5800 128mb vanilla (stock clocks 400-800) 5012 3dmarks
FX5800 128mb @ 450-900 5620 3dmarks
FX5900 128mb @ 450-900 6696 3dmarks
FX5950 ultra 256mb (stock) 6991 3dmarks
6800 ultra 256mb (stock) 14646 3dmarks
7800GS 256mb (stock) 13487 3dmarks
7800GS 256mb @ 450-1400 15796 3dmarks
7900GS 512mb (stock) 17039 3dmarks
HD3850 (stock) 30137 3dmarks (yep, that's a beast!)
Radeon 9800pro 128mb (stock) 6021 3dmarks

Comparing the overclocked vanilla fx5800 to the radeon 9800 pro it should be pretty close considering your ultra is clocked at 500-1000.

I don't think in games the radeon will ever be slower. The card has a bit more oomph so the harder it gets the better it will be in comparison. Also with AA it is much stronger.
And i only have the 128mb version....

Your 9800 scored really really low. In fact I scored better with a much slower pentium D 945 (6464 pts). In general the 9800 nukes all FX series cards in DX9 - in fact the FX series is known to perform very poorly in DX9 while the 9xxx series (with the exception of the 9000 and 9200 witch are DX 8.1 cards based on the 8500) score very very well.

3dm03 Radeon 9800PRO 128MB P4D 945.JPG
Filename
3dm03 Radeon 9800PRO 128MB P4D 945.JPG
File size
160.02 KiB
Views
4109 views
File license
Fair use/fair dealing exception

In fact your 7800GS scored too low as well.. here's what I got with a X800XT on a measly P4 631 @ 3.7GHz:

3dm03 Radeon x800xt P4 631@ 3.75GHz.JPG
Filename
3dm03 Radeon x800xt P4 631@ 3.75GHz.JPG
File size
188.85 KiB
Views
4109 views
File license
Fair use/fair dealing exception

Something must be wrong with your test setup or drivers.

As for FX 5900XT vs Radeon 9800 in DX9, here's results for the 9800pro on a measly Athlon XP 3200+ vs a Pentium 4 @ 3GHz + a overclocked FX 5900XT (425/700):

FX 5900XT @ 425/700 / Athlon XP 3200+ overclocked to almost 2400MHz:

3dm03 fx5900xt 425_700 athlon xp 3200+ 2333MHz fsb166 abit an7.JPG
Filename
3dm03 fx5900xt 425_700 athlon xp 3200+ 2333MHz fsb166 abit an7.JPG
File size
167.19 KiB
Views
4109 views
File license
Fair use/fair dealing exception

Radeon 9800 PRO (stock) / Pentium 4 3GHz:

3dm03 Radeon 9800PRO 128mb P4 3GHz 1MB GA08IPE1000.JPG
Filename
3dm03 Radeon 9800PRO 128mb P4 3GHz 1MB GA08IPE1000.JPG
File size
150.51 KiB
Views
4109 views
File license
Fair use/fair dealing exception

The 9800 is quite a bit faster.

Last edited by kanecvr on 2017-02-19, 16:29. Edited 2 times in total.

Reply 22 of 35, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

Right - I have something a bit more relevant:

Radeon 9800 pro (stock) + Atlhon XP 2600+ @ 2130 MHz:

3dm03 Radeon 9800pro XP2600+ 2123MHz.JPG
Filename
3dm03 Radeon 9800pro XP2600+ 2123MHz.JPG
File size
278.01 KiB
Views
4109 views
File license
Fair use/fair dealing exception

Geforce FX 5900XT (stock) + Atlhon XP 2600+ @ 2130 MHz:

3dm03 FX5900XT xp2600+ 2123mhz .JPG
Filename
3dm03 FX5900XT xp2600+ 2123mhz .JPG
File size
259.37 KiB
Views
4109 views
File license
Fair use/fair dealing exception

Like I said, the 9800 is quite a bit faster in DX9. It's even more apparent in games. The difference should only increase when using a fast CPU like a core 2 duo.

agent_x007 wrote:
"No driver optimisation" you say... OK :) Quake III Arena [Demo001] (1024x768 w/TF) : 9800 XT (412/770) : 367,3FPS 5950 Ultra (4 […]
Show full quote
swaaye wrote:

I really wouldn't trust 3dmark scores. The driver optimization controversy and all that.

For a good time, try Oblivion on a FX series card. 🤣 9800 runs the game fairly well.

Think of the FX 5800/5900 as a upgraded GF4 Ti. It is great in that role.

"No driver optimisation" you say... OK 😀
Quake III Arena [Demo001] (1024x768 w/TF) :
9800 XT (412/770) : 367,3FPS
5950 Ultra (475/950) : 421,3FPS
5950 Ultra > 9800 XT, by ~15%

Doom 3 v1.0 build-in benchmark (1280x1024) :
9800 XT (412/770) : 37,5 FPS
5950 Ultra (475/950) : 31,2 FPS
9800 XT > 5950 Ultra by 20%

Dunno what driver and test setup you used there, but my testing on a period correct system shows the opposite, at least in regards to the FX 5900XT and the 9800PRO. The FX 5950 Ultra is just marginally faster then the XT, the only difference being an overclock witch is only really noticeable in DX9 titles. In fact beyond a certain clock speed, faster cards scored slower then slower clocked ones in direct 3d 7 unreal - see my benchmarks here: Re: 99-03 video card performance scaling (Q3, Unreal, DK2, 3DM2k, 3DM2001)

Here's quake 3 @ 1280x1024 on a Athlon 64 3800+ (single core, socket 939, 2400MHz, HT set to 800MHz, single channel ram to emulate a socket 754 3400+):

6kmtqb6.png

The 9800 series is faster then the FX5900XT here. Even my quadro FX 3000 flashed to FX 5900 Ultra failed to take the lead from the 9800 PRO, despite much higher clocks (especially vram) but it scored closer - 281 fps if memory serves. I have a MSI 5950 Ultra, but it fails to complete most benchmarks (it hangs part-way). Pretty sure one or more of the capacitors have gone off. I'll fix her up and include test results when I have the time.

The FX series are pretty good cards - just not the 128 bit FX 5800. In all seriousness, there are faster 5700 Ultra cards out there. Have a look at this:

rPb2SGV.png

Yes, the 5800 has 4 extra texture units, but the 5700 ultra has a 75MHz core clock advantage and 150MHz on vram in the case of the GDDR3 version - but there's DDR versions of the 5700 that use 850 and even 900MHz GDDR - the Abit Siluro and Albatron 5700 Ultra come to mind. These cards should perform similarly.

The point I'm trying to make is it's not worth going after the FX 5800 unless it's to stick it in a display case. They are hot, not very fast, and very very rare (at least in eastern europe). If I had one it would sit framed in the center of my collection - using it in a system and risking it dying is not worth it when you can use an equivalent 5700 or faster 5900XT witch are easy to find and rather plentiful.

Last edited by kanecvr on 2017-02-19, 04:12. Edited 1 time in total.

Reply 23 of 35, by Tommaso72

User metadata
Rank Newbie
Rank
Newbie

I thought my score for 3DMark 2003 was rather low, but looking at what others got it is actually a good score. I got 5740 with everything set to default with a P4 Northwood HT with 533 FSB and only 512 megs RAM on the 5800 Ultra. I guess the ram makes no difference in the tests? Just curious if my score would be higher with 1 or 2 gigs?

Tommaso

Reply 24 of 35, by PhilsComputerLab

User metadata
Rank l33t++
Rank
l33t++

Double check the card, I do remember there was a 9800 SE or LE? With disabled features.

YouTube, Facebook, Website

Reply 25 of 35, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

@Phil No LE card, i'ts a normal clocked 128mb version, like the one kanecvr used.

@kanecvr Nothing wrong with my setup or the driver. We talked about this before, it is really hard to completely fairly compare these cards without first checking out which is absolutely the best driver for each card. Will take forever. Because of this we can never compare each others scores OR CALL THEM WRONG without the same setup and drivers....

I stated i took a driver that supported geforce4 to geforce7, so there will always be cards that would be faster with a more ''period correct driver'' and maybe the latest cards are best with the latest driver, who knows.

kane, you're talking about ''wrong'' and are trying to show me ''right'' but all you come up with are really small differences. Yes, the 9800pro is faster (like i stated before).
I came up with a score of around 6000, you counter with 6200......really??? Who says radeons scale to the moon? Me using a core2duo doesn't make a big difference apparantly. Someone above me came with nearly the same result and he had a 4ghz core2duo! Against my 3ghz setup. He had 60points more.....

I didn't ''pick'' a driver for the 9800pro. I simply tested my Nvidia's (from geforce4 and up) and finished with the hd3850 just for extra fun. I then bought the 9800pro card and wanted to know if it was working, the setup was still complete so plugged it in and tested (with the 3850 driver).

So no, my results aren't wrong as i was mostly playing and wanted to know if my cards still worked and wanted to compare some (got some FX quadro scores and gf4 scores too).
And no, my 7800gs results are not low, the card is NOT faster as a 6800ultra as was stated in many reviews. Both are 16P cards and the ultra has the higher core clockspeed, the 7800gs has the higher memory speed. 7900GS has both and more pixel pipes and was the fastest Nvidia. Seems right and works for me.

Your x800xt scores beeing so high i don't know, maybe 3dmark loves cpu clockspeed? 3,7ghz is no slow p4. Maybe it is because your card does not support sm3.0 and that way the scores get higher as the card has it easier in some tests compared to the 6800/7800/7900?

It's a shame that it's always ''your score is too high!'' or ''your scores are bad'' or ''your scores are too low!'' I can come up with release reviews of cards that did massively different with different reviewers, but they didn't have the same systems/games/drivers so there is no comparison.

BTW, do you really believe you OWN scores? An unlocked 6800 16p card is only about 15% faster as a 8p card? Sure that came out at 1280x1024. Try again at 1600x1200 or with AA and the 16p card is twice as fast as the 8p card. Your benches do not show that. It's hard to do it right.....

I do a LOT of testing/benchmarking and i mostly keep the scores to myself because of the reason there is always someone bitching about the results. 😎

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 26 of 35, by PhilsComputerLab

User metadata
Rank l33t++
Rank
l33t++
meljor wrote:

I do a LOT of testing/benchmarking and i mostly keep the scores to myself because of the reason there is always someone bitching about the results. 😎

Man that does resonate. I totally get where you're coming from. I get this often as well. How dare this card be faster than my beloved card from my childhood 🤣

Especially with 3D Mark I can do 3 runs and get quite different results. So whenever I do a shootout, I do write down three runs to really see what's going on and catch any outliners.

I also choose drivers for other reasons that just performance. For example I rather have one NV driver to use with all cards, although I know that an older driver might give an older card a bit of a boost.

And that's just with a single game. This issue gets bigger the more games you test. Suddenly this game gets a boost, but that one runs a bit slower. And then you try SLI and discover you can't reuse your results because, well SLI and CrossFire are even more driver version dependent.

So when people see "this card is a bit faster than that card", you should really be thinking "On this board, CPU with that amount of RAM, OS, Driver version, game and settings, this card is a bit faster than that card". I think that would help for everyone to calm down 😀

People telling me "there's something wrong with your system", man, they aren't worth the effort. They come in so hard and opinionated, there is no ground to even enter a dialogue, not worth bothering, just ignore them.

I don't have much experience with the 9800 and FX series (yet), but I agree with you on the 7800 GS. I used it for my last videos in S1 of Intel vs AMD and I had to overclock it quite heavily to have it go faster than a 6800 Ultra. At stock I reckon the 7800 GS is slower or on par with a 6800 GT when I did my tests if I remember correctly.

And I also believe that you got to find suitable benchmarks and use a mix of synthetic, built-in and FRAPS benchmarks to get a better picture. For example AMD are strong in Far Cry and Half-Life, Nvidia is better in Doom 3. And once you get 200+ FPS, it's definitely time to move as some minor CPU optimisation can affect results.

Ignore the haters, just keep on doing your thing 😘

YouTube, Facebook, Website

Reply 27 of 35, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie
kanecvr wrote:
Dunno what driver and test setup you used there, but my testing on a period correct system shows the opposite, at least in regar […]
Show full quote
agent_x007 wrote:
"No driver optimisation" you say... OK :) Quake III Arena [Demo001] (1024x768 w/TF) : 9800 XT (412/770) : 367,3FPS 5950 Ultra (4 […]
Show full quote

"No driver optimisation" you say... OK 😀
Quake III Arena [Demo001] (1024x768 w/TF) :
9800 XT (412/770) : 367,3FPS
5950 Ultra (475/950) : 421,3FPS
5950 Ultra > 9800 XT, by ~15%

Doom 3 v1.0 build-in benchmark (1280x1024) :
9800 XT (412/770) : 37,5 FPS
5950 Ultra (475/950) : 31,2 FPS
9800 XT > 5950 Ultra by 20%

Dunno what driver and test setup you used there, but my testing on a period correct system shows the opposite, at least in regards to the FX 5900XT and the 9800PRO. The FX 5950 Ultra is just marginally faster then the XT, the only difference being an overclock witch is only really noticeable in DX9 titles. In fact beyond a certain clock speed, faster cards scored slower then slower clocked ones in direct 3d 7 unreal - see my benchmarks here: Re: 99-03 video card performance scaling (Q3, Unreal, DK2, 3DM2k, 3DM2001)

Here's quake 3 @ 1280x1024 on a Athlon 64 3800+ (single core, socket 939, 2400MHz, HT set to 800MHz, single channel ram to emulate a socket 754 3400+):

The 9800 series is faster then the FX5900XT here. Even my quadro FX 3000 flashed to FX 5900 Ultra failed to take the lead from the 9800 PRO, despite much higher clocks (especially vram) but it scored closer - 281 fps if memory serves.

What driver/test setup I used is visible in attachments 😀
Pentium Dual-Core "E5700" @ 4GHz with 4GB of DDR2 and 4CoreDual-SATA2 R2.0 MB
Drivers : Catalyst 9.1 for 9800 XT and 93.71 for FX 5950 Ultra.

As for "period correct" hardware : CPU/RAM botlleneck can augment results.
In case of slower CPU/RAM setup, Radeon may be faster (has lower footprint driver ?).
BUT, if your CPU is fast enough, FX 5950 Ultra will gain more from it in DX7 games than 9800 XT.

PS. 7600 GT AGP is usually faster than 6800 Ultra AGP in newer titles (at least that's what I got from my tests).

157143230295.png

Reply 28 of 35, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

@Phil I have a bit of a problem when it comes to ignoring, but i will keep doing what i do 😎 Thanx.
I also value input and no one needs to agree but always starting with how ''wrong'' others are with benchmarks is a bit lame imho, especially as it is very hard to duplicate results with completely different hardware/drivers.

At the time i didn't even understand the launch of the 7800gs. Yes, it can be a tiny bit quicker in some tests that value memory speed more but otherwise it is the same/slower as the 6800ultra.
The former was a lot more expensive to make so in that way the 7800gs was a good deal, they just should have given it at least 20 pixel pipelines and higher clocks . The 7900gs is what 7800gs should've been.

@ agent_x007 I also have a 7600gt agp and indeed it is in the same league as the 6800ultra and the 7800gs. In some scenario's it is really quicker because of the very high clockspeeds, when the pixel shaders really matter the others can be faster. I didn't test the 7600gt with the others because it is in a system and i didn't want to take it out.

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 29 of 35, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie
meljor wrote:

@Phil No LE card, i'ts a normal clocked 128mb version, like the one kanecvr used.

@kanecvr Nothing wrong with my setup or the driver. We talked about this before, it is really hard to completely fairly compare these cards without first checking out which is absolutely the best driver for each card. Will take forever. Because of this we can never compare each others scores OR CALL THEM WRONG without the same setup and drivers....

Let me start by apologizing - I've upset you, witch was not my intent. Please let me explain - not an native english speaker, and I probably didn't manage to convey what I meant to say correctly. What I mean by "wrong" is that the system is not performing as well as it could. I am genuinely interested in how much the card would have scored with such a fast setup.

We have talked about driver setup before, and we did agree on the fact that it's very very hard to come up with a driver setup that will provide good performance on all cards - but I would have expected over 7500 with the core 2 duo @ 3GHZ and the 9800 PRO - especially since 3dmark tests are very very CPU bound.

The screenshots I posted are not a means to boast. In fact they are very mediocre scores. In my attempt to try and convince OP that the 5800 is not worth the investment, and that it's such a rare card it should be saved for benchmarking and displaying, I may cave come off as a bit of an asshole, and if I did I apologize... 🙁 I did not mean this as any sort of personal attack, and I'm not doubting your skills in any way. Old tech is "finicky" and things like this happen. It would be a shame if you'd discount the 9800 as a poor performer based on that.

meljor wrote:

I stated i took a driver that supported geforce4 to geforce7, so there will always be cards that would be faster with a more ''period correct driver'' and maybe the latest cards are best with the latest driver, who knows.

kane, you're talking about ''wrong'' and are trying to show me ''right'' but all you come up with are really small differences. Yes, the 9800pro is faster (like i stated before).
I came up with a score of around 6000, you counter with 6200......really??? Who says radeons scale to the moon? Me using a core2duo doesn't make a big difference apparantly. Someone above me came with nearly the same result and he had a 4ghz core2duo! Against my 3ghz setup. He had 60points more.....

All video cards scale very very well in 3dmark with very fast processor, because the tests also measure CPU performance, and because a faster platform will be able to feed more data to the card then a slower one - but you already know this. I personally managed to score a bit over 7k in 3dmark03 with a similar setup, using a C2D 6600 and a crappy AGP / PCI-E combo board (witch died while ago). Again, I'm not disputing the result, I'm just saying it can go higher.

meljor wrote:

I didn't ''pick'' a driver for the 9800pro. I simply tested my Nvidia's (from geforce4 and up) and finished with the hd3850 just for extra fun. I then bought the 9800pro card and wanted to know if it was working, the setup was still complete so plugged it in and tested (with the 3850 driver).

So no, my results aren't wrong as i was mostly playing and wanted to know if my cards still worked and wanted to compare some (got some FX quadro scores and gf4 scores too).
And no, my 7800gs results are not low, the card is NOT faster as a 6800ultra as was stated in many reviews. Both are 16P cards and the ultra has the higher core clockspeed, the 7800gs has the higher memory speed. 7900GS has both and more pixel pipes and was the fastest Nvidia. Seems right and works for me.

I didn't know the 7800GS and 6800 ultra were on par with each other. I frankly expected the 7800GS to be faster, despite the similar core configuration. Guess you learn something everyday 😀

Using the 3850 driver explains why the 9800 under-performed. It's most likely using a basic form of it's driver w/o any GPU-specific optimizations, in witch case your bench score is actually pretty high, considering that's the cards "raw" performance.

meljor wrote:

Your x800xt scores beeing so high i don't know, maybe 3dmark loves cpu clockspeed? 3,7ghz is no slow p4. Maybe it is because your card does not support sm3.0 and that way the scores get higher as the card has it easier in some tests compared to the 6800/7800/7900?

The P4 is much much slower then a 3GHz core 2 duo, and like I said before, 3dmark is quite CPU bound. Not only clockspeed but architecture as well. The fact that the x800 lacks SM3.0 support has nothing to do with it's performance. In fact it should slow it down. What you are seeing is "brute force". The benchmarks are genuine, and if you doubt them I can reproduce them anytime, film the benchmarks and post them to youtube 😀 The X800XT is faster then the 6800 in some games and benches, and that's a known fact. Until the 8800 series nvidia and ATi were extremely competitive with each other, after witch nvidia took a clear lead.

meljor wrote:

It's a shame that it's always ''your score is too high!'' or ''your scores are bad'' or ''your scores are too low!'' I can come up with release reviews of cards that did massively different with different reviewers, but they didn't have the same systems/games/drivers so there is no comparison.

Please don't take this personally. I look at these things objectively, as my only interest is to judge actual performance today. Scores in period correct reviews are worthless since we have at our disposal newer hardware and a wide selection of drivers for these cards, witch reviewers did not. I never said ''your scores are bad' - I said "it scored too low" - witch should imply that the card can do much better, and nothing else. It was by no means a personal attack.

meljor wrote:

BTW, do you really believe you OWN scores? An unlocked 6800 16p card is only about 15% faster as a 8p card? Sure that came out at 1280x1024. Try again at 1600x1200 or with AA and the 16p card is twice as fast as the 8p card. Your benches do not show that. It's hard to do it right.....

Yes, they are 100% genuine, taken as an average between 3-4 runs with the same card, and literally took MONTHS to compile. Before that I spent a 3 weeks looking for a suitable platform on witch the cards will scale well, and before that it took me 7 weeks to find the right driver to use for all cards - witch was Catalyst 6.2 for ATi cards, and Forceware 66.93 for nvidia cards. These two provided an increase in performance when moving up to newer cards, they ran all games and benchmarks on all cards were other versions would crash or black screen, and provided the least result anomalies.

like I said, if you or anyone else doubts the scores, I'm not above posting recordings of the cards running them on youtube to prove they are genuine. I know some of them are hard to fathom, like how the 8500 performs so well in unreal, or how in the same game lower clocked cards score better - I don't understand this myself really, I actually took a clip of the 8500 running unreal after I noticed it was so much faster then much newer cards. Still have it on my phoone.

As for the 6800LE - at 1280x1024 you can see it take a major lead over the 8p version - lead witch increases to 38% at 1600x1200. The reason I did not run q3 at 1600x1200 is because the slow cards ran it as a slideshow... so I saw no point. I did bench the faster of the cards at that resolution, and the X800XT is on top because of faster vram (the 6800LE has 700MHz DDR, witch I overclocked to 800MHz for the 16p tests - but it can't handle more then that. A 6800GT would have 1000MHz GDDR and would have come on top of the X800XT in quake 3 @ 1600x1200 as well). At lower resolutions like 640x480 the small difference between the 8p and 16p version comes down to a CPU bottleneck, witch I explained in the review. Just look at your Q3 tests - you score 400+ FPS because of the much much faster 3GHz core 2 duo. I would presume the maximum framerate a 6800GT card can achive at 640x480 w/o a CPU bottleneck is around 450-500 fps, and same for the x800xt.

PhilsComputerLab wrote:

People telling me "there's something wrong with your system", man, they aren't worth the effort. They come in so hard and opinionated, there is no ground to even enter a dialogue, not worth bothering, just ignore them.

I completely disagree with you phil. When benchmarking, taking the time to find the best platform and drivers is part of the job. I believe that as a reviewer it's your duty to bring out the best possible result for the cards you are testing, and driver / platform tests are MANDATORY, not optional. It makes sense to me to find the fastest driver (for both video card and chipset) and a platform that will run both competing cards as fast as possible with no hardware incompatibility issues. Please forgive me, but I see your approach to this as superficial.

Your videos are great tough. I can only imagine the sheer amount of time it takes to put one together - setting things up, taking cuts, adding commentary and so on. Personally I couldn't do it. My only complaint really is that I believe you could have been a bit more thorough in some of them, because of the huge amount of work it took you not only to do the benchmarks but also film and edit them in a pleasant and presentable way. Your videos are extremely relaxing to watch, and your voice somehow invokes a calming atmosphere reminiscent of the "good old days". I love coming home from work and sitting down to one of your videos with a cup of coffee, relaxing and reminiscing about the good old days. Sometimes I have one of your or LGR's or Plaul's videos running in the background while doing my own tests. It creates a great atmosphere. Great channel!

When I did my round of testing I went as far as to use two identical drives with separate windows installs as to not have old driver leftovers affect benchmark results. I tested several motherboards for performance and compatibility with all the video cards in the test - that's 28 video cards tested... 28 cards tested with 7-9 video drivers (hell for nvidia cards since there are so many drivers) and 28 cards tested with 4-5 chipset drivers... I managed to fill a notebook with tables and general scribbles. Actually ruined the AGP connector on one of my mainboards doing this - not to mention the sheer amount of time and research this entails - so please forgive me when I get a little "over excited" when I see someone has overlooked things witch I see as obvious, witch give some of the cards an edge. It is clear that I have to chance the way I "speak", as to avoid offending others, witch is what I apparently did here 🙁

As for "opinionated" - there's a lot of that going on on this forum and in IT in general. I try to avoid that, and since I love all brands (Intel, AMD, nvidia, ATi, 3dfx, you name it) I try to be impartial. If you look at my IT history you will see that over the years I've owned all brands, and found something I loved about all of them - and I reminisce fondly about all.

There is no such thing as the "fastest card" or "best card" in most cases. Today it all comes down to price / availability and compatibility with your current hardware and the games you want to play.

Reply 30 of 35, by swaaye

User metadata
Rank l33t++
Rank
l33t++
agent_x007 wrote:
"No driver optimisation" you say... OK :) Quake III Arena [Demo001] (1024x768 w/TF) : 9800 XT (412/770) : 367,3FPS 5950 Ultra (4 […]
Show full quote

"No driver optimisation" you say... OK 😀
Quake III Arena [Demo001] (1024x768 w/TF) :
9800 XT (412/770) : 367,3FPS
5950 Ultra (475/950) : 421,3FPS
5950 Ultra > 9800 XT, by ~15%

Doom 3 v1.0 build-in benchmark (1280x1024) :
9800 XT (412/770) : 37,5 FPS
5950 Ultra (475/950) : 31,2 FPS
9800 XT > 5950 Ultra by 20%

Doom3 is interesting. GeForce FX 5700/5900 have some stencil shadowing speed functions, but the game uses a lot of fancy effects and GeForce FX of course is awful for shader programs. The game uses some texture lookup tables in place of arithmetic for some effects, in order to keep the shader math load lower. This hurts ATI in some cases AFAIK (look up the Doom3 Humus tweak).

I also wonder about how the behavior varies across driver versions. Sometimes newer drivers are faster and less buggy, sometimes they have regressions. Try playing KOTOR on your 9800. I think only Catalyst 4.2 gets you all effects. GeForce FX drivers changed over the years too, with shader compiler improvements and different shader replacement and texture filtering optimizations. It's also not unlikely that FX driver support went downhill pretty quick considering how great and loved the next generation was.

Reply 31 of 35, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie
kanecvr wrote:

The reason I did not run q3 at 1600x1200 is because the slow cards ran it as a slideshow... so I saw no point. I did bench the faster of the cards at that resolution, and the X800XT is on top because of faster vram (the 6800LE has 700MHz DDR, witch I overclocked to 800MHz for the 16p tests - but it can't handle more then that. A 6800GT would have 1000MHz GDDR and would have come on top of the X800XT in quake 3 @ 1600x1200 as well). At lower resolutions like 640x480 the small difference between the 8p and 16p version comes down to a CPU bottleneck, witch I explained in the review. Just look at your Q3 tests - you score 400+ FPS because of the much much faster 3GHz core 2 duo. I would presume the maximum framerate a 6800GT card can achive at 640x480 w/o a CPU bottleneck is around 450-500 fps, and same for the x800xt.

6800 Ultra [GPU = 400MHz VRAM = 1100MHz], can do 700FPS+ on QIIIA with 1024x768, just FYI 😀

[Demo001] GeForce 6800 Ultra (400-1100) mini.png
Filename
[Demo001] GeForce 6800 Ultra (400-1100) mini.png
File size
594.22 KiB
Views
4004 views
File license
Fair use/fair dealing exception

157143230295.png

Reply 32 of 35, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie
agent_x007 wrote:
kanecvr wrote:

The reason I did not run q3 at 1600x1200 is because the slow cards ran it as a slideshow... so I saw no point. I did bench the faster of the cards at that resolution, and the X800XT is on top because of faster vram (the 6800LE has 700MHz DDR, witch I overclocked to 800MHz for the 16p tests - but it can't handle more then that. A 6800GT would have 1000MHz GDDR and would have come on top of the X800XT in quake 3 @ 1600x1200 as well). At lower resolutions like 640x480 the small difference between the 8p and 16p version comes down to a CPU bottleneck, witch I explained in the review. Just look at your Q3 tests - you score 400+ FPS because of the much much faster 3GHz core 2 duo. I would presume the maximum framerate a 6800GT card can achive at 640x480 w/o a CPU bottleneck is around 450-500 fps, and same for the x800xt.

6800 Ultra [GPU = 400MHz VRAM = 1100MHz], can do 700FPS+ on QIIIA with 1024x768, just FYI 😀

[Demo001] GeForce 6800 Ultra (400-1100) mini.png

Sweet 😀 It would be interesting to see how it would do on something even faster, like say a modern core i5 or i7. This shows how important the CPU and platform are for game performance. I'll probably start a thread about this at one point.

Can you post results with it running demo1 at 640x480? I'm curious just how much it scales from an athlon 64 to a core 2 quad.

Reply 33 of 35, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

Well, it should scale really good 😀
Here's my Pentium 4 EE OC'ed result with 6800 Ultra (it's comperable to Athlon 64), same settings as on QX9770.

Demo001 GeForce 6800 Ultra (400-1100) [1024x768] TF mini.png
Filename
Demo001 GeForce 6800 Ultra (400-1100) [1024x768] TF mini.png
File size
634.5 KiB
Views
3989 views
File license
Fair use/fair dealing exception

That QX9770 OC and my ASRock board are good to around 900 FPS (which both 7900 GS AGP, and HD 3850 AGP, reach on it).
To have Core i5/Core i7 test, PCI-e version of any GPU is required unfortunatly...

157143230295.png

Reply 34 of 35, by Tommaso

User metadata
Rank Newbie
Rank
Newbie

I just tested my 6800 Leadtek 400A / P4 3.06 HT 533 FSB with 512 megs DDR 333 RAM on Unreal 3 bench mark and I got 260 FPS with everything set to full and 1024 X 768. Just curious does this seem a little low? I was happy with my Doom 3 score of 72.9 with everything set to high, but I am a little disappointed in this Unreal 3 score. Could it be the 6800 is not as efficient with older game bench marks? Just curious on what all of you think.

I do believe I will stick to the 6800 for this build. I have a 3850 AGP that I could use but it isn't period correct and that is important to me. I wonder how much faster it would be on this system? I am going to have to see.

Tommaso

Reply 35 of 35, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

Without actually adding anything substantial to the discussion here, I just wanted to chime in to say that I find it a very interesting one to read 😀

When benchmarking things depend on many factors and the results (and opinions extracted from these results) reflect real world opinion on retro-computing quite well actually. What's important for one, does not have to be important for another.

Heck, when it comes to performance, I can vividly imagine that in some cases having a faster card in the rig (which results in better and more stable framerates and less slowdowns) may actually be sooo annoyingly loud that it will actually blow up my concentration, which then results in me performing more crappy when playing through games (head shots require focus and focus requires my rig to not be obnoxiously annoyingly eardrum-bleedingly loud, no matter if I got over 9k frames/second or not 🤣!). Or it will make my concentration run out of fuel sooner, making me take longer to actually finish playing through a certain campaign (this is easier to imagine when using a monitor that gives a headacke after a little while vs using a monitor that does not).

In such cases I'll actually prefer the actual slowdowns, kinda like how Super Mario Bros had slowdowns with 3 or more enemies..heck I even used those slowdowns kinda like bullet time 🤣!

But don't mind me, keep on doing the good stuff! 😁

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!