Pitiful performance of the V5-6000

[RoboTECH]

*****EDIT******

the card was benchmarked with texture compression disabled. The author of the article confirmed this in an email to me. Please ignore previous comments.

 

[IBMer]

I don't think VE tweaked the card at all.

You also have to remember at 90+fps the V5-6000 is CPU limited. NOT fillrate limited like the V5 5500.

I belive if they had used a T-bird 1Ghz and the new 1.04 drivers (Not the HSR ones) that it would have done a lot better.

 

[RoboTECH]

understood, but drivers or no drivers, in fillrate limited situations (1024x32 w/4xFSAA or 1600x1200x32), drivers shouldn't matter

the thing should do better, but it doesn't, so it was canned.

Doesn't instill much confidence in me for their "high-end" Spectre solution

no wonder they didn't want anyone measuring framerates when they were demo'ing/previewing it.

 

[edm]

I read on another site (forget where) that had tested the overseas version of the 6000 which was released, they reported getting over 60 FPS in 1600x1200 32-bit with 4XFSAA Enabled, not too shabby.

 

[lsd]

I think they had major driver issues. Those score are much too low. IMO I think 8x fsaa would have been a slide show.

 

[RoboTECH]

well, you can only flood the "PCI" channel so much

this is a case where 4xAGP would've been helpful, I bet.

 

[EMAN]

If you actually read the article than you would know that voodoo 5 6k drivers are only the first ones that 3dfx came out with. It's basically still in beta form and untweaked drivers. Do you remember how much controversy there was when voodoo 5 5500 scores were so much slower than what it is today? People were dissing the 5500 so hard. Give it some time or do we have another voodoo 3 3500 in our hands....


E-----MAN

 

[jpprod]

...they reported getting over 60 FPS in 1600x1200 32-bit with 4XFSAA Enabled...

I don't think so. 1600x1200 at 32bit color/Z and with 4X FSAA would requre...

1600 x 1200 x (4 bytes color + 4 bytes Z) x 4 samples per pixel x 2 buffers = 122 megabytes

...just to allocate frame buffer. That would mean there'd be 1500 kilobytes (128MB - 122MB * 1/4) RAM left for textures. I'm willing to bet that mode isn't even available on the Voodoo5 6K.

About the V5-6000 at VE, true, for card with such a bandwidth at it's disposal, the high-resolution numbers were very dissapointing. Reason could be partially a driver issue, though, as Quantum's driver set is likely based on a very old reference driver build.

 

[Deeko]

I don't think it really is the same speed as the 5500. I mean, those driver tweaks added over 20 fps for me on the V5, and I don't know that they did any of them with those drivers, and so assume that it can get as much/more than the 5500 with those tweaks, it could easily surpass 100 fps in 1024x768x32.

EDIT: I somehow switched the word driver with registry....I dumb.

 

[legion88]

Some of you people are jumping on the drivers-bandwagon without thinking. It is like a knee-jerk response or something. Why don't you look at the Voodoo5 5500 scores posted in various benchmarks (like the one from Anandtech as an example)? Then double the score that the Voodoo5 5500 provides at 1600x1200x32. Why double? The 6000 has twice the fill rate and bandwidth. So double the scores. Twice means double. 2 == double. This should not be too hard to understand. You will find that when you double the scores, it comes almost exactly to what the VE's result at 1600x1200x32. Duh! What were you people expecting anyway?

Oh, obviously, they didn't use any "HSR" settings.

And other things to look at:

1) the 16-bit scores (non-FSAA) is limited by the CPU. That is why they are not blazing fast. Should this surprise anyone? The 5500 was already limited by the CPU on such a machine.

2) it appears that the reviewer had high-quality sound enabled. If so, then it will drop the score. On a 800Mhz machine, 100 FPS at CPU-limited conditions is not out of the question. But it is more than 10 FPS slower indicating that something is using up the CPU. Perhaps it is because the sound is on.

3) the 16-bit scores (FSAA) does not follow the pattern for 4X FSAA at 1280x1024 and 2X,4X FSAA at 1600x1200. It is unclear if this is because the drivers are bad or because having sound on screwed the results. Considering the resolution, I'm guessing the problem is because the card has 4 chips needing to be fed data (i.e. the same data is copied four times), the AGP is effectively operating at AGP 1X instead of AGP 4X. 32-bit scores are okay simply because that is bandwidth-limited--the card's memory is the bottleneck. In 16-bit, something else is the bottleneck. At such high fill rate, we begin to see the limitation of AGP 1X. Maybe. Anyone else has a guess as to why the 16-bit FSAA scores at high resolution is not following the expected pattern?

 

[merlocka]

In addition, he turned Dynamic Lighing off which means these numbers are inflated 5-10%

Thank goodness they didn't release this.

 

[Jethro Bodine]

Eman-
It seems you're the one who didn't read the article. I believe it said these were the only drivers 3dfx is releasing.

As far as "tweaking" goes, I'd expect a $600 video card to be fastest straight out of the box.

LOL

 

[EMAN]

Jethro, NO REALLY

It's still the first reference drivers from 3dfx and you shouldn't be quick to judge by it's performance now. Quantum is going to optimize it after they release the card.

 

[RoboTECH]

hey guys


MY BUST

the 6000 was benchmarked with TC disabled

I asked the author himself, and here was his emailed reply:

"Compression was set to 0 in the q3aconfig.cfg file. Unfortunately, we didn't have time to run benchmarks with compression enabled."

this would explain why it would do poorly in fillrate limited situations, where drivers don't make as much of a difference

it also explains why FS and a few other sources were looking at 65-70 fps @ 16x12x32 and the VE only had it in the high 50s

I know that with TC disabled on the 5500, I lose a good 20 fps at least.