Slashdot Mirror
Comprehensive Video Benchmarks
Crusader writes: "Matt Matthews has produced an extensive series of benchmarks which examine four separate games' performance on the Voodoo5, the Rage 128 Pro, the G400 Max, and the GeForce 2 3D graphics cards under Linux. Performance against Windows 98 is also included." We also received: driveitlikeyoustoleit writes, "3dfx, NVIDIA and ATI's best are all pitted against each other in a high-end 3D video card
roundup. The authors pit six GeForce2 GTS (from ASUS, Creative,
ELSA and Hercules) based cards against an ATI Radeon 256 and a 3dfx Voodoo5
5500. Performance for a change isn't the only criteria in question (although the end scores are somewhat weighted in favor of fps) but also at full-scene
anti-aliasing, image quality and DVD performance/quality are critically looked at. The screen shots page showing off FSAA comparisons are great visual indicators of what the cards can do."
For as long as I can remember, ATI's chips have performed far better in 32-bit color depth tests than 16-bit color-depth tests. Yet, Sharky doesn't seem to show any charts comparing the cards in 32-bit except for the Re-Volt benchmark which they admit is outdated. However, they do state on page 6 of their review that the GeForce-2 cards rule both 16-bit and 32-bit.
Did I miss something?
Anyways, it seems that the Radeon is only a few FPS behind the GeForce-2 cards, and I imagine that difference is humanly imperceivable except for super-humans. Meanwhile, you gain better DVD playback and other huge multimedia offerings, especially if you look into the always-a-pleasuer All-in-Wonder line from ATI.
So, why did Sharky need to use so many pages to get these points across?
For the same resolution, run the games in a window, instead of going full-screen for Windows.
As far as the display driver is concerned, doing a page flip on a non-fullscreen surface is impossible. That should eliminate the difference produced by the blitting drivers for Linux.
Besides, issues like the efficient use of AGP bus, DMA, eliminating bad polygons when you do clipping and synchronization all can have a great impact for performance - blitting vs flipping is just a minor issue on all but the highest resolutions.
Heard anything about VIA chipsets? (Specifically more recent ones, I have a KT133 Athlon mobo)
On the whole, I've never had problems with VIA chipsets and AGP, this is the first unstable driver I've ever had, and I've been using VIA-based mobos (Epox Super7 for a K6-2 and now an Athlon KT133-based board.) for 3 years.
retrorocket.o not found, launch anyway?
One thing that doesn't seem to get addressed by ANY Linux 3D hardware review I've seen (One was also in LJ recently), is the issue of stability.
We run Linux because it's stable, even in cases where we know that we lose performance because of that stability.
Well, all of these reviews basically say, "If you can afford it, buy NVIDIA! They're fast!".
So what if your card is fast but your machine crashes in the middle of a game? NVIDIA's drivers just plain suck in the stability arena - My machine crashes on a regular basis in both Q3 and UT with those crap drivers. Even Windows looks rock-solid compared to Linux when using those accursed drivers! And all this because the only difference between the Quadro and GeForce is a PCI/AGP device ID and a few flags set in the driver because of said ID - it's the only plausible reason for being the only vendor not to release source/specs. (Check out http://www.geocities.com/tnaw_xtennis/)
retrorocket.o not found, launch anyway?
All of these cards are blazing. As such, driver stability is much more of an issue. Matrox is pretty top notch in this department. NVidia has had it's share of problems in the past, but may be improving. I would much, much prefer to get a rock solid card and driver combination than some crazy card that goes for extra performance at all costs.
The difference between Linux and Windows at 1280x1024 on the GF2 is the difference between playable framerates and unplayable framerates. 'nuff said.
Actually, I think I need to cogitate more. The NVIDIA Linux drivers are pulled from more or less the same source as the Windows Linux drivers. This means code-wise, the two are fairly even in terms of performance tweeks. Now the Linux drivers will gain a little bit as NVIDIA tweeks glue between the drivers and OS, but I doubt the speed will ever reach that of Windows. Now page-flipping could be one of the things holding back the Linux drivers, but note that even at the lower-res tests, Linux is still behind. All this ignores one point: Shouldn't Linux be undeniable FASTER? Why use the OS of it is only just ALMOST as fast as Windows. Windows is a piece of junk. Why can't Linux beat it?
A deep unwavering belief is a sure sign you're missing something...
Well this review is really is going to be a factor in the purchase of my next card when it admits that it has been "weighted" in favour of certain types of applications. Is it really too much to expect to be given reviews in terms of raw perforamce? Considering the amount of journalistic integrity seen so far on the net, probably.
Then again, this isn't really suprising from a website that somehow manages to fit more banner ads onto each page (and there are for some reason, a lot of pages) than there is actual content. With the amount of corporate whoring they're managing to acheive in their page layouts, is it any wonder that their reviews feature skewed statistics which practically invalidate their purpose? It also makes you wonder where else corporate $$$ comes into the equation in these kinds of reviews.
I'd much rather that we saw more reviews from sources that don't appear to be pandering for cash from commercial sources. Whenever you see a banner ad, you can't trust the information you're being given. Hmm, now what site does that bring to mind?
Do you call crashing routinely within an hour (sometimes within 5 minutes) of starting a game good performance?
I don't. NVidia's drivers are unstable and crash my system all the time.
retrorocket.o not found, launch anyway?
Whenever I have needed help I have found the nvidia people that sit in that channel very helpful.
Boffoonery - downloadable Comedy Benefit for Bletchley Park
Trailing behind the POS that is Windows98 by more than 10% (at the higher res tests, more like 25%) is DEFINATELY bad! I gets even worse. This disparity gets bigger when you note that QuakeIII frame-rates in NT4 are even faster (by about 10-15%) than in Win98.
A deep unwavering belief is a sure sign you're missing something...
All this from the same crew who benchmarks Q3 at 1600x1200, and spits on any card that loses that race (how many ppl have monitors that can do 1600x1200 at 100 hz anyways?)
They rated the Elsa card as SuperFantasticGetOneOrDie, yet the identical Powergene card was rated as "bleh," for those "those on an extremely tight budget" (the powergene is only $10 less than the elsa.) But according to the reviewer "you don't get a name brand" with the powergene, so stay away unless you are ghetto. Reality check anyone? BOTH cards are stock reference designs, except for a possible future tv in/out module for the elsa.
Also, by reading these "shootouts" one would get the impression that quake 3 is the only game on the market. If they benchmark some other game, it has to be a quake clone. I play the Quake series to death, but I also play strategy games like Homeworld (which can bring a video card/cpu to its KNEES during intense battles.) Where is the benchmark on some non-FPS game?
How about image quality? I personally turn on FSAA on my Geforce when playing Homeworld at 800x600, because it looks SO much better than 1024/768 without FSAA. If sharky reviewers would play something besides FPSs then stuff like image quality would get ranked way higher.
Anyways, thats the end of my rant. Whenever you read one of these reviews, keep in mind the biases of the reviewer, and remember that they sometimes get caught up in "reviewerland," which is not necessarily connected to the "real world."
:wq
X is not a slow as people seem to think -- witness the good showing of the NVIDIA Linux drivers against the Windows drivers. There is no magic bullet that will give you substantially better performance under Berlin than you have in X.