X-bit labs

Started by: leexgx | Date 02/14/08

Comments: 23 | Last Comment:  04/15/08

[1-13]

1. games only have 3 levels of settings 128mb 256mb and 512mb (openGl based games i think you can pass an command that will make it use more then 512mb framebuffer)

like on my 8800 GTX on ET:QW it fully states i have 756mb when i alt tab back into the game but says somthing like 512mb is going to be used

so the games need optimizing to use more then there is mroe then 512 Real framebuffer
(also SLI is Not 512+512 = 1gb same goes for the 3870 X2 its an 512mb card not 1gb it can Only use 512mb framebuffer)

Umm, no. Games can and do use any arbitrary amount of vram. Most games have a number of configuration settings - model detail, texture detail, shadow detail, etc etc. Each setting for each of these has an mount of memory that it requires. The total amount of vram that the game will use depends on which one of the thousands ofways that it can be configured that you use. Not to mention that the resolution, AA settings, and double/triple buffering directly affect the amount that's needed for the framebuffer.

You can watch vram usage of a game using NV's perfkit, along with all sorts of other really interesting diagnostics.

2. It would have been nice for you to prove your statement:
"There is no speed gain from 1 gigabyte of graphics memory. The performance growth of the Bliss 8800 GT 1204MB GS over the reference card is due to the increased frequencies.".
You could have shown at least one benchmark with the reference 512MB 8800GT clocked at the same speed as the Gainward 8800GT. It seems a little wreckless to make statements/claims like that without a benchmark to prove it.

I think some of the benchmarks are self-evident that the increased frequencies could not have caused such a large jump in performance. Just look in Hellgate:London, 48% performance increase whereas the frequencies themselves were not increased by such a large amount.

The statement "There is no speed gain ..." was specifically for the Battlefield 2142 benchmarks. If you read the conclusion, they say that there are games that show significant improvement with the extra RAM.

In the case of BF2142, the speed gains are 0.63%, 3.5%, and 5.9% at 1280x1024, 1600x1200, and 1920x1440 respectively. The card clocks are running at 8.3%, 8.0%, and 5.6% higher in the GPU, shader, and memory domains respectively. While it would have been nice to see a comparison with identical speeds, I'd say that it's safe to say that the increases seen in BF2142 are not out of line with those expected simply from the increased operating frequencies.

The only affect that memory has in performance is when there is not enough of it, at which point performance falls off a cliff. There is no gradual performance degradation in out-of-memory conditions, nor any increase in performance to having more RAM than the game is using. This is clearly demonstrated here, where the 1GB card is either a few percent faster (due to slightly higher clocks), or hugely faster (due to there being "enough" RAM).

The best example is the Call of Juarez benchmark. at 1280x1024, the 1Gb card is ahead by the usual 4.6%. At the next step up, 1600x1200, this has increased to 111% faster. All due to an increase of probably (since I don't know if CoJ increases other buffers with the framebuffer) 20-40 MB of RAM usage. That's a classic example of the memory usage cliff.

3. Yet another old review... with old drivers and old cards.
With ATI Catalyst 7.12 and no X2, not that Catalyst 7.12 would be much use with the x2! It just shows how old this review is.

4. Why did you use only 2GB of RAM?
This helps 1GB card becouse 4 GB limit means that you couldn't use more than 3GB while systems cards with 512 MB can use a little more RAM.
Also maybe some games could use more than 2GB. Big flaw in article IMHO.

5. Another great review. I love that you use such a wide variety of great games, but I wish you'd consider adding a section on your subjective opinion, you just to let us know you did actually play games with it, and not just benchmark.

6. All I can do is LOL at the fact you guys didnt include the 3870x2. What a bunch of "NOOBS"

Yep, and I have been reading reviews of the factory overcolocked X2 on other sites.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=5615&Itemid=40&limit=1&limitstart=0

And the water cooled X2 looks cool.
http://www.nordichardware.com/news,7309.html

Maybe I should loan Xbit my crap old 6600GT or an even a more crap FX5600 if all they do is old reviews.

Maybe you should just fuck off back to the other sites if all you come here to do is bitch and moan about the reviews :P

Like every other site out there, xbitlabs doesn't re-benchmark every card for every review. Heck, there's a good chance they don't even have the other cards any more. The performance of the cards other than the card(s) being reviewed are historical numbers for approximate comparison. In any case, single-card performance has *slowed down* between the 7.12's and 8.2's (except for minor increases in CoH), so if anything the review is giving an advantage to ATI's cards.

Finally, if you could convince a manufacturer to give xbitlabs a 3870x2 (or even give them one out of your own pocket), I'm sure they'd be more than happy to review it.

Have to agree.

The irony is that as soon as I saw 'Here we go again''s comment I though 'Here we go again - another 12 year old who simply and mindlessly looked at the driver used without reading the whole article, researching the real world difference in performance between driver iterations or considers the time it takes to benchmark so many games over multiple settings before going to press.

'Here we go again' ... please piss of to whatever tangerine and lime filled fantasy site that conforms to your anal and immature fanboyism, that validates whatever video card you own, and that does its cowboy reviews in a few days.

Your comments ceased to be amusing a long time ago, and now only demonstrate how stupid you are.

7. "sound-over-HDMI"

HDMI supports much better sound formats than this 'Dolby Digital is good enough for anyone' approach.

Let's hope that these transitional acrobatics are no longer necessary within a year.

8. Any of you think the 8800GS is a decent card and worth picking up? All the other 8800 cards are a little over priced IMO.

9. RE Crysis: "Perhaps 1GB of graphics memory would be helpful if we tested the cards with enabled FSAA."

Well, I wish you had, so we could see even relative differences - playable or not.

Still, of course, we appreciate your work and what it does show us.

Thanks,
Arbie

10. Could somebody confirm my post #4 or say that I'm an idiot. ;) Becouse I kind of think that it(2GB RAM) is a big deal.

You are an idiot.

I think they used 2gb of ram because most people use 2gigs in their machines. Does that clarify anything?

It's unlikely that the extra 512 MB of RAM would help much. In fact, most games can't use more than 2GB of *address space*, let alone RAM, since they're only 32 bit.

The only benefit you'd see would be in loading times (as more files would be cached in RAM), which are irrelevant to the benchmarks presented.

OK. TNX. Still I would nver buy 1GB card. Waste of expencive RAM. :)

11. It's a shame that you didn't mention the HD3870 run AAA in SuperSampling mode, since the switch for MultiSampling is broken with newer drivers.

ATI cards running AAA at SuperSampling has no chance against NVidia using MultiSampling and therefore the HD3870 looks worse than it is in this review.

12. Ok, where's an 8800GTS with 1 GB of memory? There are quite a few manufacturers now offering the GT version with 1GB but for some reason nobody has dared to do it for a new GTS (now 2 months old). They started doing it for GT cards no later than 1 month after it was released (Palit was one of the first, IIRC). The guy who sells that new GTS with 1 GB will be the best seller, no doubt.. but why are they not doing it yet???

My theory is that Nvidia made it clear to them that they had no permission to do that with GTS G92 cards, since Nvidia has plans to release a *slightly* faster 9800GTX card based on a very similar G92 chip. 1GB of memory is pretty much guaranteed, perhaps with a 512-bit interface this time. And I think that DX10.1 support will also be incorporated, thus meriting a jump from the 8-series generation to the 9-series. Regarding 512-bit memory--Nvidia seems to be more interested in saving money by avoiding super-high-speed memory and using 512-bit instead (note: still no GDDR4 support predicted yet!). The 8800 Ultra used 2.2 GHz-rated GDDR3 memory, but only because Nvidia knew they could sell it at $649 a pop with that awesome feature: "100GB/s bandwidth at 768-bit interface".

Guys, those G92 cards are severely crippled, and Nvidia knows it. Only 16 ROP's, come on guys! Nvidia is actually delaying their next-gen cards because their G92 cards are selling like hot cakes. Of course I do appreciate the low prices, but that is only because of the 38xx cards from ATi. We need to start playing that "hard-to-get" game with Nvidia and they will start listening to us more. I mean, hey, it's not so hard to just slap on 1GB of 2.6GHz GDDR4 memory onto that 8800GTS G92 card.. if Nvidia would let Gainward or Evga do it themselves.

Nvidia just doesn't want to lose any thunder for its upcoming 9800GTX card that is supposed to be ONLY 30% faster than an 8800GTX (not the Ultra)!!! That's all??? Yeah, that's perfectly understandable, Nvidia.. with what you're doing right now.

13. I would like to see the testresults of 2 of these cards in SLI compared to the 9800x2
(maybe the GPU is this way not the limiter of the card(s) and can take more advantage of the extra memory)

[1-13]