Sapphire Hybrid Radeon X800 XL 512MB Review

[Dave Baumann]

http://www.beyond3d.com/siteimages/b3dsmall.gif (http://www.beyond3d.com/reviews/sapphire/512/)If there has been a constant in the PC graphics market since it began its that graphics performance has increased, and with it so too have RAM sizes. Whilst we may still be fresh from asking "Do we need 256MB on our graphics boards?", the hardware vendors themselves are forging ahead, apparently spurred by developer requests, with graphics boards with 512MB of local RAM.

Today we are taking a look at one such 512MB board configuration ATI have made available to their board partners with a review of Sapphire's Hybrid Radeon X800 XL 512MB. We'll compare it against its 256MB sibling as well as the higher performance X850 XT Platinum Edition under a series of tests with high memory utilisation to see how much of a difference, if any, 512MB can make today. We also ask a few developers on their perspective of the inclusion of 512MB on a graphics board. Click here to find out more (http://www.beyond3d.com/reviews/sapphire/512/).

[tEd]

The graphiccard from my first PC had 1MB ram. Doom rocked on it.

:roll:

[Pete]

$150 extra for the RAM? Cripes. Still strange to see only HL2 showed any use for it, while D3's Ultra mode again seemed to fit comfortably within 256MB. Is it possible both ATI and nV are compressing D3's normal maps irregardless of what the app says (to win benchmarks)?

Still, surprising to see a 512MB XL at all. Thanks for the article, though, Dave.

[Dave Baumann]

As I mentioned in the review, there is a slight performance drop between Normal and Ultra modes - this is not enough of a drop to suggest that there is massive texture swapping/addressing occuring from system RAM, but could be consistent with a few more uncompressed textures/normals.

[phenix]

Extremetech and Driverheaven reviews demonstrate an increase in doom3 performance. Driver heaven claims a significant increase in Riddick performance too.

[Neeyik]

I find Extremtech's D3 result to be highly questionable - 54% increase in frame rate? Just what timedemo/level are they testing it with? Driverheaven report a frame rate increase of 29% - again, what demo are they running?

[Geo]

I found Sweeney's comments the most thoughtful, and actually pointing at a way forward on compression.

"Texture compression is a huge area ripe for IHV innovation, and I've been quite surprised at the lack of significant innovation in that area. The existing DXT formats come very close to maxing out the quality/size trade-off possible with ultra-low-computation-overhead 4x4 block compression schemes. Further compression will rely on more sophisticated random-access schemes operating on larger blocks. If you look at the JPEG 2000 results, they have achieved perhaps 10X better quality/size than DXT1-5. Hardware JPEG 2000 is probably not practical, but one could certainly achieve far better results than DXT by moving towards a scheme somewhere in between in complexity.

Anyone care to comment on his suggested way forward?

Also, any hints of 8x or greater coming in R520 Catalyst releases? I'm thinking if they are planning to offer a higher-mode than 6x, and make it available in the drivers to the X800 line as well, it makes more sense why we'd be seeing 512mb X8xx cards. . .

[Pete]

Does the X series even have enough sample positions to support 8x?

Thinking ahead to hardware that might offer 8xMSAA, would allowing for 4x per pass--vs the 2x per pass of current h/w--mean faster 4xAA in addition to the possibility of 8x and even 12xAA (LOL @ framebuffer req's)? How much faster would 4x per pass h/w be than current h/w with 4xAA (assuming same bandwidth, etc.)?

As for new compression, it was discussed in a very recent 3D Tech thread. Maybe someone'll link it before I rediscover it....

[Veridian3]

I find Extremtech's D3 result to be highly questionable - 54% increase in frame rate? Just what timedemo/level are they testing it with? Driverheaven report a frame rate increase of 29% - again, what demo are they running?

Our own custom one rather than the inbuilt demo. I spend quite a bit of time with each of our review suite games before we add them locating the most demanding sections of gameplay then use those for testing.

[Neeyik]

Could you PM any details of it? All such material and data will remain strictly confidential, of course, but I would like to see how your demo is so different to our own!

[Geo]

Thinking ahead to hardware that might offer 8xMSAA, would allowing for 4x per pass--vs the 2x per pass of current h/w--mean faster 4xAA in addition to the possibility of 8x and even 12xAA (LOL @ framebuffer req's)? How much faster would 4x per pass h/w be than current h/w with 4xAA (assuming same bandwidth, etc.)?



I played with that idea in the late R520 thread. Chalnoth pooh-poohed, but Ailuros seemed interested. At 12x you have the possibility of 24x Temporal. . .that might be quite interesting to see quality-wise.

[Dave Baumann]

Does the X series even have enough sample positions to support 8x?

The sample patterns are very sparse at these low levels. The R300-R420 series support a 12x12 sample grid, so thats a max of 144 sample positions (although, obviously will all of them populated it would end up as ordered grid). 8x is already supported, of course, through "temporal" AA, as is 12x - I believe the R300 SGI and E&S systems go up to 24 samples.

Its not really the number of sampling the positions, here, though, but how much the chip can buffer and block compress.

[Pete]

Yeah, given how well even current cards perform with available games, I wonder if temporal is the answer until Longhorn hits. Going higher than 6x sparse may not even be ideal if ATI (and nV) managed to get AA working with FP blending, though given the identical bus widths, I doubt this (in my naive little way).

[dizietsma]

Dave,

I have just read DriverHeavens review of the X800XL 512MB card

http://www.driverheaven.net/showthread.php?t=74089

and the results seem to be at odds with your own findings, especially in relation to Doom3.

The conclusion ( http://www.driverheaven.net/reviews/x800_512/conclusion.htm ) is equally different, yours seems to be a lot more conservative.

Any thoughts ?

[Veridian3]

Dave,

I have just read DriverHeavens review of the X800XL 512MB card

http://www.driverheaven.net/showthread.php?t=74089

and the results seem to be at odds with your own findings, especially in relation to Doom3.

The conclusion ( http://www.driverheaven.net/reviews/x800_512/conclusion.htm ) is equally different, yours seems to be a lot more conservative.

Any thoughts ?

I cant speak for Dave, but i'll give some of my thinking on this from a personal and DH pov.

Re: Doom3
There are several factors which may affect the results obtained... test system config, section of game used, characters on screen and so on. As most sites will be using custom demos there is likey to be confilicting results. No one set is wrong, they just give snapshots of the game performance in the tested section.

On the conclusion, having read Daves i think there are a few things which result in the final opinions differing. The most minor of these are the fan noise which we didnt have issues with (ours was a different fan design) and the overclocking potential (again ours was a better sample in that regard). These only have minor impact on the overal card conclusion however a better sounding and overclocking card will obviously give a slightly better feel than a lesser performing card.

Then of course you have the performance differences between the reviews, because of the different test suites Dave and I find different performance increases (or lack off them) and so that colours the overall opinion of the card you are testing. Again, as i mentioned in the Doom3 comment above, neither opinion/result is wrong they are just different. Also one of the more major things that affected the conclusions also is the pricing, i personally wasnt as concerned about the pricing of the card because the pricing puts the 512mb XL in direct competition with the 256mb 6800GT. Based on the performance of the XL being better than the GT in the majority of the engines we tested and the fact that the card may well be slightly more futureproof due to the extra 256mb i feel the price is spot on. Dave of course didnt see the benefits as much in performance so the extra cash doesnt seem as worth it.

In the end every review is one persons report on their experiences with the product in question. Which is why i'd always recommend reading the results from a few and seeking out the ones with the games your most likely to play before making a decision on buying a product.

[Jawed]

Presumably HL-2 is rendering reflections via a texture, and that would explain the heavy "texturing" load.

Dave, I suppose you could test this with the 512MB card by turning off Reflect All, setting it to Reflect World or Reflect None. Probably only the latter would make a meaningful difference. Just curious really.

Would the shadowing in Chronicles of Riddick also go via a texture?

Jawed

[Mendel]

Veridian.


POP:Warrior Within was one of the few games we tested that the Geforce 6800 GT was faster than the XL 256mb card.


To me it seems 6800 GT is faster than XL 256 card in 4 tests out of 7. Why would you want to put such a negative spin on gt?

[Veridian3]

Veridian.


POP:Warrior Within was one of the few games we tested that the Geforce 6800 GT was faster than the XL 256mb card.


To me it seems 6800 GT is faster than XL 256 card in 4 tests out of 7. Why would you want to put such a negative spin on gt?

Because i test more games than i include in the article and the majority were faster on the XL.

[ondaedg]

Veridian.


POP:Warrior Within was one of the few games we tested that the Geforce 6800 GT was faster than the XL 256mb card.


To me it seems 6800 GT is faster than XL 256 card in 4 tests out of 7. Why would you want to put such a negative spin on gt?

Because i test more games than i include in the article and the majority were faster on the XL.

:roll: :?: :roll:

[Veridian3]

:roll: :?: :roll:

Wow, that was a constructive post.

The other games included...Farcry, nfsu2, hl2, tiger woods 2005, Chaos Theory, UT 2004 and a few others. None of which told us anything that the tests included in the review didnt so no need to include them...there is such a thing as too much information.

[ondaedg]

Well, what "journalist" would tell people that the XL is more futureproof when their own article shows that it is slower in the majority of the benchmarks THEY provided, then clarify by saying "we didn't show all of the tests"?

Why don't you post the benchmark scores for those "other" tests so that we can see your findings? I for one am interested in the benefit of 512mb cards over 256mb cards for games that are out right now.

[Dave Baumann]

Just a quick note that I've made a couple of updates to the first and last pages with comments from Sapphire about the fan, availability and price (evidently they are doing it for €399) and also included the calculation of the frame buffer size with FSAA on. We're still looking into the Doom 3 performance with the Ultra mode enabled. [Edit] There's also some rationale for why using 6x FSAA with 16x AF on the test setup page added.

[Jawed]

Just a quick note that I've made a couple of updates to the first and last pages with comments from Sapphire about the fan, availability and price (evidently they are doing it for €399) and also included the calculation of the frame buffer size with FSAA on. We're still looking into the Doom 3 performance with the Ultra mode enabled.
Can you try HL-2 with Reflect World and Reflect None to see if that impacts the "texture memory footprint".

Jawed

[trinibwoy]

Well, what "journalist" would tell people that the XL is more futureproof when their own article shows that it is slower in the majority of the benchmarks THEY provided, then clarify by saying "we didn't show all of the tests"?

Don't get your panties in a knot over one article. Many sites have (p)reviewed the card and have given their opinions. Just read all of them and come to your own conclusion.

[Mulciber]

Well, what "journalist" would tell people that the XL is more futureproof when their own article shows that it is slower in the majority of the benchmarks THEY provided, then clarify by saying "we didn't show all of the tests"?

Don't get your panties in a knot over one article. Many sites have (p)reviewed the card and have given their opinions. Just read all of them and come to your own conclusion.

he was refering to statements made in this thread, not the article itself

[Dave Baumann]

Jawed - there is a matter of about a single FPS difference between reflect all being on or off. I think the memory requirements are down to the cube map shadowing, and there is going to be the same amount present for reflect all and reflect world, the onlt difference is that there is more processing going onto that map when reflect all is used.

[ondaedg]

Well, what "journalist" would tell people that the XL is more futureproof when their own article shows that it is slower in the majority of the benchmarks THEY provided, then clarify by saying "we didn't show all of the tests"?

Don't get your panties in a knot over one article. Many sites have (p)reviewed the card and have given their opinions. Just read all of them and come to your own conclusion.

Ya. He didn't like my emoticons in an earlier post. :wink:
I don't actually get mad about articles. He just made some suspect statements that as a "journalist" should not have been made without supporting it with objective testing and benchmarks.

[ondaedg]

Just a quick note that I've made a couple of updates to the first and last pages with comments from Sapphire about the fan, availability and price (evidently they are doing it for €399) and also included the calculation of the frame buffer size with FSAA on. We're still looking into the Doom 3 performance with the Ultra mode enabled. [Edit] There's also some rationale for why using 6x FSAA with 16x AF on the test setup page added.

Dave, was there any instance other than the benchmarks that you posted that having 512 mb showed more than an 11% increase in framerate? Do you think the 512mb 6800 Ultra would show similar results?

[Jawed]

Jawed - there is a matter of about a single FPS difference between reflect all being on or off. I think the memory requirements are down to the cube map shadowing, and there is going to be the same amount present for reflect all and reflect world, the onlt difference is that there is more processing going onto that map when reflect all is used.

Hmm, roughly 1fps difference between Reflect All and Reflect None? Blimey the benchmark you're using is doing some serious other stuff! :D

It would be interesting if is was possible to isolate the memory consumed by each of the eye candy features.

I haven't got my 9800Pro any more (fan died) so I can't play about...

Jawed

[Dave Baumann]

ondaedg, what I've tested and seen is whats in the article.

[Mendel]

Because i test more games than i include in the article and the majority were faster on the XL.

Oh okay, thanks for clearing that up!

What I would find interesting, would be some general "gaming performance score"

It could be calculated from the average of the framerates across tested games. That way you could include some data from the not so popular benchmarks, and still not have way too much information. And we end users would get a nice overall view of the general speed of different cards.

- Just an idea!

[Sunday]

http://www.hardinfo.dk/show.asp?page=5652

[cho]

i have a question, why include the z-buffer into the front buffer when calculating the pixel frame buffer requirements ?

[Dave Baumann]

Because, surprisingly its still there - at least many apps specify it to be under DX.