PCI Express & CrossFire: Scaling Explored

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
[citation][nom]avatar_raq[/nom]I think you are talking about something ASUS calls "crosslinx" claiming it can enable crossfire in 8x 8x mode..Well I'm not sure it can reach for the same results of P45..Perhaps it's somewhere in between other P35 mobos and P45...Can't you check the gaming performance gain after adding a second GFX?? (barrowing it from a friend for example!)..Then compare it to the results above and tell us!!![/citation]

Actually the bandwidth would be in line with the 975X, which is also x8x8 in 1.1. But the PCIe hub can also be a performance-reducing factor as it adds latency.
 
[citation][nom]Proximon[/nom]I think you totally nailed this. You overclocked the CPU to avoid as much bottleneck as possible. I would only add that, in other comparisons elsewhere, x38 tended to fall right in the middle of x48 and P45.I have wanted this for a few months, since drivers and BIOSes matured, but figured it wasn't going to happen once i7 hit.[/citation]

The only reason X38 would fall between X48 and P45 is because the manufacturer neglected that board: X38 and X48 are the same chipset.
 
[citation][nom]Crashman[/nom]The only reason X38 would fall between X48 and P45 is because the manufacturer neglected that board: X38 and X48 are the same chipset.[/citation]
Is there anyway to reverse that neglection? and gaining performance simslar to x48? I'm thinking of a new build and having x38 performing identical to x48 with lower price and commoner DDR2 support will be a great bang for the buck..
 
[citation][nom]avatar_raq[/nom]Is there anyway to reverse that neglection? and gaining performance simslar to x48? I'm thinking of a new build and having x38 performing identical to x48 with lower price and commoner DDR2 support will be a great bang for the buck..[/citation]

Some X38 and X48 motherboards are identical, so the way to "fix" a "slower" X38 would be to flash the BIOS from the X48 version.

Some other X48 motherboards are second-generation designs, which means they can have design advantages over first-generation versions.
 
[citation][nom]Crashman[/nom]Some other X48 motherboards are second-generation designs, which means they can have design advantages over first-generation versions.[/citation]
How can I tell if certain x48 mobo is 1st or 2nd generation?And what did you mean by "Some X38 and X48 motherboards are identical"? Identical by the number of slots? In other words what are the things that we need to checkout to say these 2 mobos are similar and the bios can be flashed safely?
 
[citation][nom]avatar_raq[/nom]How can I tell if certain x48 mobo is 1st or 2nd generation?And what did you mean by "Some X38 and X48 motherboards are identical"? Identical by the number of slots? In other words what are the things that we need to checkout to say these 2 mobos are similar and the bios can be flashed safely?[/citation]

Everything right down to the PCB revision. Gigabyte X38-DQ6 and X48-DQ6 were the same board so long as you got the same PCB, and they even have different webpages for the new PCB. Same with Gigabyte X38T-DQ6 and X48T-DQ6.

I also have an early P5E3 Premium (X48) with the same PCB as the P5E3 Deluxe (X38), but it might just be a pre-production sample.
 
So, what we can see is: x4 && X8 PCIe 1.1 is to tight for 4870.
Well, but we have no answer for 4670, 38xx and maybe some lesser performing video cards in CrossFire, where the smaller bandwith on P965 or P35 may not be a problem.
I think 4870 is too pricey for an upgrade on such motherboard, and I would look on the cheaper cards (or wait till 4870 gets cheaper, as well as the motherboards 😛 ).
I'm still expecting gains on P35 or P965 using two slower cards in CrossFire. But it's only a guess - haven't seen any benchmark.
 
I know this is an article about Crossfire speed but I think a lot of people are missing a big point of Crossfire. You get double the AA, AAA and AF if you choose to enable it. I am dead serious about a perfect picture and I crank EVERYTHING on full with my 4870/512 CF setup and get an absolutely beautiful image running an E8400 at 4 Ghz on x48. Grid is at 100 fps at 1900 x 1080. At this setup, it is very close to photorealistic. Throw that on a Sony XBR4 and there is nowhere else to go! Absolutely stunning!
 
Well done Crash, nice investigation. I was surprised that there was such an impact of Crossfire on Supreme Commander even at low resolution, it's a game we'd typically associate as system/CPU bound than GPU. Also interesting to see the variability in the World in Conflict with the P35 and P965 falling of the scale even at high res.

Thanks for the work, and nice to see you contributing more too.
 
[citation][nom]randomizer[/nom]I think 1024x768 can be dropped from benchmarks with high-end cards, but not midrange and low-end. 1280x1024 still has plenty of users with 19" LCD panels.[/citation]

The only reason I would want to see it there and for it to especially be in Xfire/SLi reviews is because the low resolution test does a pretty good job usually of showing Xfire and SLi overhead. It's not exposed so much here (a bit vs single card) as that's not the focus/goal, but if you look at many multi-GPU reviews, you'll see this magically tiping point that the overhead involved in making the multi-card solution work, impact greatly on the performance, compared to a solid single GPU solution, where a single GF9800GTX will outperform a GTX260 SLi or HD4870CF, etc. Because while bandwidth might not be the issue there, the CPU resources can be taxed. It's usually not a major issue because often we're talking about 150fps vs 120 or something, but it is a quite noticeable thing when it happens, so it's just good to know which games display that more often.
It's tough to find the right balance, but I like Crash's choices because they cover a system-centric res, a gpu-centric res, and then a typical balance. Kinda let's you pick and chose your information.

Of course for me, I always want more information so I'm definitely not truly an 'unbiased observer' in that respect.
 
cool article some suprising results - know i know why my fav mobo is the p5w-dh!

still got mine!

why no x58 this is incomplete - it could have been an 11 on 10 scale, know have to guess where x58 fits in.

8/10
 
Considering i'm trying to make a laptop purchase deicsion - as in to get Crossfire or NOT get crossfire - i don't feel like GUESSING if the performance will really be worth the extra $300. I can always GUESS on my own that it will be money well spent, but i was looking for numbers to help base that decision on. And since the laptop has a 17" panel and uses 1920x1200, I was really hoping this article would have it. Considering 19x12 is a pretty popular resolution, not including it is a lot like taking a crap and missing the toilet.
 
[citation][nom]dragonsprayer[/nom]cool article some suprising results - know i know why my fav mobo is the p5w-dh!still got mine!why no x58 this is incomplete - it could have been an 11 on 10 scale, know have to guess where x58 fits in.8/10[/citation]

This was an appels-to-apples comparison, an X58 wouldn't hold a Core 2 Duo or work with DDR2. That misses the point of the article anyway, which was to answer the question "If I have a decent card, will adding a second one give me the performance boost I want".
 
[citation][nom]dragonsprayer[/nom]cool article some suprising results - know i know why my fav mobo is the p5w-dh!still got mine!why no x58 this is incomplete - it could have been an 11 on 10 scale, know have to guess where x58 fits in.8/10[/citation]
It's not that hard to guess since you'd be using Core i7.

[citation][nom]Crashman[/nom]Overclocked memory controller. You didn't read the Synthetic benchmark page?[/citation]
People never read about synthetics :lol:
 
[citation][nom]Miner24[/nom]Hmm, how come tweaktown got quite different results?http://www.tweaktown.com/articles/ [...] index.html[/citation]

Those look like the results I've seen using the benchmark_gpu bench, rather than the benchmark_cpu bench. The benchmark_gpu is a flyby and I've noticed rendering errors on multiple configurations, so I stick with the benchmark_cpu which has a more realistic in-game appearence.
 
[citation][nom]Crashman[/nom]Those look like the results I've seen using the benchmark_gpu bench, rather than the benchmark_cpu bench. The benchmark_gpu is a flyby and I've noticed rendering errors on multiple configurations, so I stick with the benchmark_cpu which has a more realistic in-game appearence.[/citation]
My X1950 pro would render a few distant trees blue in the benchmark_gpu bench.
 
I see I am late to see this article but dropping 1024x768 would be a mistake. Ok lets put it in another way maybe have a set of resolutions for the high end cards like 1680x1050 1920x1200 2560x1600 but only for the high end cards. But when testing the lower end cards have a different set of resolutions 1024x768 1280x1024 1680x1050 this makes more sense to do it this way. I have seen so many sites test low end to low mid range cards at resolutions like 1920x1200 & when the cards fail to perform they say how bad these cards are & are a waste of money because they can't play the games tests at a good enough frame rate. Well of coarse they won't if a card costs $65 it is not gonna game at 2650x1600 thats just common sense but if these same cards were tested at the lower resolutions that they can play at they would work just fine.

Oh & yes a pretty good article lots of info to be found.

I myself play at 1080P most of the time as that is what my HDTV works best at but from time to time in the old games 1024x768 is the only option so yes I guess I still use that resolution every once & a while & I own a radeon 4870x2 2GB OC'ed to the nutz.
 
I agree with rocky1234. Low and middle and high end range cards need to have different testing specks. It's goot to have one or two resolutions that are same, but in most cases these cards have different uses.
That one "common" resolution allows direct comparison with these cards.
Low: 1024x768 1280x1024 1680x1050
Middle: 1280x1024 1680x1050 1920x1200
High: 1680x1050 1920x1200 2560x1600

Or something like that. I expect that wide screen resolutions would even be better alternative guite soon, even to low end cards.
It's a guite good way if you want to test 3 resolutions at time. Then there are these "problem cards" like 4850/4870 that was mean to be middle range cards, but still managet to compete with high end cards when released. When something like this happens, the testers have to be awake!

Widescreen 16:10 version
low: 1280x800, 1440x900, 1680x1050
middle: 1440x900, 1680x1050, 1920x1200
high: 1680x1050, 1920x1200, 2560x1600
super high end: 1920x1200, 2560x1600, 3840x2400

widescreen 16:9 version
low: 1280x720, 1366x768, 1920x1080
middle: 1366x768, 1920x1080
High: 1920x1080, (3840x2160), 4096x2160
Ultra high: (3840x2160), 4096x2160, 7680x4320

clossary:
HDTV 720p: 1280x720
HDTV 1080p: 1920x1080
UHDV 4320p: 7680x4320
3840P: 3840x2160 (this in not standard so far)
(Digital Cinema 4K: 4096x2160)
RED Digital Cinema 2540p: 4520x2540
 
I run a P45 single slot with a E8500@3.9ghz OC with a GTX260 OC and run all games at highest settings @ 1680x1050 8xAA even Crysis @46FPS average and FarCry2@120FPS average 100% NO STUTTER! and get 17,800+ in 3Dmark06 with windowsXP pro SP3 X86 DX9 so I agree that unless you are going to run extreme high resolutions it is not worth going SLI/CrossFire.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom