Why connect the cards in SLI and Crossfire?

Karma Police said:
Any differences between bridgeless 7x.xx & the 8x.xx?


Thats a good question. I honestly dont know. I do know the 8x.xx drivers havent done well with SLI for me and introduced some earlier SLI driver bugs that were alleviated in the 7x.xx series. It was kind of discouraging actually. I do know the bridgeless performance with 7x.xx is about 15-30% slower give or take. In some cases ((3dmark2k1 SE Nature test)) you get no performance boost at all.
 
ChrisRay said:
Thats a good question. I honestly dont know. I do know the 8x.xx drivers havent done well with SLI for me and introduced some earlier SLI driver bugs that were alleviated in the 7x.xx series. It was kind of discouraging actually. I do know the bridgeless performance with 7x.xx is about 15-30% slower give or take. In some cases ((3dmark2k1 SE Nature test)) you get no performance boost at all.

hmmm. Well, back to the main topic. I wonder if DB was referring to CF, then.
 
I think this is a fiarly good reason why PCIe isn't used at the moment, the performance is just too low for high resolution rendering on high end boards. Although the theory suggests 2GB/s transfer rates, the reaility is the usable bandwidth with current implementations it is much lower and I think there is also latency issues thereas well.

That being said, I wouldn't be surupised to see a bit more of a two tiered system.
 
Dave Baumann said:
I think this is a fiarly good reason why PCIe isn't used at the moment, the performance is just too low for high resolution rendering on high end boards. Although the theory suggests 2GB/s transfer rates, the reaility is the usable bandwidth with current implementations it is much lower and I think there is also latency issues thereas well.

That being said, I wouldn't be surupised to see a bit more of a two tiered system.

Nice article as usual Dave. Scaling looks to be excellent. Super AA gets a big thumbs down though.
 
trinibwoy said:
Nice article as usual Dave. Scaling looks to be excellent. Super AA gets a big thumbs down though.
Agreed, great article. I was actually supprised by Super AA, not sure if anyone else was too.

epic
 
Dave Baumann said:
I think this is a fiarly good reason why PCIe isn't used at the moment, the performance is just too low for high resolution rendering on high end boards. Although the theory suggests 2GB/s transfer rates, the reaility is the usable bandwidth with current implementations it is much lower and I think there is also latency issues thereas well.
Good article, but I would rather have seen a comparison between the performance scaling of Crossfire and SLI than between Crossfire on different ATI hardware.
 
Chalnoth said:
Good article, but I would rather have seen a comparison between the performance scaling of Crossfire and SLI than between Crossfire on different ATI hardware.

Welll Dave never does comparisons so why would you expect differently. There are a lot of other reviews out there with comparisons to SLI though.
 
trinibwoy said:
Welll Dave never does comparisons so why would you expect differently. There are a lot of other reviews out there with comparisons to SLI though.
I haven't seen one that's sufficient yet, though.

HardOCP does their usual "real world performance" crap that is utterly useless at comparing the technologies.
Tomshardware doesn't list their dual-card scores along with single-card scores, and also benches the x850 XT Crossfire against the 7800 GTX SLI, a rather unequal pairing.

Have you found a decent one?
 
techreport++
Although apparently they're having problems with SLI aa so the scores are useless.
 
Chalnoth said:
I haven't seen one that's sufficient yet, though.

HardOCP does their usual "real world performance" crap that is utterly useless at comparing the technologies.
Tomshardware doesn't list their dual-card scores along with single-card scores, and also benches the x850 XT Crossfire against the 7800 GTX SLI, a rather unequal pairing.

Have you found a decent one?

Here are a couple of reviews:

http://www.beyond3d.com/forum/showpost.php?p=577747&postcount=46
 
radeonic2 said:
techreport++
Although apparently they're having problems with SLI aa so the scores are useless.
Ah, thanks, that was a decent read.

Looks to me that at this point in time, Crossfire really is pointless. Without a usable (only 60Hz) 1600x1200 resolution in Crossfire modes, the only good thing about it would be the new AA modes available, but those seem to be really low-performing.

As near as I can tell, then, Crossfire may become a good solution with the launch of ATI's R5xx hardware, but only if they fix the problems with resolution and/or Crossfire AA. The performance scaling is definitely good enough to show some potential, but without high resolution support or good AA performance, what's the point?
 
Chalnoth said:
As near as I can tell, then, Crossfire may become a good solution with the launch of ATI's R5xx hardware, but only if they fix the problems with resolution and/or Crossfire AA. The performance scaling is definitely good enough to show some potential, but without high resolution support or good AA performance, what's the point?

Well, to be clear, AA performance is good as long as you stick to the "standard" 6X modes and lower. It is only specifically Super AA modes where performance is currently lacking.

But overall I agree...x-fire on R4xx is a tough sell.
 
Joe DeFuria said:
Well, to be clear, AA performance is good as long as you stick to the "standard" 6X modes and lower. It is only specifically Super AA modes where performance is currently lacking.
Right, that's what I mean: the selling point of Crossfire is supposed to be higher resolution or better AA. But the better AA comes at a big performance hit, and the resolution is limited to 1280x1024 is you want at least 85Hz.
 
Well, I just see this Crossfire version as one step towards R5xx multi-card setup. I really think a R5xx CF will be different than a X850XT setup. I'm not talking performance differences, but implimentation differences.
 
Really? Expectations seem to be for the same external, dongle solution, albeit with no real res and refresh limitations, owing to dual-link TDMS. That just leaves Super AA, which will surely be corrected in the next gen XF implementation, too.
 
I expect the external dongle to still be there, but not as much data will be passed through it. Instead I think ATi had enough time with the R5xx to have some of the info that was passed through the dongle with the R4xx to be passed more efficiently through the PCIe ports with the R5xx.

I know DB said the PCIe part isn't working very well in his article, but with ATi issuing CF for the R4xx as a "knee-jerk" reaction, I believe they had enough time to change the way it's implemented on the R5xx. You follow what I'm thinking?
 
Are we expecting R5xx master Crossfire samples for reviewers at the same time as the underlying R5xx cards? Any sense if some will be available at the "Tech Day" whenever that is?
 
Karma Police said:
Well, I just see this Crossfire version as one step towards R5xx multi-card setup. I really think a R5xx CF will be different than a X850XT setup. I'm not talking performance differences, but implimentation differences.
One would hope.

Karma Police said:
I expect the external dongle to still be there, but not as much data will be passed through it. Instead I think ATi had enough time with the R5xx to have some of the info that was passed through the dongle with the R4xx to be passed more efficiently through the PCIe ports with the R5xx.
Er, the external dongle is more efficient, performance-wise, than using the PCIe bus. According to Dave's writeup, passing info over the PCIe bus is the primary reason for the massive performance hit with SuperAA.

The reason why passing over the PCIe bus is bad is simply this: with most current dual-PEG motherboards running with 2 8x ports, bandwidth is already significantly lower for each card in SLI/Crossfire than it is for one card alone. Passing more data across that bus only decreases the available bandwidth, and serves to reduce overall performance.
 
Back
Top