NF200 "True" 3-Way SLI Preliminary Results

[enaberif]

[H] Enthusiast - NF200 "True" 3-Way SLI Preliminary Results

[SKYMTL]

"True"? What the heck is that supposed to mean? I believe some other sites have already posted NF200 SLI results.

[enaberif]

Originally Posted by SKYMTL

"True"? What the heck is that supposed to mean? I believe some other sites have already posted NF200 SLI results.

Perhaps I'm behind the times, just the first I saw.

And by "true" i'm assuming not using the NF200 chipset.

[ImmaPC]

Anyway, True means ACTUAL 3 x16 PCI-E lanes, instead of x16-x8-x8 or x16-x16-x8.

I was looking at the graph they have under the video while it loaded and it seems that the NF200 chip is actually beaten by the not-so-"true" EVGA X58 SLI. It's strange this would occur, perhaps it's a scaling issue, as in the NF200 doesn't know what to do with the extra bandwidth. Would anyone know why this decreased performance is so?

[enaberif]

Originally Posted by ImmaPC

Anyway, True means ACTUAL 3 x16 PCI-E lanes, instead of x16-x8-x8 or x16-x16-x8.

I was looking at the graph they have under the video while it loaded and it seems that the NF200 chip is actually beaten by the not-so-"true" EVGA X58 SLI. It's strange this would occur, perhaps it's a scaling issue, as in the NF200 doesn't know what to do with the extra bandwidth. Would anyone know why this decreased performance is so?

Because this is how the NF200 supposedly works.
Make sense now?

[MpG]

Kyle (in video) was theorizing that the NF200 introduces a small amount of latency into the process, accounting for the slight drop in frames, which is pretty reasonable imo. And really, both Nvidia and ATI probably understand the limitations of the PCI-E bus (1.1 or 2.0) on their cards better than anyone, and have coded accordingly, to ensure that data transfers don't saturate the bus. Thus, even if the NF200 actually worked the way marketing would have us believe, the benefits would likely still be negligible.

[ImmaPC]

XD, thank you for the clarification, enaberif. That design is somewhat foolish; they should have the NF200 hooked up directly to the 3 PCI-E x16 slots, THEN to the X58. I dunno if it would make a huge difference, but it would definitely be less latency-ridden than the current design.

[3oh6]

the big thing guys is that 16X / 8X doesn't show any performance hit compared to 16X / 16X. in my EVGA X58 SLI review, it was the first thing i was interested in and spent a lot more time than just the results i posted on it. to this day, i can not find any repeatable or remotely significant difference with a pair of GTX 280s. i just spent last night benching the CPU under LN2 again and with the cards clocked at 756/1512/1296 and the i7 965 at 5800MHz...there wasn't a lick of difference between the cards in 16X slots versus 16X/8X in 3DMark 06. some other configuration, possibly 3-way SLI, may show gains to the larger width PCI-E lanes but so far i can't find a thing with a couple cards...but i'll keep testing

good to see some more testing on this subject though

[MpG]

The X58 is inherented limited to only 36 PCI-E lanes, and I'm not sure how you could truly expand that, short of reengineering the chipset itself. I expect that if there were multiple devices NOT working together, the NF200 could give each one 16 lanes as needed, as long as they didn't all need the lanes at the same time.

Incidentally, if you check out the Evga X58 board review that just put up, it's got a section testing out the effects of 8x and 16x lane combinations. Conclusion: difference = pretty much squat. So even if the NF200 wasn't gimped in application, there doesn't seem to be room for improvement in the first place. And by the time there is, PCI-E gen 3 will probably be showing its head.

Edit: 'd. You gonna make a thread for that board review?

[ImmaPC]

@MpG & 3oh6

I was just reading the X58 SLI review, and I was so excited to see the results of x16/x16 vs. x16/x8 that I pretty much skipped right ahead to it. I was actually surprised that 2x16 slots actually showed very little, if any improvement, vs x16/x8, or even 2x8! Perhaps it's because the video cards simply don't need all that bandwidth/throughput yet, maybe in future video card models (like GTX295, GTX350 ) we might see the PCI-E maxed out a bit, and the difference between number of lanes might make a difference. Until now, I'm shan't fret over wether I have enough bandwidth, and no-one should!

Also, 3oh6, I'd just like to thank you for the review, I thought that every aspect of it was thoroughly covered, and I especially liked that you took the extra time to test it with 2x16 SLI, X16/X8 SLI, and 2x8 SLI. Thanks to your review, I have no doubts that I bought the right board for my i7 setup . Now if only that dang RAM would arrive...