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

Brahmzy said:
So, is there any latency involved when using 2-way SLi?

I have a a pcie1x X-i T, and obviously 2 dual-slot GPUs if I go SLi... what would the best slots be used for this? And again, would I be bypassing any latency of the NF200?

So...
Slot 1 - GPU 1
Slot 2 - Empty
Slot 3 - GPU 2
Slot 4 - Empty
Slot 5 - Empty
Slot 6 - X-Fi Ti

Would the NF200 even be used in this config?
Click to expand...

..no,:mad::mad::mad: ..sorry, just being cynical about this - my point is it adds latency, most chips along 'interconnects' do...TBH, if it is it wouldn't help:rolleyes::rolleyes::rolleyes: according to above, slots 1 & 3...doesn't make sense to me:confused::confused::confused:


hey Kyle..new catchphrase, to 'by sixteen someone'!!!!;););) To sell someone fanboi crap that doesn't make a bit of difference....:D:D
 
Hi everybody. I'm not so active on this forum but I've always been reading HardOCP reviews with interest because I've always found them the most technically focused. But I must say that this time I'm bit disappointed: I think that the game chosen for the test and most of all, considering only the average FPS makes the result totally meaningless; a game that in worst case scenario runs at 80FPS is clearly not stressing the system at all, I know that Crysis is awfully coded as Kyle wrote here (Warhead runs so smoother and it's the same game! :eek:), but it's the only title out there able to put on its knees a 3x280GTX system (whatever motherboard you have).
Most of all I think that for a bandwidth vs latency test like this one, we should be looking at the moments when the game stresses most the system so not at AVERAGE FPS but at MINIMUM FPS or even better the TIME-FPS graphs the HOCP introduced and that i love (microstuttering? Anyone? :cool:).
I'm sure that soon HOCP will provide a more complete review of the situation and I invite all people who say "NF200 is bs" or "it's needed anyways for true 3-way SLI" to wait a little more time before judging.
I have 3 280GTXs and I think I'll get the "Revolution" (as soon as they release a working BIOS): it costs almost the same (in some shops even less) than the R2E while offering much more features like SAS, high stability and no fancy-useless chipset cooling system that often are just a big hassle when it comes to placing big heatsinks on the really hot parts (CPU, GPUs).

P.S.: I see Kyle is a good looking guy and not the usual nerdy-style reviewer you find on IT-related sites and so he may want to show himself to the female public out there ;), but i still prefer the good old text reviews on videos.
 
That said I did study the actual timeline graphs and look at min and max framerate values. Like the average framerate there was nothing to report. All looked exactly the same. I played the game on both boxes too. I could not tell any difference. So I understand what you are saying, and I would have brought it up had it shown any importance. But I should have at least stated what I just typed here in the video. I will keep that in mind next time NFS! Thanks for the feedback.
 
Nice review Kyle
but you miss the setting that shows a bottleneck in your test which I think is
2560*1600 ultra high 8 aa.
My watercooled 3 way sli 280s are not bottlenecked by not having 16,16,16
because 3x 16 is a standard from the 790s. Compared to the ,from 280s 3 way sli point of view, middle class x58s 16,16,16 is a must for 280s 3 way sli. Since I still run the 790ultra and I am waiting for the nf200 board the bottleneck is that I am running 16x2.0,16x1.1,16x2.0 so the 16x1.1 slot. The best way
to see the bottleneck is Crysis 64 2560*1600 very high,here it stutters. If I overclock the 16x1.1 2500 mhz to 3300 mhz I almost completely release the bottleneck at 2000*1500. Unfortunatly, I cannot
overclock the slot to the needed 16x2.0 5000 mhz. So having 16x2.0 5000mhz x3 on the nf200 should
release the bottleneck completely. So please try Crysis 64 2560*1600 very high,if it runs the bottleneck is gone and thats where the nf200 shines. Then it is a must for you to use a nf 200 board
in your home system if you run 3 280s
 
EtheEnthusiast said:
Nice review Kyle
but you miss the setting that shows a bottleneck in your test which I think is
2560*1600 ultra high 8 aa.
CUT
Click to expand...

That can be already desumed by the trend in Kyle's test results: as long as the resolution is raised the difference between the X58-only system and the X58+NF200 tends to lower till 0 at the highest res; "interpolating" this results one could say that at even higher settings (or with a more demanding game like Crysis) the NF200 board should overtake the X58-only system.
 
Got the new BIOS and 3 216-core gtx260's. I don't have anything to compare it to (other than butter), but it runs SMOOOOOTH.

This is compared to trying the same thing on the old BIOS.. which was smooth like sandpaper.

I was gaming at 2560x1600 playing Left 4 Dead at max settings and it never dropped below 50 and was well above 70 most of the time.

Dang, it would have been so much better if I had 1 more frickin' FPS in my FPS.

Oh, and it folds like a beast. Getting well over 20K PPD without doing any overclocking (yet).
 
I want a giant monitor and 3 GTX 280 graphics cards....

BTW, I thought the video was great I enjoy seeing what you guys actually do for testing. I'd like to see longer ones with you guys setting everything up. Besides in stores, it's not often that you see so many GPUs together.
 
NeedForSpeed said:
That can be already desumed by the trend in Kyle's test results: as long as the resolution is raised the difference between the X58-only system and the X58+NF200 tends to lower till 0 at the highest res; "interpolating" this results one could say that at even higher settings (or with a more demanding game like Crysis) the NF200 board should overtake the X58-only system.
Click to expand...

True but the difference is much bigger, FC 2 2560*1600 ultra high 4 aa has no bottleneck but 8 aa has. When the bottleneck comes visible the middle class x58s and 790s will stutter,crash or drop unplayable.The NF200 will provide even 60 fps with the very same 3 way sli gpus.
Although I was boycotting Asus for 3 years now I just ordered the nf200 p6t6 ws revo.
@ xaxxon nice to see it running :)
Another thing is that the other slots on the nf200 are indirectly blocked because bandwidth
would be shared.Like putting in a soundcard it would be 16,16,8,8 only for 3 way sli 280s with soundcard. I hope the hd sound chip is good but if Kyle says that I believe it.
The same problem comes when a fourth gpu is added as dedicated physix gpu.It would
be 16,16,8,8 for 3 way sli with dedicated physix. Only 3 295s would have the full bandwidth like 295 sli 16,16 with dedicated physix 295 at 16. Still physix is not allowed in vantage and there is no game that needs dedicated 280,285 or 295 dedicated physix.
But in the end we need a board with 2 nf 200 chips to have full bandwidth next to 3 way sli
Although I preferre EVGA I try this Asus nf 200 board lets see
 
xaxxon said:
Got the new BIOS and 3 216-core gtx260's. I don't have anything to compare it to (other than butter), but it runs SMOOOOOTH.
Click to expand...
Where didi you get it? Kyle sent you? Because on Asus sites they still have only the "first release BIOS 0208". Does this address also the CPU voltage problem?
Tomorrow will be my day off from work, I'm planning about going out to get the "Revolution" but I wanna have a working BIOS downloaded before doing it.
 
Wondering as well, and if I can snag a copy as I have the P6T6 and 3x 280s in hand. Haven't finished putting it together yet, but don't want to wait for Asus to release it before I start, so would like to grab that if I can.
 
ND40oz said:
NF200 is PCIe 2.0. NF100 was PCIe 1.1. It's 8 GB/s to the NF200 chips and from there 8 GB/s to each of the two cards. If you're using all 4 slots, then it's 4 GB/s to each slot.

If you look at Everest with a 9800GX2, you'll see each of the cards in the 9800GX2 has a full PCIe 2.0 x16 connection because of the NF200 chip on the card. Same thing applies here.
Click to expand...

http://www.fudzilla.com/index.php?option=com_content&task=view&id=5193&Itemid=37
the nForce 200 or n200 chip is an interesting little beast, as it is a PCIe 2.0 controller which adds 32 lanes of PCIe 2.0 bandwith, but only communicates with the chipset using 16 lanes of PCIe 1.0 bandwidth.
Click to expand...

As I said before, the connection between the two individual cards may be 2.0 x16; but the connection from the nF200 to the Northbridge is 1.0 x16. The 9800GX2 must have a different implementation that properly supports 2.0 x16, or else it's just reporting the connection from each individual sub-card to the nF200; not the connection from the nF200 to the Northbridge.
 
hmmm...that is odd, why should a x16x16x16 loss to x16x8x8. I know he said there maybe some latency add, but still.
 
CharonPDX said:
http://www.fudzilla.com/index.php?option=com_content&task=view&id=5193&Itemid=37


As I said before, the connection between the two individual cards may be 2.0 x16; but the connection from the nF200 to the Northbridge is 1.0 x16. The 9800GX2 must have a different implementation that properly supports 2.0 x16, or else it's just reporting the connection from each individual sub-card to the nF200; not the connection from the nF200 to the Northbridge.
Click to expand...

Well, I doubt I'd trust a link from Fud, but anyway:

http://techreport.com/articles.x/14661/2

With 780a they're definitely using a PCIe 2.0 x16 interconnect between 780a and NF200.

They should also trim traffic on the 16-lane PCI Express 2.0 link that connects the nForce 200 with the 780a SLI MCP. This link delivers a full 16GB/s of bi-directional bandwidth, which Nvidia says is plenty even for three-way SLI. However, it is worth noting that 16GB/s of interconnect bandwidth is only half of what would be needed to saturate all 32 of the nForce 200's PCIe lanes.
Click to expand...
 
Man, pass on the board for the reasons already mentioned, the ASUS P6T6 WS Revolution is still the one I lust over, but man, that video of FC2 makes me want to go out and get a pair of GTX 280s right now. I got bored of it after four or five hours but it looks sooooo smooth there....
 
I would be very interested to see how this board does in a dual card config. Using the main 16x slot, and one of the two that reportidly get funneled into one theoretically 16x channel. Maybe even see how it acts with dual 4870x2 cards.
Great review on this board though, this seems to be my favorite thus far.
 
Hi again, yesterday I bought this card but still ASUS hasn't released the new BIOS (that they told Kyle would have been released by the end of last week).
Can any of you who have at least the beta version please tell me where to download it or send me a PM?
Thanks in advance for your help.
 
I find it pretty interesting that even with new gen video cards we haven't tapped even 1/2 of the BW in a 16x link.. as for why 16/8/8 is faster than 16/16/16 in many cases, maybe there is some overhead with the multiplexing that the nf200 chip does to get a full 16 lanes for each slot in the 16/16/16 config?
 
Great piece of work, Kyle. I was just wondering what made you want to explore this NF200 vs. Sans NF200 topic? Did you notice something in your work that cued you in to a difference or you just had the time to delve into it?
 
revenant said:
I find it pretty interesting that even with new gen video cards we haven't tapped even 1/2 of the BW in a 16x link.. as for why 16/8/8 is faster than 16/16/16 in many cases, maybe there is some overhead with the multiplexing that the nf200 chip does to get a full 16 lanes for each slot in the 16/16/16 config?
Click to expand...

Two things:
- The NF200 chip is adding latency, this is inevitable, as it adds more transistors and buffers and switching actions between the PCIe devices and chipset.
- The two PCIe devices behind the NF200 chip may have 2x x16 links, but the link between NF200 <-> chipset is still just a single x16 link. With double the BW of a single GPU and the need to switch between two PCIe devices (since PCIe isn't a shared bus, it only supports one device per link), things will be slower.

If BW really was the issue with x8 slots, NF200 would make no difference either, as the link from the two 'NF200'-ed GPUs to the 3rd GPU on its own x16 slot would still be the same, BW-wise (2x x16 <-> NF200 <-> 1x x16 <-> chipset <-> 1x x16).

In other words, NF200 isn't for tri-SLI. It is to increase the BW between two PCIe devices connected to it. It makes absolutely no difference for any other PCIe devices on the system (in a positive sense at least). For a multi-GPU videocard like the 9800-GX2 it'd be an okay solution (no need to tweak anything, just drop it in), but that's about it.
 
Elledan said:
Two things:
- The NF200 chip is adding latency, this is inevitable, as it adds more transistors and buffers and switching actions between the PCIe devices and chipset.
- The two PCIe devices behind the NF200 chip may have 2x x16 links, but the link between NF200 <-> chipset is still just a single x16 link. With double the BW of a single GPU and the need to switch between two PCIe devices (since PCIe isn't a shared bus, it only supports one device per link), things will be slower.

If BW really was the issue with x8 slots, NF200 would make no difference either, as the link from the two 'NF200'-ed GPUs to the 3rd GPU on its own x16 slot would still be the same, BW-wise (2x x16 <-> NF200 <-> 1x x16 <-> chipset <-> 1x x16).

In other words, NF200 isn't for tri-SLI. It is to increase the BW between two PCIe devices connected to it. It makes absolutely no difference for any other PCIe devices on the system (in a positive sense at least). For a multi-GPU videocard like the 9800-GX2 it'd be an okay solution (no need to tweak anything, just drop it in), but that's about it.
Click to expand...

Good explination, but it still begs the question of whether or not there would be a speed in/decrease if one GPU was in the 'normal' PCIe 16x slot, and a second card was thrown in one of the NF200 driven 16x slots. The second NF200 slot of course stays empty. This would (theoretically) create two full 16x slots.

I would be very interested to see if there is any slowdown in this type of setup vs the same cards used in a non-NF200 board. I think this test would really prove how efficient the NF200 chip is.

Though probably not needed, but still interesting, I would also like to read a review on two cards, both of which on in NF200 driven slots. I would assume that this would be slightly better than a dual card setup in 2 8x PCIe slots... depending out how the bandwidth is controlled going from the NF200 chip to the chipset
 
jfreund said:
Can't watch the vid at work, but the chart tells the story. Not a large difference, but doesn't look like the NF200 is worth it for tri-SLI. Could be other applications for the PCIe lanes, though.
Click to expand...

i agreed
 
I was really expecting a stellar performance from the NF200. Then i saw the results. Then i read
Originally Posted by Elledan
The two PCIe devices behind the NF200 chip may have 2x x16 links, but the link between NF200 <-> chipset is still just a single x16 link. With double the BW of a single GPU and the need to switch between two PCIe devices (since PCIe isn't a shared bus, it only supports one device per link), things will be slower.
Click to expand...
and it all made sense

Originally Posted by Elledan
 
I saw that review on Tom's with the MSI eclipse Plus, it is a nice board but the BIOS are not the best. I do not think x16 3 is worth it at all, plus the NF200 chip loves to blow heat
 
jakethesnake011 said:
I saw that review on Tom's with the MSI eclipse Plus, it is a nice board but the BIOS are not the best. I do not think x16 3 is worth it at all, plus the NF200 chip loves to blow heat
Click to expand...
I used to have an Nvidia SLI motherboard and it would just fucking run hot. Is that typical?
 
Seems like these nf200 reviews may be cpu limited. You really do not get the full 3 way sli experience probably until you over 5ghz. At around 3.2ghz "2-way" and "3-way" scores are almost identical. Even at 2560x1600 there is only a 20% improvement over 2-way sli. In some cases tri-sli hurts performance when below 4ghz and 1920x1200 res vs dual-sli which does better. Check out this link for proof: "http://www.benchmarkextreme.com/Articles/I7%20920%20Bottleneck%20Analysis/P6.html". This all makes me wonder what the scores would be if you take the gloves off the cpu and allow all three video cards to max out the pci-e lanes??? Would the nf200 catch up??
 
I have triple sli on a 3.5G Q9450 790i - 1920x1200 the minimum framerates go up for me. Maximum fps - not so much. Also the video drivers have improved where I have no issues on any games like I did early on with the 3x280's. I have no doubt an i7 would get me a good 10-15fps more though.
 
Last edited:
Love the video and thanks for posting the data along with it. :)
 
You must log in or register to reply here.
Back
Top