EPT![[image loading]](http://www.overclock.net/attachments/intel-motherboards/191693d1295759743-sb-cfx-8x-8x-vs-16x-m2033.jpg)
dx11... yeah its old but from this generation o.O
Computer Build Resource Thread - Page 624
| Forum Index > Tech Support |
When using this resource, please read FragKrag's opening post. The Tech Support forum regulars have helped create countless of desktop systems without any compensation. The least you can do is provide all of the information required for them to help you properly. | ||
|
Shikyo
Finland33997 Posts
dx11... yeah its old but from this generation o.O | ||
|
JingleHell
United States11308 Posts
On October 04 2011 09:24 Medrea wrote: The performance difference between running a GPU in x16 mode compared to x1 is a little under 10 percent. Its really that small of a difference. And remember there is x16 on the first card in the pro3. Opposed to x8. x8x8 is really fucking close to x16 x4. Do you have any recent benchmarks? I can't find any, but, as you'd expect, the older ones I found show that it varies based on games and cards. It's also painfully biased against nVidia. Ah, never mind, shikyo found something much better than the old bullshit I found. And of course, this is coming from me, but any loss of performance that can be mitigated for a reasonable cost should be. Poor scaling due to poor hardware selection is criminal. | ||
|
Shikyo
Finland33997 Posts
| ||
|
Medrea
10003 Posts
Is a video for x16 x16 x8x8. But more interestingly. This is what I was operating under. If you dont want to watch the video, it basically shows that there is a very small difference. I mean he runs the 6990 in x2 x2 mode. Thats pretty low IMO. | ||
|
Shikyo
Finland33997 Posts
| ||
|
Medrea
10003 Posts
On October 04 2011 09:44 Shikyo wrote: afaik crossfire/sli are different for that because of how the second card must be connected or something o.O I read that 16x -> 4x is around 100% -> 96% but those benchmarks I found don't seem to support it Well, in this case we are using an internally crossfired GPU in 2x 2x mode and seeing very little performance drop. About 3 or so percent I think. In an abomitable setup nonetheless. For crossfire at least, it seems that the card just doesnt give a major fuck. This runs parallel to what I have found. Where I have Radeon 5 series cards all playing off the x1 minislot and they run at expected performance, can even boot from them. i was actually wrong when i said 10 percent earlier, seems like it matters even less. | ||
|
JingleHell
United States11308 Posts
On October 04 2011 09:46 Medrea wrote: Well, in this case we are using an internally crossfired GPU in 2x 2x mode and seeing very little performance drop. About 8 or so percent I think. In an abomitable setup nonetheless. He's using 3dmark11. While it's wonderful for standardizing benchmarks, it has serious limitations as far as "realistic situations" like he's claiming to shoot for. It's kind of similar to Furmark, in that it's designed to push to limits that games don't reach, except instead of power draw and thermals, it hits the GPU as hard as it's able. Meaning the card is being pretty seriously limited by the GPU's rather than memory usage. The problem is, there's pretty much no game that can hammer the GPU that hard without having heavy duty, memory intensive textures and shitloads of AA and AF, in particular, using significantly harsher AA than the 4x MSAA on the 3dmark11 Extreme preset. Any realistic gaming situation maxing a game will be beating bandwidth harder, due to memory use. You'd run into similar issues with Bitcoin mining, I would expect, which, IIRC, is your primary use? | ||
|
Munk-E
United States672 Posts
| ||
|
JingleHell
United States11308 Posts
On October 04 2011 09:52 Munk-E wrote: So what I'm getting from all this is most of you think I should get the 90$ Motherboard, am I right, or just confused? Uhm, only one person. We've gotten into the marginally academic argument of how much it matters, although it is pretty much agreed that there will be some difference with the more expensive board. But now we're arguing theory as to how much. | ||
|
Medrea
10003 Posts
On October 04 2011 09:52 Munk-E wrote: So what I'm getting from all this is most of you think I should get the 90$ Motherboard, am I right, or just confused? It's honestly not an enormous deal either way. Im coming with pretty hard evidence that bandwidth is by no means a modern bottleneck but others seem to think otherwise with there own, also valid, reasons. | ||
|
skyR
Canada13817 Posts
And if you follow Linus, he does a follow-up video with real world performance and there is a difference. | ||
|
Medrea
10003 Posts
EDIT: Ok well that was interesting. Seems like bandwidth becomes a bottleneck once we go from x8 to x4. But thats comparing x8 x8 to x4 x4. In an external crossfire configuration, how is bandwidth divided up between the x16 slot and the second slot? Is it even, is it weighted towards the first card with the higher bandwidth? | ||
|
Shikyo
Finland33997 Posts
| ||
|
Bambipwnsu
Canada698 Posts
Seeing as people on the evga forums that have this mobo have gotten terrible temps in this config and that I am really anal about the loudness of the fans, my only option was to smack this 2nd card into the x4 slot that was below. I've read several "recent" reviews that seem to point that a x16/x4 in SLI for nvidia cards isn't too atrocious. http://www.overclock.net/nvidia/819348-16x-16x-vs-16x-4x-gtx.html ~ 2% - 6% depending on game http://www.hardocp.com/article/2010/08/25/gtx_480_sli_pcie_bandwidth_perf_x16x16_vs_x4x4/ ~ shockingly x4/x4 has negligible (read almost none) effects on BFBC2 @2560x1600 4xAA 16xAF @ very high settings and similar results on Metro2033...which gives me hope because I bought this second card really just for BF3. Of course there are also other reviews such as that one about the 5870 that show some pretty substantial losses going on. I'm personally mentally ready to take up to about a 10% hit on the scaling which I hope i can make up just by OCing. As Jingle mentioned I did some very "light and unprofessional testing" by swapping the single card I have currently from the 16x to the x4 slot, my 3dmark graphics score went from 4100ish to 3700ish so on 3dmark there is a difference as far as I know. Also from what I read x8/x8 is something like 2% drop only. | ||
|
Medrea
10003 Posts
The bridge is there for a reason, I assume part of it is that so the GPU's don't need to communicate through the board. I don't think its an x4 bottleneck because i doubt both cards have the exact same amount of information they are communicating through the bus with. I always assumed cards in Crossfire work under some kind of master/slave orientation under the hood. | ||
|
Shikyo
Finland33997 Posts
| ||
|
Medrea
10003 Posts
On October 04 2011 10:08 Shikyo wrote: Btw that's one thing I'm interested about, if you can only run a game at 30 fps or so, does that even matter? do you even use up the 8x or 16x bandwidth? It probably causes framedrops above 60 yeah but those are useless frames pretty much... Exactly. From what I have seen, a card running in x16 mode only pulls very far ahead of the other modes when frames are very high. When framerates are below 60, the width of the bus seems to matter less and less, but that last part is anecdotal. | ||
|
CharlieBrownsc
Canada598 Posts
On October 04 2011 09:56 skyR wrote: This is another article which is fairly recent on the P67 platform as well on this issue: http://www.tomshardware.com/reviews/pci-express-scaling-p67-chipset-gaming-performance,2887-9.html And if you follow Linus, he does a follow-up video with real world performance and there is a difference. http://www.youtube.com/watch?v=vfXALgE7mVM Does he have a tiger? | ||
|
Medrea
10003 Posts
It came with his PCIE cables. I want to see this later article on tomshardware that was mentioned. | ||
|
JingleHell
United States11308 Posts
On October 04 2011 10:11 Medrea wrote: Exactly. From what I have seen, a card running in x16 mode only pulls very far ahead of the other modes when frames are very high. When framerates are below 60, the width of the bus seems to matter less and less, but that last part is anecdotal. No, that's pretty much what we've all been talking about, is GPU use vs bandwidth use. In cases like bitcoin mining and folding, the slot really barely means shit, because it's GPU intense. Same as a synthetic benchmark like 3dmark 11. Yes, it's the best you can get for standardized benchmarking with gaming similarities in a synthetic benchmark, but the higher presets are smashing the GPU harder than any game does, limiting memory and bandwidth use. It all comes down to what resources actually get utilized, instead of what resources are available. It's kind of like pointing to SC2 performance as a benchmark for Metro. No matter how well you benchmark something, if it isn't working the same way, it's not providing useful data. Also, for gaming purposes, I'd rather have 2% improvement on minimum FPS than 8% on maximum. ^Only time you'll catch me saying that... | ||
| ||
