How Many PCIe Lanes Do You Need?
We've already proven that the performance impact of reducing PCIe bus bandwidth on a single card is essentially negligible (see The Myths Of Graphics Card Performance: Debunked, Part 2). But does the same claim hold true in multi-GPU configurations?
A while back, I was struggling with what seemed like poor performance scaling from GeForce GTX 980s in SLI. I was seeing numbers 30% lower than what I was expecting. And my CPU wasn't the bottleneck, either. Furthermore, the issue only materialized when I had a G-Sync-capable display connected, regardless of whether G-Sync was on or off. I came to suspect some sort of driver issue, before Nvidia and EVGA forum users suggested that I was the only one having this particular issue. If you're a system builder, this is unambiguous guidance: you are the one doing something wrong!
I couldn't make sense of what I was seeing until, about a week later, I remembered that I had left the PCIe Generation setting in my motherboard's firmware at first-gen transfer rates. I was running the two cards at x8 PCIe 1.0! That's a paltry 2 GB/s bandwidth, equivalent more or less to two third-gen lanes. ASRock has a nifty setting in its UEFI that lets you specify this parameter. It's something you never want to mess with unless you are, as I was, testing different PCIe link speeds.
Switching back to PCIe 3.0 in the motherboard UEFI alleviated that 30% performance handicap. But, perhaps even more interestingly, it allowed me to make some interesting inferences.
- Eight lanes (x8) of PCIe 3.0 is more than enough for two top-tier Nvidia cards in SLI. That is, from a PCIe bus effectiveness standpoint, you won't benefit materially from a motherboard with an (expensive) PLX chip or the leap to Intel's LGA 2011-v3 interface.
- Four lanes (x4) of PCIe 3.0 would most likely be fine for SLI. Nvidia does not support three-way SLI on x8/x4/x4-capable PCIe 3.0 motherboards. That's an uncommon scenario, so it matters little anyway. Three-way SLI is pretty rare to start, and if you're in the market for $1500 worth of graphics cards, you can most likely afford an extra $100 for a PLX bridge-equipped motherboard that'll give you the number of lanes Nvidia requires.
- G-Sync somehow materially increases usage of the PCIe bus in SLI. This isn't an issue per se, but it's a somewhat interesting fact if you are curious about understanding precisely how new technologies work. I still wonder why this is the case.
Do you think we'd see 1080p monitors with 200hz+ in the future? Would it even make a difference to the human eye?
I also believe that alternating frames is utter crap. The fact that this has become the go to standard is a travesty. I dont care for fake fps, at the expense of consistent frames, or increased latency. If one card produces 60fps in a game. I would much rather have 2 cards produce 90fps and both of them work on the same frame at the same time, then for 2 cards to produce 120 fps alternating frames.
The only time 2 gpus should not be working on the same frame, is 3d or vr, where you need 2 angles of the same scene generated each frame. Then ya, have the cards work seperatly on their own perspective of the scene.
However, If i need to buy 2 980s to run a VR set or a 4K display Ill just wait till the prices are more mainstream.
I mean, in order to have a good SLI 980 rig you need a lot of spare cash, not to mention buying a 4K display (those that are actually any good cost a fortune), a CPU that wont bottleneck the GPUs, etc...
Too rich for my blood, Id rather stay on 1080p, untill those technologies are not only proven to be the next standard, but content is widely available.
For me, the right moment to upgrade my Q6600 will be after DX12 comes out, so I can see real performance tests on new platforms.
I had my eye on the two Acer monitors, the curved 34" 21:9 75Hz IPS, and the 27" 144HZ IPS, either one really for a future build but this piece of info tells me my i5 will be a problem.
Could it be that Intel CPUs are stagnated in performance compared to GPUs, due to lack of competition?
Is there a way around this bottleneck at 1440P? Overclocking or upgrading to Haswell-E or waiting for Sky-lake?
While true from a certain perspective, it should be clarified that you need 2 of the same number designation. As in two 980's or two 970's. I fear that new system builders will hold off from going SLI because they can't find the same *brand* of card or think they can't mix an OC 970 with a stock 970 (you can, but they will perform at the lower card's level).
PS. I run two 670's just fine (one stock EVGA and one OC Zotac)
What you say -was- true with 6xx class cards. With 9xx class cards, requirements for the cards to be identical have become much more stringent!