Widescreen Gaming Forum

Hi players,
I have successfully set up a 5x2 configuration that I can run as extended Windows desktops or as two independent SLS using Eyefinity. I am using two VisionTek 7750s with six displayports each.

I have the top row of 5 screens hooked up to one card, and the bottom row hooked up to the second card.

It works, and graphics performance within a window is pretty good. When I have a window that only spans the bottom row of screens, performance is fine. However, if I position that same window such that it spans both the top and bottom rows of displays, performance drops significantly: the program still reports a high frame rate, but the display is visibly choppy/erratic/low-frame-rate.

Can one of you educate me about why this is the case?

Do any of you have a non-SLI, dual-card setup where a window that spans screens powered by two separate cards exhibits a performance decrease?

Thanks for any tips...

Hi, are you able to tell us much more about your system configuration?

Motherboard, position of GPU's on motherboard, PSU, CPU, AMD driver version etc.

Between this post and another post in another thread, I have gathered that you are using 2x VisionTek 7750 in non CFX mode on a Windows 8.1 operating system.

_________________
| Gigabyte X79S-UP5-WiFi Skt2011 | i7 4820k @4.4GHz Watercooled (Ivy Bridge-E) | Corsair Vengence 32GB @2133MHz 11-11-11-27 2T Quad-Channel (8x4GB) | AMD R9-Fury-X 4GB @1140/505MHz + EKWB Watercooled | 256GB Crucial MX200 SSD | 8x 1TB Samsung F3 SATA-II | Corsair 850W TX PSU | 3x Dell S2209W 22"wide TFT's (5760x1080res) | Logitech Z-5400 5.1ch speakers | Win7 Ultimate x64 |

Ok, here's some info.
The symptom is: a window w/ 3d animation runs fine on the lower row. When displayed on the top row, it has a subtle "jerk" that occurs every ~1 second or so. When the window spans both the top and bottom row, the program reports the same FPS but there is some significant jerkiness to the animation.

Motherboard: ASUS Sabertooth P67.
Proc: Core i7-2600 3.4Ghz, 4 core
Ram: 16GB
Graphics: Two VisionTek 7750s
OS: Windows 8.1 x64

This is a dual 5x1 setup where each card controls one row of 5 1920x1080 60Hz displays. Top card controls top row, bottom card controls bottom row.
Using Eyefinity to configure two SLS (5x1).
Catalyst 14.12.
AMD Driver version 14.501.1003.0

One additional detail is that the top row of screens are physically flipped 180, and I have configured the 180 filp in the Catalyst Control Center.

Hi, have you tried to run the top row without rotation?
It may be the flip that is causing the issue

_________________
| Gigabyte X79S-UP5-WiFi Skt2011 | i7 4820k @4.4GHz Watercooled (Ivy Bridge-E) | Corsair Vengence 32GB @2133MHz 11-11-11-27 2T Quad-Channel (8x4GB) | AMD R9-Fury-X 4GB @1140/505MHz + EKWB Watercooled | 256GB Crucial MX200 SSD | 8x 1TB Samsung F3 SATA-II | Corsair 850W TX PSU | 3x Dell S2209W 22"wide TFT's (5760x1080res) | Logitech Z-5400 5.1ch speakers | Win7 Ultimate x64 |

I tried out a handful of different combinations and can confirm that the issue is not due to the flipping/rotation. I can even replicate the issue with Eyefinity off, with just one monitor plugged into each card, and no rotation (or with, it doesn't matter).

In other words, I seem to be taking a performance hit as soon as I have a window that spans two monitors that are being driven by two separate cards.

Is there some kind of over-the-bus synchronization or communication that takes place in this scenario? It's the only thing I can think of, but I really don't have any idea what I'm talking about.

Some more experimenting...

I changed the resolution of my window and when the total number of pixels (width x height) are small enough (eg: 2200 x 1080) I cannot replicate the issue.

When I exceed a certain threshold of pixels, the framerate drops dramatically when I span the window across displays attached to the two separate cards. While I could understand how a gradual drop in performance might occur, do any of you have any ideas why I'd see such a sudden drop in performance? I literally go from 60fps to 30fps very suddenly.

So here is the deal. You have a total of 16 PCI-E lanes available from your CPU. Then you have two x16 full length PCI-E slots on your board. While they are both wired for x16 with 2 graphics cards occupying the slots they go to a x8 x8 configuration.This normally wouldn't be an issue but if you have anything plugged in to the x1 slots that will draw a lane per slot, also if you have any sata devices plugged into the off colored sata ports which are from the Marvell sata controller that will also take away from your available lanes. If you look at the top x16 PCI-E slot you'll notice it is a different color. This is the one they suggest you use for your GPU. The ASUS BIOS gives that slot which will be labelled as primary graphics adapter some priority over other devices to ensure a more uniformed experience across the most common configuration of only a single GPU. The other thing is that these are all PCI-E 2.0 lanes. Normally this wouldn't be an issue even with a dual GPU setup but with 10 monitors going, the amount of bandwidth that is being requested just isn't available. So even if you are doing nothing of note on the bottom row it is not going to give you any of the lanes set aside for the first x16 slot whether they are idle or not. If they were PCI-E 3.0 slots you would have more than enough but that is not the case. A quick way to check this (Which is something I'm sure you've already checked in your quest to find the issue) is to switch the cards with one another. If nothing changes even with a clean driver install you know it is a hardware issue.

I am a little delirious from long hours so I hope this is clear enough to make some sense. Have a nice rest of the week

Makes total sense, @Peer_Review, thanks for taking the time to explain.

Why do you say that PCI Express 3.0 would alleviate the issue?

Looks like I'm going to upgrade the motherboard and report back with the results.

Sorry for not responding sooner. My mother had her hip replaced so I've been a little occupied taking care of her.

The reason that PCI-E 3.0 would most likely fix the issue you are having is that PCI-E 2.0 lanes are capable of 800 MB/s, while PCI-E 3.0 is somewhere around 4,000 MB/s. This would give you a substantial increase in the amount of overall throughput available before saturating the available lanes. I am not a fan of using the term bottleneck since it is very rarely used correctly but with 10 monitors and the CPU/Motherboard/GPU configuration you have this is most likely the case. It all comes down to the way board partners (Asus, Gigabyte, MSI etc) wire the boards, and the way they handle the switching of lanes when populating different PCI-E slots. Even on a board with PCI-E 3.0 when adding your second card it is going to drop from x16 to x8x8 but with all the extra overhead you should be fine. You might also look into a board with some extra PCI-E lanes given from a PLX chip. If you have a row of 5 monitors you are not going to be using for gaming or anything heavy you could put that card in a slot with lanes provided by the PLX chip and keep your CPU lanes all for your main card and its corresponding x5 row of monitors.

Thanks @Peer_Review, hope your ma's recovering all right.

I did some digging into PCI lanes and throughput, and bought some new hardware. I know everyone says that video cards can't saturate x16, but I don't think they're considering driving ten 1080p monitors. I haven't done the math.

However, I did purchase an i7 with CPU 40 lanes and a mobo with true dual x16 support, so I hope to get true PCI 3 x16 on both cards for maximum throughput.

What do you think? Unnecessary? (Guess I'll find out once the hardware arrives.)