Linus tests PCI-E lane speeds.

Gregster · 6 Oct 2014 at 08:27

An interesting watch for those that are curious on the differences between PCI-E X16/X8. I have always said it would be such a minimum difference, it wouldn't be noticeable

Chuk_Chuk · 6 Oct 2014 at 08:45

Surely it would be a good idea for him to have done the test with AMD cards since they no longer require crossfire bridges?

Kaapstad · 6 Oct 2014 at 08:48

Terrible review

4 cards do make a significant difference @4K

Why did he not use four 290Xs to test the 5820 ?

Why is the 4K Watch Dogs score so low for four 980s, my 4 Titans beat that easy. There is a good answer but he should tell people what it is lol.

If people are going to do these type of reviews they need to do a though job.

smogsy · 6 Oct 2014 at 08:53

good video

so tri-fire try-sli 5820k quad 5830k

however if your running say 2 GFX plus say soundcard & raid controller or network card 5830 would be a better choice

Gregster · 6 Oct 2014 at 08:59

Don't read too much into the results guys and it is merely showing the differences between bandwidth of the lanes. I wouldn't expect XDMA to add any speed either because it will still be limited by the PCI-E lane speeds and as shown for the 4x980s, you would need the 5930K but if you never intend to go 4 GPUs, the 5820K would be ideal and this will save you a few squid.

I thought it was interesting anyway.

Disco_P · 6 Oct 2014 at 09:28

Kaapstad said:
Terrible review

4 cards do make a significant difference @4K

Why did he not use four 290Xs to test the 5820 ?

Why is the 4K Watch Dogs score so low for four 980s, my 4 Titans beat that easy. There is a good answer but he should tell people what it is lol.

If people are going to do these type of reviews they need to do a though job.

I don't think they have 4 290X's.

Also they may have ran out of VRAM in W_D. But I am only guessing here.

Kaapstad · 6 Oct 2014 at 09:28

Gregster said:
Don't read too much into the results guys and it is merely showing the differences between bandwidth of the lanes. I wouldn't expect XDMA to add any speed either because it will still be limited by the PCI-E lane speeds and as shown for the 4x980s, you would need the 5930K but if you never intend to go 4 GPUs, the 5820K would be ideal and this will save you a few squid.

I thought it was interesting anyway.

If he tried 4 x 290X cards with the 5820k the drop in performance would be terrible on some games and I would like to think in his position he knows that. I have posted benches before with 4 cards @4K using PCI-E 2.0 and 3.0, as we know PCI-E 3.0 x4 is the same as PCI-E 2.0 x8. When switching between PCI-E 3.0 x16/x8/x8/x8 and PCI-E 2.0 x16/x8/x8/x8 the performance goes through the floor with the 290Xs on some games @4K and the Titans also suffer quite a bit as well.

Just ran these off for Linus benefit to show the 4K scaling on 4 cards lol.

290Xs @1200/1500
4930k @4.0
Tomb Raider totally maxed @4K

One card

Two cards

Three cards

Four cards

You get the same sort of scaling in lots of games on both AMD and NVidia.

Kaapstad · 6 Oct 2014 at 09:32

Disco_P said:
I don't think they have 4 290X's.

Also they may have ran out of VRAM in W_D. But I am only guessing here.

They did run out, it is totally unplayable on a 4gb card. You need 4 x 6gb Titans and a bit of overclocking to get it playable without stutter @4K.

Watch Dogs is also another game with good scaling from 1 to 4 cards at high resolutions, shame it is such a dog to run and I hate going through that long winded opening sequence to do some benching.

Silent_Scone · 7 Oct 2014 at 18:45

Just as well there is plenty bandwidth on GEN3, number of lanes aside.