4K using PCI-E 1.0 , 2.0 and 3.0 what a shocker !!!

Kaapstad · 4 Jul 2014 at 16:16

I have just run the Sleeping Dogs bench @4K maxed to answer a question about PCI-E 1.0 , 2.0 and 3.0 I got in another thread.

Confused Stu said:
If you get the chance, is there any possibility of seeing if PCI-E generation has a higher impact with the higher resolution? I know PCI-E 2 to 3 has a limited impact at 1080P (less than 5% from memory?) but wondering if that increases with 4K. To be honest, I just want to know how much my 2600K is holding back my two 290s at 3K as to whether it's worth the money to upgrade to Ivybridge/Haswell - I'm hoping not as I haven't got the money to spend!

4 x 290X @1000/1250 stock

4930k @4.0

4K

Sleeping Dogs maxed

PCI-E 3.0 x16/x8/x8/x8

PCI-E 2.0 x16/x8/x8/x8

PCI-E 1.0 x16/x8/x8/x8

There is a big difference in FPS between 2.0 and 3.0 which was more than I was expecting.

Using PCI-E 1.0 resulted in another big drop and worse still the bench looked like a slide show. Even more serious was when selecting settings my mouse was barely usable and it was much quicker to use the keyboard to select bench settings. All in all PCI-E 1.0 was terrible.

Tute · 4 Jul 2014 at 16:21

Am I reading it wrong? I was actually impressed with 52fps in 1.0, I thought it would bomb into single figures - that's hardly a slide show?

LtMatt · 4 Jul 2014 at 16:23

Interesting results Kaap. Is it possible you could test pci-e 2.0 x16 x16 in a two way setup on this game and maybe another game? Considering 4K myself tbh but want to do plenty of research before hand so i don't get caught with my trousers down anywhere.

Kaapstad · 4 Jul 2014 at 16:29

Tute said:
Am I reading it wrong? I was actually impressed with 52fps in 1.0, I thought it would bomb into single figures - that's hardly a slide show?

To watch it running it is very cartoony using 1.0 and a lot worse than the figures suggest.

When it comes to gaming I am not too fussy about image quality having started on DOS games but I could not put up with using PCI-E 1.0 on this.

bru · 4 Jul 2014 at 16:32

Nice.

When I saw the thread title with the 'shocker!!' I almost expected there to be no difference between them.

As matt suggests two way and if possible single card results would be great.

Kaapstad · 4 Jul 2014 at 17:03

LtMatt said:
Interesting results Kaap. Is it possible you could test pci-e 2.0 x16 x16 in a two way setup on this game and maybe another game? Considering 4K myself tbh but want to do plenty of research before hand so i don't get caught with my trousers down anywhere.

Two card results I have just done are much closer to each other.

2 X 290X @1000/1250

4930k @4.0

PCI-E 3.0 x16/x16

PCI-E 2.0 x16/x16

PCI-E 1.0 x16/x16

I will do more testing when I get time.

Remember I am using an IB-E hexcore, on SB things may be different. I know from past experience the IB-E CPUs are very strong with PCI-E bandwidth.

LtMatt · 4 Jul 2014 at 17:05

Thanks Kaap, much appreciated. That's put my mind at rest in one regard, but made me think i might need three cards.

I think i might be ok regarding bandwidth as ive noticed the Plx chip on my mobo takes strain off the cpu and pci-e lanes. I get better firestrike physics and combined scores when i use the Plx chip lanes on my mobo. Maybe that would help a bit?

Kaapstad · 4 Jul 2014 at 17:38

LtMatt said:
Thanks Kaap, much appreciated. That's put my mind at rest in one regard, but made me think i might need three cards.

I think i might be ok regarding bandwidth as ive noticed the Plx chip on my mobo takes strain off the cpu and pci-e lanes. I get better firestrike physics and combined scores when i use the Plx chip lanes on my mobo. Maybe that would help a bit?

There is a lot of stuff we don't know about 4K this is why I am posting stuff whether it makes hardware look good or bad. It is more important to know what the limits of 4K is.

It was no accident that I used the 290Xs to test the PCI-E bandwidth, as they come with a 512bit bus this tends to rule out any short comings with the cards. I will test later with the Titans and their 384bit bus but I expect the results to be similar at best and possibly worse than the 290Xs in a 4 way setup.

Another thing on the horizon is Maxwell full fat, these cards are going to be about 50% faster than anything we are using now, what happens when they run @4K - will the PCI-E slots bottleneck them?

Disco_P · 4 Jul 2014 at 17:55

Why is VRAM showing as only 3GB as opposed to 4? Game limitation I guess?

layte · 4 Jul 2014 at 18:18

Disco_P said:
Why is VRAM showing as only 3GB as opposed to 4? Game limitation I guess?

Yea. This and other games often show 2/3GB on my Titans.

coupe69 · 4 Jul 2014 at 18:39

Is there really any need for 4k screenshots of benchmark figures!?

Mal X · 4 Jul 2014 at 19:11

LtMatt said:
Thanks Kaap, much appreciated. That's put my mind at rest in one regard, but made me think i might need three cards.

I think i might be ok regarding bandwidth as ive noticed the Plx chip on my mobo takes strain off the cpu and pci-e lanes. I get better firestrike physics and combined scores when i use the Plx chip lanes on my mobo. Maybe that would help a bit?

you're daft going for 3 cards, because it's bad enough replacing one card in a years time; let alone 3.

our cards aren't powerful enough for 4K, but they might be late next year

Mal X · 4 Jul 2014 at 19:14

Kaapstad said:
Another thing on the horizon is Maxwell full fat, these cards are going to be about 50% faster than anything we are using now, what happens when they run @4K - will the PCI-E slots bottleneck them?

yea and bloody expensive too.....yea the mobo will be too slow ha ha !

Eurofighter · 4 Jul 2014 at 22:53

Interesting. Better make sure you have PCI-E 3 if using 4 cards!

IceShock · 4 Jul 2014 at 23:28

Jesus, first post gave me almost a heart attack, then I read down the two card tests and my heart started beating again

im running pci-e 2.0 7950cf, so I guess im not missing much using 2.0 over 3.0

Silent_Scone · 5 Jul 2014 at 00:16

Interesting results Kaap. New bus before long.

Kanoe · 5 Jul 2014 at 00:44

What's the difference in single card between the PCI-E modes?

Is the fact that AMD are not using an external bridge between the cards for the 290x contributing to higher PCI-E bandwidth being used now that they are routing the communication between the cards in crossfire across the PCI-E so therefore using them in PCI-E 3.0 mode yields better performance?

Orangey · 5 Jul 2014 at 00:45

MjFrosty said:
Interesting results Kaap. New bus before long.

Strangely not on X99, which I was hoping to play with soon.

Confused Stu · 5 Jul 2014 at 01:31

Thank you so much Kaap for taking the time to run these. My wallet panicked badly after seeing the results of the 4-way tests in post one but thankfully you tested dual cards too and they're much closer. Obviously I'm only running 3/4 the resolution you are but it's good to know paying out to go PCI-E 3 wouldn't make a noticeable difference in my case.

Now just need to keep persuading myself not to jack the Eyefinity setup in for a 4k screen. Why is nice shiny tech always so expensive but so, so tempting?!

Vega · 5 Jul 2014 at 02:30

Good tests. These mirror PCI-E tests I've done for years on my multi-GPU/multi-display setups. A lot of people didn't believe me of the large differences, but when stressed the PCI-E speed differences can be dramatic.