Final PCIe Bandwith Comparison Test - Real Word Performance Benchmarks Linus Tech Tips

Linus Tech Tips ·Linus Tech Tips ·2011-05-08 · 669 words · ~3 min read
Floatplane YouTube

Transcript

JSON SRT VTT 54
0:00 all right i'm going to take another crack at my PCIe 16x versus 8x versus 4x
0:05 testing this time i'm going to use a couple of games that i've benchmarked
0:08 quite a bit in the past we're going to go with battlefield bad company 2 as
0:12 well as metro 2033 since i had people
0:15 asking for not a purely synthetic test so i'm going to use the 69 90 again
0:20 single card gonna run at 16x 8x and 4x
0:24 PCIe bandwidth and see what kind of performance differences we see at 1080p
0:29 running in these games
0:32 so this was interesting uh not only did i test battlefield bad company 2 and
0:37 metro 2033 actually i've got my fraps one of my fraps results up here
0:42 right now oh apparently i have a lot of them up here right now okay so i also
0:47 added crisis 2 to the testing
0:50 methodology so here you go here are my results in a very tiny little graph that
0:55 i have made on a piece of paper along with a list of stuff that i had to do
0:59 the other day okay so here we are oh please focus come on you'll be kidding
1:04 me focus
1:07 focus there we go yeah yeah
1:11 okay so crisis two from 16x to 8x we're looking at pretty
1:15 much within margin of error but then as soon as we drop down to 4x PCIe
1:20 bandwidth we lose about six to well six
1:24 to eight FPS so that's going to be about 10 so we do see a tangible real world
1:28 performance decrease on a very high performance card in crisis 2 going down
1:33 to 4x that'll feel bad company 2 once again we see kind of within the margin
1:37 of error and then as soon as we go down to 4x we
1:40 see a very dramatic decrease in performance about 30
1:44 so metro 2033 same story this is margin of error right here and then moving down
1:50 to 4x we see a yeah 12-ish percent
1:54 performance drop so basically this goes
1:57 against the findings that i found before and i believe i know the reason so
2:01 running the extreme preset on 3d mark 11 basically meant that the
2:08 processors on the Radeon 6990 were completely taxed
2:14 so we were creating a processor bottleneck which meant that there didn't
2:19 have to be as much communication between the card and the rest of the system
2:22 because we weren't able to actually we were we were limited by how much
2:27 rendering these cores could do rather than by how quickly the card could
2:31 communicate with the system so running at 1080p with some modern games we were
2:37 able to create a scenario whereby the
2:40 card was not actually the bottleneck in the performance of the system in that
2:45 particular game so the processors were able to do more than the pci express
2:50 bandwidth was capable of communicating
2:54 so basically what we see here then
2:57 is that as soon as PCIe 3.0 comes out
3:00 we're probably looking at a scenario remember pci express is doubling in
3:04 bandwidth every time we see a new iteration but we're probably looking at a scenario where any modern card will
3:09 perform about like this at PCIe 3.0
3:13 4x and will perform about like this at PCIe PCIe 3.08 x and then there will be
3:19 a hypothetical probably like incremental 0.2 FPS beyond that so for any modern
3:25 card and remember this is two modern high-end gpus those are each 6970s on
3:30 there for any two modern high-end gpus
3:34 PCIe 8x is pretty much enough and 16x is
3:37 definitely enough based on how little improvement we see by doubling the
3:40 bandwidth from a PCIe 8x slot to a 16x slot so thank you for checking out this
3:44 episode of Linus tech tips don't forget to subscribe for more unboxings reviews
3:48 and other computer videos