PCIe Speed Test v0.1 CPU-GPU/GPU-CPU bandwidth test

[wez]

Hi Mechromancer,
Here is the image of GPU-Z

You are running x16 pcie already

[Stoner133]

I ran the test again today and saw something I didn't notice the first time. It had to do with grabbing the screen shot with Snag It. As soon as I grabbed the screen shot the test shut down. So this time I let it run longer and saw with Crossfire it actually tests both cards as seen here:


By knuckles6 at 2009-06-02

Then I ran it again with Crossfire disabled showing a big loss in the results, also note on the first one it listed it found two devices and without Crossfire it only finds one:


By knuckles6 at 2009-06-02

Anyone know of a way to keep the screen up after the test is complete?

[EniGmA1987]

You are running x16 pcie already

But not 2.0 speeds...

[Mechromancer]

You are running x16 pcie already

But not 2.0 speeds...

Yea, +1.

He has to find the BIOS options that allow PCIe slot adjustments. I don't have his mobo so I can't put up a guide to help him. Can anybody with that board give him a hand with the BIOS setup?

[HeamothoraxUK]

Hi Mechromancer,
I went into bios and did Ctrl F1 to unlock hidden features
then went Advanced Chipset Features -> PCIE Configuration.

This is what i see.

Primary Dual Slot Configuration [ AUTO/DISABLED]
Secondary Dual Slot Configuration [ AUTO/DISABLED]

PCIe_16_A_GEN2 SPEED MODE [AUTO/DISABLED/SW SWITCH/HW SWITCH]
PCIe_8_A_GEN2 SPEED MODE [AUTO/DISABLED/SW SWITCH/HW SWITCH]
PCIe_16_B_GEN2 SPEED MODE [AUTO/DISABLED/SW SWITCH/HW SWITCH]
PCIe_8_B_GEN2 SPEED MODE [AUTO/DISABLED/SW SWITCH/HW SWITCH]
PCIe_1 GEN2 SPEED MODE [AUTO/DISABLED/SW SWITCH/HW SWITCH]

PCIe Delay Time [0/255]

I've tried different combinations but not had much luck so far, maybe Lightman can shed some light on this as he has a similar setup.

[ahmad]

Mech this is interesting stuff! Thanks for creating the threads, I'll test this when I go home.

We are such benchmarking geeks. Love to bench stuff

[EniGmA1987]

PCIe_16_A_GEN2 SPEED MODE should be set to "HW Switch" probably. That should make the graphcis card tell the MB to use 2.0 speeds. If not, use "SW Switch" and Catalyst should tell it to use 2.0 speeds.

[Mechromancer]

PCIe_16_A_GEN2 SPEED MODE should be set to "HW Switch" probably. That should make the graphcis card tell the MB to use 2.0 speeds. If not, use "SW Switch" and Catalyst should tell it to use 2.0 speeds.

+1

On my Asus board, the SW (software) Switch doesn't force PCIe 2.0. I don't know if this is just a Cat 9.5 issue or what. HW Switch should work for you.

[Slagathor]

Same HT Link?, only change is higher PCI-e?

Sorry I missed this, only change is the higher PCIe.

It is default everything as I cant overclock it right now.

[Daveburt714]

Swap your 3870 and 4850 and see if the numbers change between boards.

Thats a good idea man, I'll try the 4850 on the GBT board within the next few days and post the results.

[HeamothoraxUK]

Hi Mechromancer & EniGmA1987,

so far i have tried,

Primary Dual Slot Configuration {Disabled}

PCIe_16_A_GEN2 SPEED MODE [SW SWITCH/HW SWITCH]

With no luck.Each time i uninstalled drivers and hardware before making the changesin bios, then reinstalling drivers still no change.
I've added a few screenshots that look a little odd, one of the IRQ's assigned



one from Everest of the pcie channels in use none of which are the graphics card



one from Sandra that says it's using a pcie1 X16 channel



and one from Sandra that says the graphics card has no interrupt assigned.



I'm not exactly sure whats going on.

[EniGmA1987]

Maybe try setting it to auto and see if it changes to 2.0?


Also, maybe you dont have your graphcis card in the primary slot, try setting all of your x16 slots to the 2.0 mode and see if GPU-Z reads it correctly.

[HeamothoraxUK]

Hi EniGmA1987,
Graphics card is in primary slot, Have tried all on Auto/SW Switch and HW Switch no change I'm not sure what the problem is. I might try dropping back to CAT 9.4 to see if it's a driver issue with 9.5.

[i found nemo]

HeamothoraxUK



if u read system busses 1,5,6 are the only ones 2.0 compatible.

[HeamothoraxUK]

Hi i found nemo,
Does this mean my board is not capable of running PCIe2 although it state it is? is there a way to change which bus the card is running on
because at the moment my score in this test is lower than your AGP score.Also in windows system info why is it assigning an IRQ of 4294967293 i thought this would have been a lower number meaning higher priority.

[i found nemo]

it's a physical limitation, i don't think u can force 2.0 on pci-e lanes

as for the windows irq, i really dunno, maybe in bios u can set other irq's for ur gfx and idk how to configure that bro

[EniGmA1987]

If you notice though, his first slot (bus 1) is a 2.0 x16 slot. So theoretically he should have a full 2.0 speed PCI-E slot on his motherboard that the graphcis card can use... Right? And if his card is in the primary slot, shouldnt that he the first bus on the board that uses the 2.0 speed?

[ahmad]

could use some help here. Intel e8400 mild bus oc. I figured it'd get more attention here than the intel one.

What the hell..

I thought my oc was stable, reduced it, and this thing still wont go through all the way to 1073741824 bytes test. It does the 536870812 one, and I get the "display driver recovered" message.

How many people have ran this test on C2D and had it go all the way through (where the window closes)??

Vista x64 4870, 9.5 cats.

[i found nemo]

If you notice though, his first slot (bus 1) is a 2.0 x16 slot. So theoretically he should have a full 2.0 speed PCI-E slot on his motherboard that the graphcis card can use... Right? And if his card is in the primary slot, shouldnt that he the first bus on the board that uses the 2.0 speed?

he's on bus 3 and 4 tho.....

[HeamothoraxUK]

Hi Hi i found nemo,
The board has a pci-e 2X16 on it i just can't figure out how to get the graphics card on that lane.I'll go into bios and and play with IRQ Assignment to see
if that sorts it out.Could do with some input from Lightman when he comes online as he has the same motherboard and chip to see if he's running pci-e 2 or not on his card
according to GPU-Z.Thanks for your help i found nemo.

[EniGmA1987]

hey ahmad, the test is very intensive. it doesnt matter if your FSB is overclocked at all. It will probably still crash near the end.

[jcool]

Can we get any Intel Core i7 X58 and Core 2 guys with ATI GPUs to run this program? This will let us know if it is a chipset or display driver issue, if those at all.

My queue again, I guess

4870X2, Server 08 x64 SP2, Cat. 9,4 - lower than stock QPI (5,6GT/s) and stock PCIe (100Mhz)



GPU --> CPU seems kinda low? Well at least it's consistent...

Took all of 2 minutes to get past the 512k test by the way.. I wonder why it seems to take 10-15 minutes for you guys?

[i found nemo]

just plain closes out for me after 128k

[EniGmA1987]

Took all of 2 minutes to get past the 512k test by the way.. I wonder why it seems to take 10-15 minutes for you guys?

It doesnt take that long on my system. My computer takes a minute or two to get past the 512 test. Who said their computer takes 10-15 minutes? They definitely have some sort of problem...

I would say that your GPU -> CPU bandwidth is slighter higher than AMD systems, but it is only slightly higher than our single card setups in bandwidth. Everyone that has done this test so far with dual cards or an X2 gets over 4GBps on the GPU -> CPU side of things.

[Mechromancer]

My queue again, I guess

4870X2, Server 08 x64 SP2, Cat. 9,4 - lower than stock QPI (5,6GT/s) and stock PCIe (100Mhz)



GPU --> CPU seems kinda low? Well at least it's consistent...

Took all of 2 minutes to get past the 512k test by the way.. I wonder why it seems to take 10-15 minutes for you guys?

The GPU to CPU bandwidth is the whole question. That is NOT PCIe x16 2.0, or even 1.0 bandwidth. Crossfire system run at the full speed of the PCIe x16 2.0 spec though.

All indications point to it being a software issue. Whether is is throttled deliberately or if it is a problem in the driver is yet to be figured out. It's not AMD chipsets since Intel users experience the same so I ruled that out. We've also had the same results spanning several different ATI GPUs. You tested with Cat 9.4 and most of us with 9.5. If it is a driver issue, it may be on purpose.

This has gotten serious to me. I may have to post this discussion on Beyond 3D! They have a couple of Obi Wan Kenobis over there that will eat this stuff up.