8K pro in W680 lag in 8x, no lag in 4x

Hello!

We are using (trying to use ) declink 8K PRO cards in W680 motherboards. ( 14900kf, 64GB 5600Mhz, tuf 4090 or inno3d 4070ti x3 ). Our problem is whether or not what gpu we use (also tried 20xx, 30xx) the input or output is lagging so hardly if the card placed in PCIEX16(g5)_1 (16x) or PCIEX16(g5)_2 (8x) slot.

In the g5 slots its impossible to let a single 3G feed to work normally independently which slot we are using and what gpu we are using. We checked the pcie link speeds in several apps like hw info and everything looks normal. The systems runs in 8x / 8x with the correct gen for the cards.

This is not a single machine. We have 4, all of them behave the same.

And the strangest thing is: If we put the card in PCIEX16(g3)_1 (4x) or PCIEX16(g3)_2 (4x) which are moreover chipset lanes there is no lag at all. There isnt a single frame dropped even its receiving 2x 2160p60 and sending 2x 1080p60. (this is the highest overall bandwidth we had the possibility to try currently). In the past, with other boards we always had lagging problems in 4x slots especially in chipset lane slots and never used chipset lane slots for years, thats why we even not tried those for the first.

We have checked the bios and tried all pci-e related settings and default settings also without a tiny difference achived.

Softwares: resolume, obs, notch, unreal ....

Does anybody has a sure explanation?

Thanks in Advance!

Hi who am I talking to?,

I think you should tell uns the mainboard model. Only mentioning the chipset W680 is too few information since every mainboard vendor may connect device differently. With the exact name of board we could see the schemantic of that board and maybe find out the cause of your issue.

Best regards
Floh

Yeah you're right, sorry! Forgot to mention the model.

Mainboard is: Asus w680-ace

Thanx. I don't think there is special about the main board.

Do you mean ‘lag’ as in stuttering? If so, doublecheck the task manager. Also doublecheck if your drivers are all up to date. Unfortunately I don't think I can help you about the issue and yes, this is weird. Maybe contact Blackmagic for support.

Best regards
Floh

The difference will be how those slots are wired.

The 4x slots go via the chipset. This is usually the worse option, but not always.

The 8x slot is connected directly to the cpu. This is usually better, but when using this slot your GPU is now also connected at 8x instead of 16x.
So my guess would be that when doing this it is causing the GPU to be the bottleneck, with it having too little bandwidth to do its work.
Keep in mind that when you are doing video input/output, you will probably want the frames to go to or originate from the gpu. So if the decklink needs 8 lanes to operate, then it will 'consume' 8 lanes of the gpu bandwidth. Which in this setup could result in there being very little or no bandwidth left for the other content/textures needed

This is cannot be the case, because its gen5 running @ gen4 x8 (4090), and the rendering itself is very fast and smooth. While the 8k pro is lagging hardly on inputs and outputs also the outputs on 4090 are perfect in the same time, and all output monitor also perfect on gui anyway. gen4 x8 is way enough for example to texture a simple test card in resolume and transfer 1080p60 to a secondary card. Or its way enough to capture a single 1080p60 and transfer it to cpu / gpu, but it not works as it should, dropping every second frames.

FYI:
gen4 x8 = 15.75GBytes / sec = 126gbps (4090)
gen3 x8 = 7.87GBytes / sec = 63gbps (8k pro)

If its not enough for texturing a simple 3g signal, than what is?

The difference will be how those slots are wired.

The 4x slots go via the chipset. This is usually the worse option, but not always.

The 8x slot is connected directly to the cpu. This is usually better, but when using this slot your GPU is now also connected at 8x instead of 16x.
So my guess would be that when doing this it is causing the GPU to be the bottleneck, with it having too little bandwidth to do its work.
Keep in mind that when you are doing video input/output, you will probably want the frames to go to or originate from the gpu. So if the decklink needs 8 lanes to operate, then it will 'consume' 8 lanes of the gpu bandwidth. Which in this setup could result in there being very little or no bandwidth left for the other content/textures needed