Looking for capture card. Limit GPU to 8 lanes?

fatryan

[H]ard|Gawd
Joined
Feb 19, 2004
Messages
1,402
I've been trying to find a capture card or HDMI recorder to record some shows. Ideally, I'd prefer a standalone, but the options are pretty limited. Id prefer 4k recording, but there's only like 3-4 devices on the market that can do that and they all have one issue or another. So I'll settle for 1080p recording with 4k input signal. That opens up my options, but I'm also looking for multi-channel audio recording (5.1) which cuts those options back down again.

The recording devices available offering the best in class features are capture cards for a desktop. I'm not opposed to using one, assuming I can use the dGPU or iGPU HDMI output as the HDMI input on the capture card. We only use streaming services or various kinds, so any show I can watch on TV i can also watch on the computer. My problem is that my boards specs only allow for one x16 device in the 16 lane slots on the board, and most top end capture cards are at least x4. There are a couple PCEix1 slots available on my board, but the x1 cards are pretty limited in features. So I'm wondering how bad it would be to just cut my GPU down to 8 lanes, giving me 8 lanes for video capture.

I'm not a gamer at all, and honestly my 2060KO is probably way overkill for my use case. It's heaviest workload is Plex transcoding (almost always just a single transcode). Other than that, it just pushes a 4k monitor and handles the day to day graphics tasks. No CADD, no video editing, no 3D graphics, etc. I have no idea how to gage this, so i was hoping someone else would be able to tell me whether or not I'll be fine knocking the GPU down to 8 lanes.
 
If that's all you do, you won't even notice a difference if you drop it to x8. Even in gaming the difference is relatively small if you aren't memory constrained.

This is for a 1080:
https://www.gamersnexus.net/guides/2488-pci-e-3-x8-vs-x16-performance-impact-on-gpus

Less than 1fps on average...
So no worries, drop it to x8 and use what you need!
OK thats good to hear.

I guess my next question is should I move the LSI HBA? The HBA is PCIe 3.0x8 and is currently in the bottom 16 lane slot (PCIe 3.0x4 PCH lanes). The open 16 lane slot is PCIe 3.0x8 CPU lanes. Honestly, I don't really know the use cases for PCH vs. CPU lanes, other than GPU should be on CPU. I only have a whopping 3x WD121KRYZ SATA HDDs on this SAS HBA, and I think its spec'd up to something like 128 drives...so I have to imagine I don't need 8 lanes, or probably even 4, right? But then again, if I get a x4 capture card, theres no way it can use 8 lanes anyway. So is there any particular reason to have the capture card on the x8 CPU lanes instead of the x4 PCH lanes?
 
3 SATA HHDs? Why are you not just using the motherboard SATA ports?

Some of this is hard to answer without further info. What motherboard and CPU are you using? I only have a win the card with coaxial input (OTA broadcast TV), so I'm not positive on the requirements for an HDMI device (like does it compress on board or does it transfer uncomopressed through pcie)? I don't imagine the HDDs tax the system to much. Just an approximate, but a single pcie 3.0 x1 is around 1GiB/s. Uncompressed 4k @ 30fps is around 800MiB/s... So even a single x1 should handle a single stream @ 4k 30hz... Or up to 4 1080p streams at 30hz. Similarly, your 3 HDDs (minus what the cache can transfer at) won't saturate a single x1 link. The PCH versus direct CPU links depends on what chipset your talking about, but in general PCH links have a few lanes from CPU then multiplexes those to multiple devices, so performance depends on how many things are in use at the same time (and depends on how many lanes and what speed it has to the CPU).
 
3 SATA HHDs? Why are you not just using the motherboard SATA ports?

Some of this is hard to answer without further info. What motherboard and CPU are you using? I only have a win the card with coaxial input (OTA broadcast TV), so I'm not positive on the requirements for an HDMI device (like does it compress on board or does it transfer uncomopressed through pcie)? I don't imagine the HDDs tax the system to much. Just an approximate, but a single pcie 3.0 x1 is around 1GiB/s. Uncompressed 4k @ 30fps is around 800MiB/s... So even a single x1 should handle a single stream @ 4k 30hz... Or up to 4 1080p streams at 30hz. Similarly, your 3 HDDs (minus what the cache can transfer at) won't saturate a single x1 link. The PCH versus direct CPU links depends on what chipset your talking about, but in general PCH links have a few lanes from CPU then multiplexes those to multiple devices, so performance depends on how many things are in use at the same time (and depends on how many lanes and what speed it has to the CPU).
Sorry, I assumed everyone could see my sig, but I guess if you're not on the website on a desktop browser then maybe you can't. My system specs are below.

MSI MPG Z390 Gaming Edge AC
i7-8700
EVGA RTX 2060 KO
Corsair Vengeance LPX 4x8GB 3000MHz DDR4 (Default clocks ATM)
Samsung Pro 970 512GB NVMe (Windows 10 Pro host + Win 10 Pro VM for work)
2x Samsung Evo 860 500GB (Storage/Plex Cache)
WD Purple 10TB (Surveillance Storage)
3x WD Gold 12TB (Plex Media)
WD Blue 6TB (Misc Storage)
Asus BW-16D1HT 4k burner

As you can probably guess from that list, my board's SATA ports are all filled, hence the HBA.

I've got 32GB RAM, but 8GB is dedicated to a RAMdisk for Plex transcoding, up to 8GB is dedicated to my Win10 VM (dynamic memory), and 2GB is dedicated to a Kali VM that I recently setup to mess around with (by no means is this a necessity). So I'm actually potentially looking at as little as 14GB available for Win10 on the host. I still feel like thats plenty though. It normally sits in the 30-45% usage range,; however I managed to pickup a trojan last week, and ever since then its sporadically been in the 40-75% range. Supposedly the trojan has been removed, so I dont know why its running that high still.

So what are CPU vs. PCH links used for? Are there restrictions on the types of cards to use on each one (assuming same # lanes)? Will one perform better for certain tasks than the other?
 
Sorry, I assumed everyone could see my sig, but I guess if you're not on the website on a desktop browser then maybe you can't. My system specs are below.

MSI MPG Z390 Gaming Edge AC
i7-8700
EVGA RTX 2060 KO
Corsair Vengeance LPX 4x8GB 3000MHz DDR4 (Default clocks ATM)
Samsung Pro 970 512GB NVMe (Windows 10 Pro host + Win 10 Pro VM for work)
2x Samsung Evo 860 500GB (Storage/Plex Cache)
WD Purple 10TB (Surveillance Storage)
3x WD Gold 12TB (Plex Media)
WD Blue 6TB (Misc Storage)
Asus BW-16D1HT 4k burner

As you can probably guess from that list, my board's SATA ports are all filled, hence the HBA.

I've got 32GB RAM, but 8GB is dedicated to a RAMdisk for Plex transcoding, up to 8GB is dedicated to my Win10 VM (dynamic memory), and 2GB is dedicated to a Kali VM that I recently setup to mess around with (by no means is this a necessity). So I'm actually potentially looking at as little as 14GB available for Win10 on the host. I still feel like thats plenty though. It normally sits in the 30-45% usage range,; however I managed to pickup a trojan last week, and ever since then its sporadically been in the 40-75% range. Supposedly the trojan has been removed, so I dont know why its running that high still.

So what are CPU vs. PCH links used for? Are there restrictions on the types of cards to use on each one (assuming same # lanes)? Will one perform better for certain tasks than the other?
Ahh, that explains it, I'm on my phone didn't know it hid signatures.

Yeah, 14GB is decent for every day uses.
Easiest way to think of it is like a USB hub. If you plug something straight into a port it has 100% of its speed available (CPU lanes). If you plug a huh into a port, then plug a device into it, you get almost full speed. If you however plug multiple devices in they have to all share than one USB link. This is similar how the PCH works.

Intel's documentation is slightly confusing, but the gist of it is: CPU supports 16 pcie 3.0 lanes directly for devices. Typically this is your GPU. But it can be split with 2 x8 or 1x8 with 2x4 so it adds up to 16 total. The PCH is connected to the CPU with DMI 3, but it's basically the same as pcie 3.0 x4, around 4GB/s total. This PCH divides the bandwidth up to 24 additional lanes for use. Keep in mind other peripherals may be attached to this as well sharing bandwidth (Ethernet, wifi, some USB ports, SATA ports, etc).

So, anything tied to the CPU directly gets 100% of its bandwidth, about 1GB/s per lane.. so a x16 connection gives about 16GB/s link. If you connect through the PCH and use x16 link you still have a max speed between PCH and CPU of 4GB/s, so your not getting full speed. Also, if other things are running you are sharing that 4GB/s with them.

Their are restrictions on a per MB bases on what it offers, but you can plug anything into anything and it'll work, just maybe not optimally. You can run a GPU on a pcie x16 slot (lots of miners did this) but if you're gaming it would probably be noticeable.

Hopefully this was clear enough about how it all interconnects, it can get really confusing especially since some MB offer different configs/options. Depending on what goes where can have a difference on performance. The DMI 3.0 link between CPU and PCH is the limit that can affect performance if you have many devices using it, but it's still 4GB/s (which is 16gbps) which is a good amount to try to saturate. This is why many people run x8 GPU if they are running an NVME.

Sidebar: This is where AMD differs, you get 20 pcie lanes from the CPU, so x16 for the GPU and x4 for nvme. They also (b550 and x570) have a pcie 4.0 x4 link to the chipset which gives them an 8GB/s link, so double the bandwidth of the Intel PCH. I'm not positive on the new 10 series stuff from Intel as I haven't looked into it, just wanted to point out the reason I wanted to know what you where running as the answer can be slightly different.


Edit:. Curiosity got the better of me. According to https://www.anandtech.com/show/15723/the-intel-z490-motherboard-overview, 10 series is still using DMI 3.0 and CPU still only has 16 lanes. Not sure if their next CPU bringing pcie 4.0 support will upgrade the link to the PCH or only the CPU lanes.
 
Last edited:
Ahh, that explains it, I'm on my phone didn't know it hid signatures.

Yeah, 14GB is decent for every day uses.
Easiest way to think of it is like a USB hub. If you plug something straight into a port it has 100% of its speed available (CPU lanes). If you plug a huh into a port, then plug a device into it, you get almost full speed. If you however plug multiple devices in they have to all share than one USB link. This is similar how the PCH works.

Intel's documentation is slightly confusing, but the gist of it is: CPU supports 16 pcie 3.0 lanes directly for devices. Typically this is your GPU. But it can be split with 2 x8 or 1x8 with 2x4 so it adds up to 16 total. The PCH is connected to the CPU with DMI 3, but it's basically the same as pcie 3.0 x4, around 4GB/s total. This PCH divides the bandwidth up to 24 additional lanes for use. Keep in mind other peripherals may be attached to this as well sharing bandwidth (Ethernet, wifi, some USB ports, SATA ports, etc).

So, anything tied to the CPU directly gets 100% of its bandwidth, about 1GB/s per lane.. so a x16 connection gives about 16GB/s link. If you connect through the PCH and use x16 link you still have a max speed between PCH and CPU of 4GB/s, so your not getting full speed. Also, if other things are running you are sharing that 4GB/s with them.

Their are restrictions on a per MB bases on what it offers, but you can plug anything into anything and it'll work, just maybe not optimally. You can run a GPU on a pcie x16 slot (lots of miners did this) but if you're gaming it would probably be noticeable.

Hopefully this was clear enough about how it all interconnects, it can get really confusing especially since some MB offer different configs/options. Depending on what goes where can have a difference on performance. The DMI 3.0 link between CPU and PCH is the limit that can affect performance if you have many devices using it, but it's still 4GB/s (which is 16gbps) which is a good amount to try to saturate. This is why many people run x8 GPU if they are running an NVME.

Sidebar: This is where AMD differs, you get 20 pcie lanes from the CPU, so x16 for the GPU and x4 for nvme. They also (b550 and x570) have a pcie 4.0 x4 link to the chipset which gives them an 8GB/s link, so double the bandwidth of the Intel PCH. I'm not positive on the new 10 series stuff from Intel as I haven't looked into it, just wanted to point out the reason I wanted to know what you where running as the answer can be slightly different.


Edit:. Curiosity got the better of me. According to https://www.anandtech.com/show/15723/the-intel-z490-motherboard-overview, 10 series is still using DMI 3.0 and CPU still only has 16 lanes. Not sure if their next CPU bringing pcie 4.0 support will upgrade the link to the PCH or only the CPU lanes.
I think i get most of that. So basically you can think of the direct CPU lanes sort of like VIP access to the CPU? Lol. They get their own box seats, while the rest of the cards are vying for standing room only?

I guess that means it's best to have the capture card on the cpu lanes to eliminate or minimize any chance of performance degradation or interruption during capture. If I'm going this capture card route, I'll likely be getting either the elgato 4k60 Pro or Avermedia live gamer 4k - both of which have on-board h264 & h265 encoders. I just need to confirm some technical aspects first, like multi channel audio capture.

You said something about running x8 with nvme ssd. Does nvme share the 16 cpu lanes too? My MB manual mentions M.2 sata drives sharing sata bandwidth, but it made no mention of using pcie cpu lanes. I only have the 1 nvme drive, because if I use the second m.2 slot for nvme, i lose 2 sata ports on the board. Obviously it's even worse with m.2 sata drives. Seeing as how these capture cards are only x4 anyway, can it still work with x8 gpu + x4 capture card + x4 nvme?
 
L
I think i get most of that. So basically you can think of the direct CPU lanes sort of like VIP access to the CPU? Lol. They get their own box seats, while the rest of the cards are vying for standing room only?

I guess that means it's best to have the capture card on the cpu lanes to eliminate or minimize any chance of performance degradation or interruption during capture. If I'm going this capture card route, I'll likely be getting either the elgato 4k60 Pro or Avermedia live gamer 4k - both of which have on-board h264 & h265 encoders. I just need to confirm some technical aspects first, like multi channel audio capture.

You said something about running x8 with nvme ssd. Does nvme share the 16 cpu lanes too? My MB manual mentions M.2 sata drives sharing sata bandwidth, but it made no mention of using pcie cpu lanes. I only have the 1 nvme drive, because if I use the second m.2 slot for nvme, i lose 2 sata ports on the board. Obviously it's even worse with m.2 sata drives. Seeing as how these capture cards are only x4 anyway, can it still work with x8 gpu + x4 capture card + x4 nvme?
Sorry if I made some confusion. Intel kind of ours you in a tight spot with NVME. You either just run the GPU @ x8 to give the nvme its own x4 lanes to the CPU or the nvme could be on the PCH and sharing bandwidth with everything else. This would be ok though if your motherboard allows you to do a 8/4/4 split. You can run GPU @ x8, nvme at x4 and the elgato 4k60 pro uses x4. This would be all your direct CPU lanes. Everything else would have to share the PCH. The onboard and encoder on the card should keep the bandwidth much lower than raw video. This means all your SATA drives and HBA (more SATA drives) would share bandwidth. More than likely you won't notice much difference unless you happen to be using all HDDs, network going all out, etc. I haven't looked at the manual for your specific board, I can probably check it in the morning to see what options it gives you.
 
L

Sorry if I made some confusion. Intel kind of ours you in a tight spot with NVME. You either just run the GPU @ x8 to give the nvme its own x4 lanes to the CPU or the nvme could be on the PCH and sharing bandwidth with everything else. This would be ok though if your motherboard allows you to do a 8/4/4 split. You can run GPU @ x8, nvme at x4 and the elgato 4k60 pro uses x4. This would be all your direct CPU lanes. Everything else would have to share the PCH. The onboard and encoder on the card should keep the bandwidth much lower than raw video. This means all your SATA drives and HBA (more SATA drives) would share bandwidth. More than likely you won't notice much difference unless you happen to be using all HDDs, network going all out, etc. I haven't looked at the manual for your specific board, I can probably check it in the morning to see what options it gives you.
See attached photos from my manual. It seems to indicate that only 2 of the 3 x16 slots are CPU slots. The rest, including the bottom x16, are PCH. If my nvme is taking up CPU lanes, can i even use that middle x16 CPU slot? Can it run in x4, or will it not work at all?

I'm still confused about the nvme part. Is this a different issue from M.2 nvme vs M.2 sata? I definitely have an M.2 nvme drive. Aside from me obviously confirming specs before purchase, all 6 of my sata ports work fine, which they wouldn't if it wasn't nvme. If this issue is just about the lanes used by the m.2 slot, I'm not sure where to find that info.

IMG_20200803_173035.jpg

IMG_20200803_235717.jpg

IMG_20200804_000038.jpg

IMG_20200803_235354.jpg
 
Here's a shot inside. Sorry, tight quarters, so i didn't feel like going through the hassle of removing glass.
IMG_20200803_173116.jpg
 
If you're running an NVME drive already, your GPU is only running at x8 and you didn't even know it. What should happen is GPU will run at x8, nvme will get x4 which leaves 4 CPU lanes remaining. You should be able to have those 4 available on the pcie slot you pointed out. (May be auto or manual setting not sure). So whatever is plugged in will be able to use at most 4 lanes. This also means your second nvme slot would get zero lanes of pcie and would have to be a SATA m.2. If I get a spare few minutes later I'll try to get on a desktop as it's much easier to type and view things. I can see what you have and what you want and try to give you an idea on how things can get plugged in and what downsides it could have. Seeing your MB manual you can see you lose some SATA ports depending on nvme drives as well.

According to the SATA table, if you have one NVME (pcie) and nothing in the second m.2 you have access to all 6 SATA ports. If you were to plug something else into the second m.2 you would lose some of your SATA ports.

I'll try to find some more info a bit later if I can.
 
Last edited:
Just realized I was mistaken on how your board was layed out. Your NVME is connected to PCH, not CPU. So the slot you are looking at will have x8 available, the nvme, SATA drives, and LSI HBA card will all be sharing PCH bandwidth.

My apologies, some boards have nvme to CPU, got myself confused. Looking at everything I would just drop the capture card in that free slot and split x8/x8 and it should all work just fine.
 
Just realized I was mistaken on how your board was layed out. Your NVME is connected to PCH, not CPU. So the slot you are looking at will have x8 available, the nvme, SATA drives, and LSI HBA card will all be sharing PCH bandwidth.

My apologies, some boards have nvme to CPU, got myself confused. Looking at everything I would just drop the capture card in that free slot and split x8/x8 and it should all work just fine.
How can you tell if NVMe uses CPU lanes or PCH lanes?
 
How can you tell if NVMe uses CPU lanes or PCH lanes?
Your second diagram, the Block Diagram shows the 2 m.2 connected to the PCH. 1 is directly connected (why a single m.2 as nvme doesn't affect SATA ports), the second goes through a switch that is shared (hence you lose some SATA ports if this is in use).
 
Your second diagram, the Block Diagram shows the 2 m.2 connected to the PCH. 1 is directly connected (why a single m.2 as nvme doesn't affect SATA ports), the second goes through a switch that is shared (hence you lose some SATA ports if this is in use).
Ok thanks. I'm still working out the details on a card. I need to have one capable of capturing 5.1ch audio. Elgato just got back to me to tell me none of their devices can do this. Avermedia's LG4k claims it can, but I've read conflicting reports on that. If i can't find one that captures 5.1, I'm going to be rethinking this whole thing.
 
Ok thanks. I'm still working out the details on a card. I need to have one capable of capturing 5.1ch audio. Elgato just got back to me to tell me none of their devices can do this. Avermedia's LG4k claims it can, but I've read conflicting reports on that. If i can't find one that captures 5.1, I'm going to be rethinking this whole thing.
Sounds good. Hopefully you at least learned some stuff, lol. Good luck finding what your looking for.
 
Sounds good. Hopefully you at least learned some stuff, lol. Good luck finding what your looking for.
Yeah, thanks. I was always a little confused about the pcei thing. Now that I'm filling up slots, i need to understand it lol
 
I ordered the Avermedia Live Gamer 4k, though it's backordered for a week or 2 :-/
 
Keep me updated, but I don't foresee any issues.
Will do. Just got an email from B&H saying it's shipping direct from avermedia and is estimated to ship out in 7-10 days. B&H was one of the only retailers that even had the thing for sale. Many others just said 'out of stock', didn't even have it listed, or had it at some absurd 50-100% markup. This thing is crazy popular now from what I understand.
 
Got the capture card and an HDFury Vertex2 in today. I can capture video from Netflix or Amazon Prime now, but the resulting video is all fucked up. I still need to figure out all these settings. Its pretty complicated with the EDID data. I'm hoping the recording issues are simply due to the EDID. I tested it out recording to my RAMDisk, so write speed should be no issue. Crystal Disk puts my RAMDisk at twice the speed of my Samsung 970 Pro NVMe, so thats fast as shit. My 2060KO should have no problem encoding h264 or h265 as well. Its late and Im tired of messing with it, so I'll have to figure it out later. Initial setup was very easy though.
 
Got the capture card and an HDFury Vertex2 in today. I can capture video from Netflix or Amazon Prime now, but the resulting video is all fucked up. I still need to figure out all these settings. Its pretty complicated with the EDID data. I'm hoping the recording issues are simply due to the EDID. I tested it out recording to my RAMDisk, so write speed should be no issue. Crystal Disk puts my RAMDisk at twice the speed of my Samsung 970 Pro NVMe, so thats fast as shit. My 2060KO should have no problem encoding h264 or h265 as well. Its late and Im tired of messing with it, so I'll have to figure it out later. Initial setup was very easy though.
Yeah, could it possibly be some sort of protection going on as your hardware should be up the task?
 
Yeah, could it possibly be some sort of protection going on as your hardware should be up the task?
That's what the vertex2 is for. It completely strips all hdcp, up to 2.2. and it send back a signal with the device model of my choice (the EDID). I currently have it set to send back the EDID for a generic 4k60 hdr 12bit 444 screen. But there are dozens of EDIDs to pick from, so maybe one will work better.
 
Back
Top