4

I have a PCI USB riser like this one: PCI x1 USB to PCI x16 riser

I have 2 identical video cards with one plugged directly into the motherboard and one plugged in via the PCI riser. I am seeing a hash rate of 290 KH/s on the one directly connected to the motherboard and 210 KH/s on the one connected via the riser. The riser is connected to the PSU so it is powered. There is no need to short the PCI slot. I have tried this riser in both a PCI x1 slot and PCI x16 slot with the same results. The video cards are both GTX 580.

Could the added latency of going through a USB wire account for this decrease in hash rate?

Using another video card (GTX 750 Ti FTW), I was able to confirm a decrease of about 100 KH/s while connected to the PCI USB riser. I have tried 2 separate risers with 2 separate video cards and every time a PCI USB riser is involved, my hash rate is lower than when connected directly to the motherboard. The motherboard is a Gigabyte GA-X58A-UD3R. The BIOS has been updated and I only see BIOS options to change the PCI Express Frequency and PCI Express Clock Drive.

I am having a hard time finding an answer online and I need experts, please help. If these USB risers decrease the hashing power then I'm not sure why people want them. I want them for the added distance I need between my motherboard and GPUS in my setup but perhaps I should alter my setup to allow for the ribbon cable riser version. I appreciate your recommendations. Could there be a setting that accounts for this latency and keeps an optimal hashing power?

  • 2
    Swap the cards, see if the difference is still there. Eliminate variables. The latency in the cable is negligible anyway. – user13413 Feb 27 '14 at 2:09
  • 1
    That doesn't make much sense to me, I can't explain it. – user13413 Feb 28 '14 at 3:16
  • 1
    What version of USB are the port and the device on? USB is not just a hardware connector specification, it comes with its own layer-1 protocol - a serial protocol, by the way - so the rate at which it transmits info. back and forth would definitely be affected depending on the version (I wouldn't use anything but 3) – Joe Pineda Mar 16 '14 at 13:20
  • 1
    Another question, at what difficulty are you mining? You could tweak your miner params (increasing diff?) so the cards do as much work as possible on their own without relying too much on the motherboard to provide them data - that way the USB comm. bottleneck would be avoided. – Joe Pineda Mar 16 '14 at 13:30
  • 1
    yes. You are transferring data through a device with smaller bandwith. Also today USB3 should be used. A longer thread is here: cryptomining-blog.com/tag/pci-e-usb-riser – pebwindkraft Feb 25 '18 at 9:06
4

The general consensus is like Robert said: there SHOULDN'T be any difference, but there is!

Having the same problem here!

I'm running Windows 8.1 Pro (64-bit) with five (5) of the new GeForce 750 Ti cards by EVGA (FTW model).

They hash at 320 KH/s when connected to the PCIe 16x slot, but get capped/maxed at 295 KH/s when connected using 1x to 16x riser cables.

  • All riser cable molex plugs are connected to power supply.

  • All cards have their PCIe 6-pin power plugs connected as well.

  • Updated BIOS of cards to take up 65 watts max (or even up to 100 W) to no avail.

Power supply: 1000 watts (more than enough)

Motherboard: BIOSTAR H81S2 (made for mining) ver. 6 - latest BIOS

UPDATE: I thought the culprit was the motherboard, so I bought the ASRock H81 Pro BTC (made for mining) motherboard. SAME EXACT STORY!

I was about to try a USB riser cable instead, but I see that the OP (Scott) has the same issue with those as well.

Now I'm leaning towards a driver or OS issue, but I'm quite certain it's because of using the 1x slots, though, AMD cards are hashing at much faster rates! So that's interesting. What's going on here?!

I will try running on Linux.

I contacted BIO-STAR for tech support and they'll get back to me. Their R&D department is looking into this.

I already tried setting the PCI speeds for the slots, in the BIOS, to Gen2, manually (instead of Auto), and that didn't help. (Because I found someone on the internet with a similar issue and they said this fixed it for them, but it's not exactly this issue they've had).

In the BIOS I tried changing PCI Latency Timer up to 248 PCI Bus Clocks - didn't help!

In the BIOS I tried changing the Maximum Payload & Maximum Read Request to 4096 bytes - didn't help!

I've disabled redundant services and components in Windows - didn't help!

I ran the DPC Latency Checker tool and things seem stable so that's not the issue either.

Interesting to note:

While overclocking above a certain MHz for the core or memory, through the 1x riser, once the hash rate reaches 295 KH/s, it does not go any higher no matter how much you overclock, however, the overclocking utility of choice (EVGA Precision X, ASUS GPU Tweak, MSI Afterburner) will confirm the higher clock rates running on the cards. If connected through the full PCIe 16x slot, this 295 KH/s limit is non-existent, and the more you overclock past that same amount of MHz as before, the higher the hash rate.

Doesn't matter which version of CUDAminer is being used.

UPDATE: Looks like it's a CUDAminer issue:

http://www.overclock.net/t/1468166/gtx-750-ti-mining/190#post_21890197

Cudaminer apparently has a high PCIe bandwidth utilization...thus when going from x16 mode to anything lower, as you get closer to x1 mode (risers or not), you're going to see degradation of performance (10-30 kH/s for 750 Ti in my experience). Cudaminer author is apparently going to be working to fix this ASAP since he'll soon be operating a 750 Ti rig with riser assemblies.

  • Excellent debugging, thank you. It would seem that increasing the PCI latency timer via the BIOS would help, too bad. – Scott Mar 5 '14 at 4:08
2

Nvidia +cuda miner need high PCIe bandwidth while Radeon/AMD+cgminer use very small PCIe bandwidth so using 1x to 16x riser doesn't give any speed reduction

  • 1
    But I am getting lower rate (VEGA 64) when using riser with xmr-stak-amd – Hassan Tareq Dec 23 '17 at 14:37
2

It would seem that the GPU cards plugged into a motherboard will transfer on the PCI-e x16 bus, but after plugged into a extension header will only be able use the PCI-e x1 bus. May make a difference.

1

Basically the effects would be negligible, so no. The risers would on a very tiny scale slow down the transmission of block information between the card and the motherboard however the transmission of data is not the bottle-neck when it comes to 'mining' since the actual size of the data files is quite small on the order of Mega bytes in size. The bottle neck in mining speed comes from the cards ability brute force calculate in-order to decrypt the block-chain. This is another reason 16x pcie card slots are not any faster than 1x slots because the rating refers to data transmission and is only really important to video intensive use, like gammers.

  • Even if the motherboard is first going through a USB controller/cable and then to the video card? – Scott Feb 27 '14 at 16:53
  • 1
    I would say that is negligible to, you are correct to say it will slow things down just not anything to worry about. – HappyBob Feb 1 '15 at 9:26
1

2/24/18: Having the same issue - I use an ASUS B250 MINING EXPERT with GeIL Pristine 4GB and Intel Core i3-6100T. 2: RAIDMAX RX-1000AP-S 1000W and 1: Corsair RM450

The RM450 runs the motherboard, SSD while the 2: 1000W runs 10: Ver006C Mining Dedicated PCIe Riser Card and the 2: MSI RX480 ARMOR 8GB, 2: XFX RX480 CORE 4GB, 1: MSI RX 570 ARMOR 8GB, 5: MSI RX580 ARMOR 8GB. The RIG is stable no doubt, performs great!

"BUT I know it can be better!"

The same problem occurs with loss in hash rate about 80-90 H/s per card. All i did was move from the PCIe 16x long slot on the old computer i use for bios modding to the riser cards on my main rig with the ASUS B250 Mining Expert.

Example: After Bios MOD - Mining XLC Leviar Coins, I am currently getting 816 H/s with one RX580. I used this same RX580 in my ECS P67H2-A2 BlackEdition running in the PCIe 16x Long Slot. Started EthOS / SGminer on a thumb drive. The RX580 solo in the PCIe 16x Long Slot was hashing it out at 905-915 H/s.!

****I do not understand the 80-90 h/s difference from the PCIe16x Long slot vs Ver006C Mining PCIe Riser Card** There is definitely a loss of hash rate and it is very annoying. With 10 Cards- I am losing almost 1000 H/s Like others say, no mater what setting you change it is throttled like a bottle neck. I hope someone can provide some insight on this issue.**

Makes you want to run 10 solo rigs to gain back my 1000H/s loss.

Thanks for reading...

  • Do you think there is any difference with the USB 2.0 and the USB 3.0 risers? or does the bottleneck exist elsewhere in the system? And do you see some of the GPUs hashing significantly higher than others? Thanks for sharing your experience. – Scott Feb 25 '18 at 16:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.