Does the CPU affect the speed of the GPU?

[Meki01]

Hello! I have a main board with 8 PCI-E slots. A GeForce 730-4G is in each slot. Sometimes I have processing times of over 2 days. Could this be due to the slow CPU? Unfortunately, the board only has a Celeron 847 @ 1.1 GHz. If I just replace a GeForce 730 with a Radeon RX550, which should be much faster, then I have a processing time of up to 34 days on this one card! The Celeron is constantly at its limit. I suspect that's the problem. Would it all work better with a faster processor (i7)? I deactivated folding with CPU. Thanks for your info. Kind regards, M

[JimboPalmer]

[I am just a user like you, I have never had a PC with 8 GPUs. this is rumor]

F@H attempts to reserve a CPU thread for each GPU to move data for it across the PCI-E bus, so ideally you would have a 8 threaded CPU.

https://ark.intel.com/content/www/us/en ... 0-ghz.html

Your CPU only has 2 threads, so is overwhelmed by 8 GPUs. A 'faster' CPU will not help as much as a multi-core CPU.

Your existing CPU is described as 'Mobile' which may mean upgrading the CPU on the existing Motherboard is not possible/difficult.

[iero]

That is an old ooooold CPU. Apart from the 1 thread per GPU, i would guess that a motherboard with 8 pci-e splots must be a mining one, thus only limited to 1 lane per slot?

[Meki01]

Hi! Thanks for the quick repley! Yes, it's a mining board I get another one where I can swap the CPU. I have an i7 with 4 cores and 8 treads. Should it go better with that?

[Joe_H]

It should be less CPU bound with the i7. You may still run into some slowdowns from PCIe bus speeds. If you go with that, try starting up one GPU at a a time to see where overall processing stops increasing.

[Meki01]

Thanks, I'll try that and report back here as soon as I get the other board.

[Meki01]

Here is a photo of the board (for the unbelievers )

[gunnarre]

What is the model number of that board?
What is the PCIe width of the slots? Are they all x1? If so, even replacing the CPU might not be worth it - though the GT 730 is a small GPU so it might still work: viewtopic.php?p=338263#p338263

I've already replied multiple times to this question.
It's up to you to find and accept those settings, or just go against it.
In most cases:
PCIE 1.0 x4 / 2.0 x2 / 3.0 x1 is not recommended for folding. You could fold on it, but will experience serious PPD penalties on budget to mid range GPUs.
PCIE 1.1 x16 / 2.0 x8 / 3.0 x4, good enough for GTX 1600 series GPUs under Linux (an RTX 2060 gets 975k PPD in Linux, which is slower than a Core 21 WU), or up to a GTX 1060 in Windows.
PCIE 2.0 x16 / 3.0 x8 is good enough for up to a 2080Ti under both Linux and Windows. Estimated upcoming 5000+ shader/core gpus might be limited in Windows, but should fold fine in Linux.
PCIE 3.0 x16, there's currently no GPU that would exceed this PCIE bandwidth, but if there were, it would be a GPU with 8000 cores or more.

[Meki01]

The board is called BTC-D37. Unfortunately I don't have a manual ...
Technical specifications:
Memory: 1 * DDR3 sodimm socket
Storage: 1 * SATA3.0 + 1 * SATA2.0 + 1 * m SATA2.0
USB: 4 * USB2.0 + 2 * front USB2.0 pin
Multifunction card: ITE IT8772E
Case size: 600 * 360 * 180mm
CPU: Intel Celeron CPU 847
8 * 6PIN PCIEx16 graphics card slot

The distance between the slots is 55 mm

Each card now works with around 4,200 PPD and requires around 14 hours for a WU.

[gunnarre]

Those are all PCIe 2.0 x1 slots according to the AliExpress listing. The slots accept x16 cards physically, but there are only connections for a x1 width connection.
Folding performance is likely to be very limited. You can test that by running one of the cards in your regular PC and compare the performance. There is some variability in performance between projects and Work Units, but if you get dramatically better performance on a regular motherboard, the CPU isn't the only issue.
I have a hunch that even a very old PC motherboard (like 10 years old) could run a couple of those cards faster than the mining board can.

[Meki01]

Thanks for the info, I'll try a card in a "normal" system and compare the values.

[iero]

An RX 480 runs, for all intents and purposes, unsaturated on an PCIe 3.0 x8 connection, but could benefit ever so slightly from a x16 connection. In perspective, a PCIe 2.0 x1 connection translates to PCIe 3.0 x0.5 speeds, speeds that are unbelievably low for a GPU.

[Jupis_folder]

I must do something wrong. I use a PCIe riser with PCIe 2.0 x1->x16 (x16 only mechanical), and have no performance penalties on a GTX1050Ti.

[bruce]

I must do something wrong. I use a PCIe riser with PCIe 2.0 x1->x16 (x16 only mechanical), and have no performance penalties on a GTX1050Ti.

For a very small protein, the performance penalty of 1x will be small enough that you probably won't notice. For a larger protein (with more atoms) the GTX1050ti will definitely be choked for data. What project(s) are being assigned and how many points-per-day are you getting. I run several 1050Ti's in 16x slots with at least 1 free CPU thread per GPU.

[bruce]

The GT 730 4 GB is a Fermi device which means it only supports OpenCL 1.1 That will be an additional limitation as it probably only supports OpenCL and not CUDA. (CUDA is somewhat faster depending on the project.) What projects are being assigned and what PPD are you getting. Are you using a passkey?