NVidia + AMD iGPU -- another data point

[pcwolf]

RTX 4070 Super + RTX 4060ti + AMD 7900 "Raphael" integrated GPU = 10,000,000 +/- PPD

RTX 4074 Super + RTX 4060ti = 16,000,000 +/- PPD

ASRock B650e Taichi Lite running Manjaro Linux. Windows11 produces significantly less (~20%) on identical dual-boot hardware.

I chose the 4060ti over another 4070+ card because the 4060ti is competitive on 165 watt TPD limit.

I thought the integrated GPU would be a little lagniappe but turns out it must be stealing some capacity from the 4070 Super.
I can break out the PPD per GPU if anyone would like a more direct comparison.
Total system draw is less than 500 watts. I run on AMD Eco mode in UEFI to limit processor heat. It runs 24x7x365 at 45-50C
System memory seems not to matter. I have run 16MB to 32MB without noticing any change in output.

[bikeaddict]

4060 Ti on dedicated Fedora Linux machine calculated from logs with script.

Code: Select all

2024-03-10: 7,953,234
2024-03-11: 7,281,703
2024-03-12: 6,783,035
2024-03-13: 7,067,035
2024-03-14: 7,588,339
2024-03-15: 7,896,240
2024-03-16: 7,470,053
2024-03-17: 7,303,260
2024-03-18: 6,295,545
2024-03-19: 7,187,024
2024-03-20: 7,175,860
2024-03-21: 7,402,360
2024-03-22: 6,288,899
2024-03-23: 7,414,960
2024-03-24: 7,411,506
2024-03-25: 7,438,633
2024-03-26: 7,020,760
2024-03-27: 7,323,833
2024-03-28: 7,815,143
2024-03-29: 7,281,424
2024-03-30: 6,809,562
2024-03-31: 7,329,328
2024-04-01: 6,922,357
2024-04-02: 7,638,836
2024-04-03: 6,399,712
2024-04-04: 8,187,872

Same for 4070 Super. Dip on 3/30 was from an overnight thunderstorm power blink when it wasn't plugged into a UPS.

Code: Select all

2024-03-10: 15,257,239
2024-03-11: 14,753,076
2024-03-12: 14,926,872
2024-03-13: 14,703,849
2024-03-14: 15,106,469
2024-03-15: 15,120,796
2024-03-16: 15,278,248
2024-03-17: 14,955,264
2024-03-18: 14,495,390
2024-03-19: 15,271,976
2024-03-20: 15,098,547
2024-03-21: 14,704,636
2024-03-22: 14,748,969
2024-03-23: 14,287,746
2024-03-24: 14,407,845
2024-03-25: 14,366,444
2024-03-26: 14,860,432
2024-03-27: 14,682,758
2024-03-28: 14,358,010
2024-03-29: 14,176,235
2024-03-30: 12,565,582
2024-03-31: 15,023,009
2024-04-01: 14,848,257
2024-04-02: 14,478,199
2024-04-03: 14,949,413
2024-04-04: 14,925,074

Your two GPUs should be totaling 20-22M PPD on Linux. My totals are a little higher than the LAR Systems database.

[pcwolf]

Current status:
4060ti - 7,448,102
4070S - 13,287,070

bikeaddict are you running both in one chassis, or separate machines?
What CPU chip, please

[bikeaddict]

They are in dedicated machines. The 4060 Ti is with an i7-13700K with the FahCore process affinity set to use an E-core with taskset. The 4070 Super is with an old i7-7700K. There is no CPU folding running. But these two GPUs were swapped between the machines about a month ago and had similar PPD before.

[muziqaz]

OK, let's start from the beginning:
What is actully folding:
I can see that 4070 and 4060ti are folding, and assume they are installed in the same system, which has 7900x, with enabled AMD iGPU. Is iGPU folding (not that it would, since there isn't many projects which are enabled for this iGPU), is CPU folding alongside. How many cores is left for both GPUs?
How are the pcie slots arranges and how are pcie slots assigned? Are you sure that one (or both) of your GPUs are running on reduced pcie lanes due to both of them installed? remember, mid range or low end motherboards cut corners with various pcie/usb/sata related matters.
nVidia drivers tend to ask for full high clocked core for each of the GPUs, this is especially true with Windows, however Linux is not immune, since nVidia drivers for Linux are paradoxically crap.
Why do you have iGPU enabled? For a moment it will not bring any benefits to FAH project, wince it was developed to display a desktop, that's it. Later in the future when HIP is out, we might revisit it as more viable product, but on OpenCL - no. Unless you need that iGPU for something specific, I would advise you to disable it in BIOS, that will also save you quite a bit of power usage and heat output.
Please note, that PPD is very volatile on higher end (modern) GPUs. It varies a lot from project to project, swings might be in millions of PPD. It only takes for TPF to shift by 2-3 seconds and you have million of PPD gone.

[pcwolf]

Yes, all folding taking place on one chassis. ASRock B650E Taichi Lite motherboard and Ryzen 7900 (no X). Manjaro Linux

The two cards are installed on PCI slots 1 and 2. Specification is:
CPU: 1 x PCIe 5.0 x16 Slot (PCIE1), supports x16 mode
Chipset: 1 x PCIe 4.0 x16 Slot (PCIE2), supports x4 mode

I only attempted CPU with F@H recommended number of cores (twenty, I think) and the iGPU Raphael to test for combined output.
Each one of these delivered on the order of 7 - 900,000 PPD while showing a penalty to the 4070 Super in the first slot PPD of roughly three times those amounts.

My experience confirms your advice: disabling CPU and iGPU to focus on the RTX cards delivers best result.

I dual boot with Windows 11. The F@H beta 8 is a simplified installation on Windows; beta 8 is not yet available on Manjaro.
I am testing combined output with the "pooling" feature of all folding resources that I understand the beta 8 offers.
Results are not clear yet with my limited time spent in Windows. For now the machine runs Manjaro for most of the month.

By the way, in order to be most economical in terms of energy usage, the Ryzen is set to ECO mode in UEFI.
4070 Super draws ~200w out of rated 220w TDP
4060ti draws ~140w out of rated 165w TDP

[muziqaz]

Your primary GPU airflow is severely restricted by secondary GPU being in second slot which is quite close by.
On windows, you can check quite a few things with GPU Z (what pcie speeds each GPU is using, and loads of each GPU as well as temps, and voltages). Have you tried playing around with BIOS settings if it is possible to run both cards on pcie x8 speeds.