Is anyone running one or more Nvidia RTX 4070 Ti Super GPU ?

Back on topic, here's something that might satisfy your curiosity. Note it's not yet stabilized on RAC.

It's the PNY XLR8 version of the card, with a gigantic 3.5-slot cooler that kept the GPU below 50 degC when limited to 180W with nvidia-smi. It's incredibly efficient at that temperature and wattage, doing about 1.4M credits of mixed 3xO3AS/2xBRP7 per day with that power target. The whole system is pulling 290W, with the CPU limited to 65W and doing Prime95 at the same time. It is actually slightly cheaper than the base-clock MSI offering in local area, presumably due to comparative lack of brand recognition.

It's a lovely piece of kit that did well enough in games too, except that it blocked the second PCIe x16(4.0 x2 electrical) slot on the motherboard out of use. No more expandability there. It's also somewhat noisy, even at minimum fan RPM while running. Someone elsewhere took issue with the fact that the minimum the fans could be set to run is 30% PWM for some reason, and that actual measured PWM is even greater.

And yes, BOINC can definitely tell the difference.

Thank you!

And a bit of an update: Increased GPU utilization factor back to 1 for all apps, when 1x O3AS consumed more than 2GB of VRAM and 3 started getting in the way of other things I do. They trip over each other in GPU usage in any case. BRP7 does not appear to scale very far either, even though it only takes 0.7GB. BRP7 would also slow down O3AS something grand, to 1/10 the speed if they happened to run together, for some reason. Some 30% loss of throughput expected, but there should also be even less power consumption. BOINC really should have a better way of managing GPU use, for applications that require GPU only part of the time.

Strangely, running just one app at a time produces coil whine, while running two does not audibly do. I may go back to 0.5 for that reason alone.

The card itself seemed about comparable to an RTX 3090, if not even faster. Quite amazing developments for a single generation. 4080 / 4080 SUPER should do even better, being the same silicon and more-or-less the same board and the same overhead, with more computational resources enabled. Unfortunately, price increase outstrips performance increment there.

Tom,

Thanks for the info. I think I would look into it. A brief skim of MPS doc has left me wondering why it was not enabled by default for supporting setups. Need to read further.