Rtx 40xx cards should get workunits. Are they?

Rtx 40xx cards should get workunits. Are they?

RTX 40 cards wont work with the current applications unless you build it for cuda 11.8 or newer.

gencodes for Ada are not even available to add in earlier cuda versions.

---

According to this cuda11.0 should be enough:
https://docs.nvidia.com/cuda/ada-compatibility-guide/index.html

So do they get workunits and crash or they don't even get it?

gdf

They get but crash after a few seconds https://www.gpugrid.net/show_host_detail.php?hostid=605125

My 4080 seems to be running the Python app OK.

My 4080 seems to be running the Python app OK.

yes it will work for Python and ATM (which is also python based)

doesnt work for acemd3 since it has an architecture check built into the code. they either need to remove that check, or rebuild the app with CUDA 11.8 or CUDA 12

---

ATMbeta still crashing on 4070Ti

ATM betas seem to crash after a few mins on my 4090 every time

First one finishing successfully on RTX 4090:
http://www.gpugrid.net/workunit.php?wuid=27474364

First one finishing successfully on RTX 4090:
http://www.gpugrid.net/workunit.php?wuid=27474364

40-series cards have always worked for the Python based "PythonGPU" and "ATM" tasks.

ACEMD3 is the app that will not work with 40-series.

---

ATM betas seem to crash after a few mins on my 4090 every time

ATMbeta still crashing on 4070Ti

Same here with a RTX 4080.

I'm thinking of replacing my old graphics cards with RTX 4060s.. the lower power consumption should be great for GPUGrid crunching. Does any of the work units work on the Ada Lovelace generation of GPUs yet?

I'm thinking of replacing my old graphics cards with RTX 4060s.. the lower power consumption should be great for GPUGrid crunching. Does any of the work units work on the Ada Lovelace generation of GPUs yet?

acemd3 wont work on 40-series until they update the application to allow them.

ATM/ATMbeta and PythonGPU do work on 40-series (at least on Linux)

---

acemd3 wont work on 40-series until they update the application to allow them.
ATM/ATMbeta and PythonGPU do work on 40-series (at least on Linux)

Thanks. I'm running on Windows. Before taking the plunge just wanted to make sure there's work units available. If not there's no rush - I can wait a few weeks more to see if RTX4060 prices will come down first.

Seems that the project has had only acemd3 task for distribution for the past couple of weeks.

I think you can delay your 4060 purchase a while longer.

I think you can delay your 4060 purchase a while longer.

Thanks for info on the 40x0 WU support. Will hold off buying a RTX 4060 for another month and see the situation again.

Prices of the Ti range are already being discounted. The non-Ti will likely drop in price soon too.

How much GPU memory does each task require? I have 8GB of memory on each of my GPUs.

---

How much GPU memory does each task require? I have 8GB of memory on each of my GPUs.

I'm running ATMBeta now on a RTX4060 and a RTX4060 Ti (2 separate machines). GPUZ reports only slightly over 1GB of graphics card memory being used, and the memory controller load is in the high teens (fluctuates around an average figure of 17%).

It seems to be compute bottlenecked as the GPU load is in the mid 90% (fluctuats around an average figure of 94%). So it would appear that 8GB is plenty for GPUGrid purposes.

Hello,

is there a solution out there to get the project running again on BOINC (Windows) with the RTX Cards? I dont get any worksloads not even the project initialisation does anything. The scheduler always is on "communication delayed" state.

Best regards,

Artemis

---

Likely nothing to do with RTX40 cards. Problem lies with Boinc and scheduler connection issues.

Set http_debug in the logging from the Manager Options menu and look through the http_debug output for the reason connections can't be made.

Likely something to do with SSL and certificates for security.