Nvidia Titan V

Ubuntu 20.04.5

BOINC 7.19.0 stand-alone compile

Platform:
Asus P9X79-E WS
Xeon E5-1680v2
32 GB DDR3-1600 ECC UDIMM

GPUs are watercooled, temps stay about 45-48C

Runtimes average about 30 seconds running 5 at a time.

---

I'm almost ashamed to post this -- seeing/reading about all these beautiful and high performing GPUs

Well, I posted recently this... :D

---

Well, now that is/was something I missed to read ...
So I don't feel that bad.

Who needs those expensive and Watt eating things ?

I know - bad joke !

cheers
S-F-V

Any idea what the optimal setup is for a Ryzen 9 5900 X CPU, running Windows 11? Has 12 cores 24 threads although I usually leave one free for the GPU
While I'm asking what is the optimal amount of tasks for a 3080 in the same machine as above. I have tried a maximum of 3 using 0.33 CPU run times in the high 270 – 286 seconds range. This is pulling between 340 and 360 W according to GPU Z.
TIA

While the Ada Lovelace architecture (40x0) is different than the Ampere generation (30x0), I found that running multiple WUs on my 4090 not only uses more power but run slower per WU.
Running a single WU does have dips in % utilization: 100% for 2-3 seconds, with a dip to ~25-50% every 4 seconds, I complete at WU (opencl_nvidia_101) in 32 seconds at 280W at a temp of 36°C
Running multiple WUs (2-4) puts utilization at 100% for most of the time, power to 290W at a temp of 38°C, but an average WU time of 35-38secs. (.135secs per cobblestone)

I'm also seeking optimization on CPU running....many processors running at 'default' settings currently: 5950x, 5900x, 6800H, 3x 6800U, 2x i7-8750H, 2x i7-7700K with an average of 4 secs per cobblestone.

I don't think any project so far had gone back through their applications and recompiled correctly for Ada Lovelace architecture. I've read many reports of projects either not sending work for RTX4000 cards or as you've noted, very sub-optimal gpu utilization for the cards.

---

Looks like Asteroids@home has taken the step for Ada Lovelace so far:
https://www.boincusa.com/forum/index.php?threads/new-cuda-application-released.2570/

I am back after a 2 year hiatus. My last assembled ASUS/AMD/Radeon died and because of supply chain problems I was unable to obtain the motherboard and cpu I wanted. I am now running an Apple Mac Studio M1 Max. I am trying to get a handle how this Mac compares with that ASUS with a AMD 9590. The Mac seems to be faster, crunching the numbers here and at Einstein@home, and where the AMD 9590 was a 200 watt cpu this M1 Max rarely goes above 20 Watts, I briefly saw 32 watts once. But it mostly stays around 10 Watts. I am not sure how to run the Mac Studio Benchmarks used for this thread plux don't know how the M1Max is being used, all I see in the description is the M1 Max has 10 processor cores. I am in the checking off stage with this machine, the PC did it like this, how does the Mac do the same thing.

---

The M1 processor is much faster than that old AMD architecture. However, unfortunately, you will find very little BOINC project support for the the Apple architecture. Very few native applications and no gpu support. Some project apps can be run in x86 emulation mode if you install the Apple Rosetta x86 emulator.

---

The M1 processor is much faster than that old AMD architecture. However, unfortunately, you will find very little BOINC project support for the the Apple architecture. Very few native applications and no gpu support. Some project apps can be run in x86 emulation mode if you install the Apple Rosetta x86 emulator.

I have not heard anything about that Microsoft sq1 chip in their Surface Pro X and assume the same problem. Probably a lot more Macs with be sold than that "X".

I kind of figured that with the majority of people using PC's that Mac's especially Apple Silicone Mac's would not be utilized to their potential whatever that might be.

I clued in someone over at MacRumors about this and he took a look at the Bionic Manager and he was excited to see a version for the old Apple Power PC systems. He said that it was old but that he was willing to code a modern version of it. I pointed out who he should contact.

---

I kind of figured that with the majority of people using PC's that Mac's especially Apple Silicone Mac's would not be utilized to their potential whatever that might be.

I clued in someone over at MacRumors about this and he took a look at the Bionic Manager and he was excited to see a version for the old Apple Power PC systems. He said that it was old but that he was willing to code a modern version of it. I pointed out who he should contact.

Many years ago, the company I worked for was getting rid of their Sun and Apollo workstations. I asked over at SETI it their app would run on them and was told they would provide me with the source code if I would build and support the app. That was the last time I asked about getting apps to work on old systems.

Hello,

I recently built a dedicated system for MW: https://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=963574

Used a Radeon Pro VII I got on eBay for $345 after tax paired with a 5900X. My advice would be to save a search for "pro vii" and then get alerts when new ones go up for auction. One just sold for $405 yesterday. It is an animal for separation tasks and doesn't get too hot when doing them. I have it running 4 tasks concurrently and they take an average of about 32 seconds each.

Best, Eric

Hello,

I recently built a dedicated system for MW: https://milkyway.cs.rpi.edu/milkyway/show_host_detail.php?hostid=963574

Used a Radeon Pro VII I got on eBay for $345 after tax paired with a 5900X. My advice would be to save a search for "pro vii" and then get alerts when new ones go up for auction. One just sold for $405 yesterday. It is an animal for separation tasks and doesn't get too hot when doing them. I have it running 4 tasks concurrently and they take an average of about 32 seconds each.

Best, Eric

Yea, I picked one for under 400 last year and price is still dropping.

Unaccountably, I cannot run my VII using a riser although the miner I bought it from had no problem with 8 of them on risers. Possibly the problem is windows 10 but that is a guess.
I would like to add more but the motherboard I am using only has a single x16 socket.

I am also running 4 at a time. Occasionally I get a stuck task. Sometimes 3 to 4 a day. Other times I can go for weeks at a time with no problem. A suspend followed by a resume :"unsticks" the tasks and I use boinctasks to handle the fixing automatically.

This discussion actually got me experimenting and for me it is slightly more efficient to run 6 tasks at a time: 46/6=7.6 seconds/task and 32/4=8 seconds/task.

How are you getting a consistent feed of GPU separation tasks? I just posted elsewhere that I'm having trouble getting sent them automatically and I have to keep hitting the update button:

"I built a whole machine just for Milkyway with a 5900X and a Radeon Pro VII (6.528 TFLOPS FP64) that finishes the stack of 300 separation GPU tasks in around 40 minutes doing 6 concurrent tasks. After that, even with Milkway as the only enabled project, it just runs CPU tasks, even after the 10 minute runoff time that I see in "properties" of the project in BOINC Manager that I don't understand. The only way for me to get more tasks is to hit update on the porject after all 300 have been completed.

I've combed through the documentation for app_config.xml and cc_config.xml to see if there were any serttings for mandating updates or something, but couldn't find anything. Is this a temporary issue with Milkway or do you have any other advice for keeping a constant stream of GPU tasks fed to my GPU?"

This discussion actually got me experimenting and for me it is slightly more efficient to run 6 tasks at a time: 46/6=7.6 seconds/task and 32/4=8 seconds/task.

How are you getting a consistent feed of GPU separation tasks? I just posted elsewhere that I'm having trouble getting sent them automatically and I have to keep hitting the update button:

"I built a whole machine just for Milkyway with a 5900X and a Radeon Pro VII (6.528 TFLOPS FP64) that finishes the stack of 300 separation GPU tasks in around 40 minutes doing 6 concurrent tasks. After that, even with Milkway as the only enabled project, it just runs CPU tasks, even after the 10 minute runoff time that I see in "properties" of the project in BOINC Manager that I don't understand. The only way for me to get more tasks is to hit update on the porject after all 300 have been completed.

I've combed through the documentation for app_config.xml and cc_config.xml to see if there were any serttings for mandating updates or something, but couldn't find anything. Is this a temporary issue with Milkway or do you have any other advice for keeping a constant stream of GPU tasks fed to my GPU?"

This has been cussed and discussed before.
There are several ways to fix the delay. One it to use a batch file runs boinccmd.exe in a loop with a time delay. That app can issue an update to a local or networked system.

I put together a "mod" of boinc version "7.15' that a number of uses still work with. Check the statistics leaders. It is not compatible with 7.20 and I update the mod here

https://github.com/JStateson/Milkyway-7-21

replace boinc.exe with the boinc--7-21.exe and rename to boinc.exe

AFAIK GPU in Apple Silicon does not have FP64 support at all. Is it a problem for MilkyWay@home?

AFAIK GPU in Apple Silicon does not have FP64 support at all. Is it a problem for MilkyWay@home?

According to this, it even doesn't have OpenCL.

---

Can anybody comment on what run times are like for the AMD Ryzen 9 7900X 12 Core/24 Thread AM5 CPU?

What application?? I assume you mean N-body since that is the only one left and you only mention a cpu.

You probably can find one somewhere in the Top 100 lists.

---