Message boards :
Number crunching :
AMD A-2 thru A-10 and new AMD cpu/gpu integrated chips
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 7 · Next
Author | Message |
---|---|
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I have started shutting down the A4-5000 because it is my "writing" machine and the only SSD in the house. I would hate to run out of writes early. I got some advice from another thread that has suggested adding -sbs 384 to my command line. I also transplanted all the configuration files I have been using on the A4-5000 to the A6 and it seems to have perked up a bit. Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I have started shutting down the A4-5000 because it is my "writing" machine and the only SSD in the house. I would hate to run out of writes early. The -sbs 384 down to -sbs 192 all cause CPU-Z to report unstable processing. I am running the A6 as a gpu only machine to see just how fast it appears to process without the cpus. I will then try adding back 1 cpu and then 2 since I believe it only has 2 Floating Point processors. Tom A proud member of the OFA (Old Farts Association). |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
I have started shutting down the A4-5000 because it is my "writing" machine and the only SSD in the house. I would hate to run out of writesHeh heh, "writing" machine...disk "writes"... Anyway--I have a Linux virtual machine dedicated to running BOINC that records metrics about the guest system. This gives some insight into the wear on the disk. Over the thirty days that it crunched so far, it processed over 1000 World Community Grid tasks and wrote about 75 GB to disk (including non-BOINC data, though there's probably not much of that in Linux). That's about 2.5 GB per day. SETI seems less disk intensive than WCG, and I know from past measurements that the Windows OS itself (indexing, anti-malware, page file, updates, etc.) writes several times this amount of data with daily use. At this rate, it would take at least a couple of decades for SETI crunching to wear out an SSD...longer than I'd trust a spinning HDD to last even if it was doing nothing. The -sbs 384 down to -sbs 192 all cause CPU-Z to report unstable processing. Tom, how does CPU-Z report unstable processing? I can't seem to find this area. I used -sbs 512 on my APU and I didn't see any signs of instability. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
Perhaps I should rephrase the sentence as "GPU-Z is showing brief interuptions in processing via the Gpu Load and Clock graphs in the sensor tab." I am interupting those interuptions as "instablity." Since any interuption slows things down and those interuptions go away when you drop the -sbs N parameter for both 192 and 384 I assume the -sbs parameter is incorrect. And is causing something that interupts processing. Tom A proud member of the OFA (Old Farts Association). |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
Tom, how does CPU-Z report unstable processing? I can't seem to find this area. I used -sbs 512 on my APU and I didn't see any signs of instability.Perhaps I should rephrase the sentence as "GPU-Z is showing brief interuptions in processing via the Gpu Load and Clock graphs in the sensor tab." I am interupting those interuptions as "instablity." Since any interuption slows things down and those interuptions go away when you drop the -sbs N parameter for both 192 and 384 I assume the -sbs parameter is incorrect. And is causing something that interupts processing. Ah, interesting observation...I'm curious to try this out. I don't remember seeing a jagged load graph, but I'll have a look with different -sbs values. |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
A quick update about the Intel NUC box that I ordered: it has been crunching for a couple days now. I was a bit worried about compatibility because the host that Kissagogo27 linked to (same model) showed a bunch of invalid AMD GPU tasks. Maybe its owner overclocked it--I'm not getting any invalids yet. For the purpose of this discussion, I can confirm that Kaby Lake-G designs are not "classic" APUs, at least for the Vega M. The cooling system is the only resource that the Vega M GPU shares with the CPU. This GPU gets its own memory, power limit, and PCIe bus, and GPU loads did not significantly impact CPU performance. It's just a tiny dedicated GPU that I can't upgrade if I want to later. The standard Intel integrated GPU behaves like the other Intel and AMD APUs we talked about. When loading this GPU, the CPU throttles frequency because the chip hits its 65 W power limit. As described for other APUs, undervolting the CPU and Intel GPU allows both to run at full clock speed. While it does add a small net increase to overall SETI productivity, the integrated Intel GPU ties up an entire core and generates enough latency in the system to degrade the performance of both the CPU and the Vega M GPU. This seems to match the experience of others with Intel GPUs. I may try tuning it some more, but I'm not sure the Intel GPU's work is worth the extra 10 to 15 W of power consumption and heat output. I'm a little concerned about the longevity of this system as a BOINC machine. The cooling system is quite capable (the box never feels more than slightly warm to the touch under full load thanks to the jet stream of hot air coming out the back), but core temperatures hover around 80 C (even with the "cool" fan speed scaling preset in the BIOS) and the dual blower fans roar under this load. The fans and heatsink are not easy to clean without disassembling the entire enclosure, and I expect it to need cleaning often with the amount of air it moves. Now I need to work on getting the correct command-line arguments for the Vega M apps, but that's for another thread. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
There are a couple of utilities that will limit the temperature that either the cpu or the gpu or both are allowed to get up to. The catch is they will slow the Seti processing down. Here is the URL to the Seti website page https://boinc.berkeley.edu/addons.php I have experience with TThrottle. It is really useful for when I am running laptops because of their tendency to run very hot when they are crunching Seti. Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
A quick update about the Intel NUC box that I ordered: it has been crunching for a couple days now. I was a bit worried about compatibility because the host that Kissagogo27 linked to (same model) showed a bunch of invalid AMD GPU tasks. Maybe its owner overclocked it--I'm not getting any invalids yet. I have been examining some of the AMD gpu tasks. A number of them are running under 10 minutes. That probably makes the gpu faster than a gtx 750Ti (as low as 15 minutes) and slower than a gtx 1060 3GB (around 7+ minutes). [Times in Windows with the SOG task]. A VERY respectable processing pace! Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
http://setiathome.berkeley.edu/show_host_detail.php?hostid=8518624 the host that Kissagogo27 linked to (same model) That host hasn't connected since the 10th of October. And I am not sure it is crunching any CPU tasks at all. Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
@monospice Did you also order one the new APU's too? Tom A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I am interupting those interuptions Sigh. Typo alert! "I am interpreting those interruptions..." A proud member of the OFA (Old Farts Association). |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
This computer https://setiathome.berkeley.edu/show_host_detail.php?hostid=8535523 is running an AMD Ryzen 3 2200G with the Radeon Vega (8) Graphics chip. The CPU is cranking out some very respectable processing. The gpu looks like it is not processing hardly at all. I would dearly like to find another host of this type (there are 30 some of them) and compare the gpu processing times. Tom A proud member of the OFA (Old Farts Association). |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
...I would dearly like to find another host of this type (there are 30 some of them) and compare the gpu processing times. Tom, I ran a quick parse on the hosts stats export to find some more Ryzen 3 2200Gs. Here's what came up: 5153399 6196425 6822565 7153793 7405371 8096159 8124409 8152278 8239266 8249384 8251100 8353789 8467612 8470764 8470894 8478006 8479872 8481920 8483704 8485343 8486774 8489427 8489717 8493180 8495722 8495845 8498690 8498846 8499372 8500012 8501861 8507209 8510603 8513663 8514379 8516475 8516694 8518783 8518990 8520539 8521097 8527576 8531191 8533345 8535523 8537473 8538646 8541081 8542928 8543982 8545096 8546911 8547321 8547517 8554696 8555805 8557352 8557673 8558151 8558351 8558549 8560098 8560659 8562529 8563082 8566385 8567107 8569657 8571872 8572372 8574662 8575009 8575069 8576639 8577054 8577260 8578462 8580183 8583089 8584248 8584284 8586996 8587831 8588123 8589528 8591539 8593850 8594090 8594777 8595655 8595676 Edit: the subset where RAC is greater than 100, ordered from highest RAC to lowest: 8580183 8562529 8584248 8558549 8584284 8577054 8518990 8535523 8554696 6196425 8538646 8558351 8558151 8594090 8595655 8591539 8566385 8547517 7153793 7405371 8542928 5153399 8575009 8560659 8575069 8595676 8578462 8514379 8470764 8516475 8593850 8588123 8527576 8589528 |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
A quick update about the Intel NUC box that I ordered: it has been crunching for a couple days now. I was a bit worried about compatibility because the host that Kissagogo27 linked to (same model) showed a bunch of invalid AMD GPU tasks. Maybe its owner overclocked it--I'm not getting any invalids yet. It surely is faster than I thought it could be, but something seems to be holding it back. There are a few other users with the same model that seem to be crunching faster by a significant margin. These hosts seem to have a faster "average processing rate" for both CPU and GPU, but no valid tasks that I can examine: 8524176 8527615. This one's more in line with my box: 8528551. The other 10-or-so active i7-8809G NUC hosts fall somewhere in between. Maybe I need to try the Lunatics apps that I keep reading about. That host hasn't connected since the 10th of October. And I am not sure it is crunching any CPU tasks at all. I had a look at it back around that time before the history fell off. It stacked up around 100 invalid tasks for the Vega GPU. Seems like the owner noticed the invalids and unplugged it in the meantime. I noticed another couple of these NUCS with a few invalid GPU tasks. Maybe an automatic driver update rolled out within the last few weeks that botched things up. Here is the URL to the Seti website page https://boinc.berkeley.edu/addons.php I have experience with TThrottle. It is really useful for when I am running laptops because of their tendency to run very hot when they are crunching Seti. Thanks for that. The cold weather that just blew in seems to have resolved my heat concerns and fan noise for now. I'll keep that one in mind next time the seasons change :) Did you also order one the new APU's too? I did consider it! But no, the Vega M seemed like the more interesting "APU", so I picked that one instead. Turns out it's just a discreet GPU on a chip. This NUC is a nice little box after all, though. Pricey, but not unjustly so. It's loaded with features (that I may never use--serial IO header, infrared port...what?) |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
...I would dearly like to find another host of this type (there are 30 some of them) and compare the gpu processing times. Thank you. There are more than a few areas of Seti that I am either not very savy, or I haven't been able to get to work the way others have. I have looked at the first two top performers. The first is running both the cpu and the gpu. The gpu times seems to be very respectable in the 10 minutes per task range. The Gflops # is 200+. The 2nd top performer is running pure cpu only tasks. :) A number of the low RAC people on your sorted list just started in October :) Maybe Seti's active processors grows? I hope this satisfy's my itch to create one so I can see how it performs :) Tom A proud member of the OFA (Old Farts Association). |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
I have looked at the first two top performers. The first is running both the cpu and the gpu. The gpu times seems to be very respectable in the 10 minutes per task range. The Gflops # is 200+. The 2nd top performer is running pure cpu only tasks. :) The first one looks like they're actually crunching with their GTX 1060 instead of the GPU on the APU. The third host in the sorted set consistently crunches on the Ryzen's GPU. SoG tasks seem to run for about an hour. This may not represent the full potential of the GPU...it's finishing tasks about as fast as my 3-year-old A8-7650K, which is a bit disappointing. The fourth host in the set takes twice as long as the third. Maybe the chip is hitting its power limit like the other APUs seem to do. The sixth and seventh hosts do a bit better at around 45-60 minutes per GPU task. James's Ryzen (the host you originally linked to) is number 8. He doesn't seem to crunch many GPU tasks...they don't seem to perform well. I stopped at #9, which seems to crunch GPU tasks fastest at about 35-50 minutes. The only (dubious) pattern that I can see between the slower the faster hosts is that the faster hosts run on Windows 10 Pro instead of Core or Education versions (or Linux). Unfortunately, the stderr history doesn't show memory speed of the host, which may play a role for these APUs. |
rob smith Send message Joined: 7 Mar 03 Posts: 22202 Credit: 416,307,556 RAC: 380 |
The first one (https://setiathome.berkeley.edu/results.php?hostid=8580183) has only (recently) returned results from the GTX1060. It also apears to be a "sporadic contributor". So I think I'd put that one off the list. Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
I still had a tab open to that 1st machine and re-checked the details for "pre-Gtx 1060" and it certainly is faster than "James" gpu. All the AMD Vega tasks were sporting 53 to 68 Gflops which while not racehorse were certainly burying my A4 and A6 gpu's :) I am not certain where I got the "200 Mflop" #. It might have been from the peak Gflops # of a task. Now I wonder how good the AMD Ryzen 5 2400G Processor with Radeon RX Vega 11 Graphics runs? Tom A proud member of the OFA (Old Farts Association). |
monospice Send message Joined: 20 Sep 18 Posts: 30 Credit: 1,905,918 RAC: 11 |
All the AMD Vega tasks were sporting 53 to 68 Gflops which while not racehorse were certainly burying my A4 and A6 gpu's :) I am not certain where I got the "200 Mflop" #. It might have been from the peak Gflops # of a task. As desktop APUs with 65 W TDPs and five times the compute units, it's difficult to compare them to the 15 W TDP chips in those laptops. I have a hard time comparing their GPUs to dedicated cards as well, and differences between APUs and Intel's integrated graphics are also hard to define because Intel's solution comes with so much system baggage. They really do appear to fill a niche of their own, and performance seems quite reasonable. Now I wonder how good the AMD Ryzen 5 2400G Processor with Radeon RX Vega 11 Graphics runs? ...and you shall receive :) Here's RAC greater than 100 for Ryzen 5 2400Gs, ordered from highest RAC to lowest: 8586507 8566538 8545679 8572001 8571043 8547917 8342450 8566933 8548533 8546280 8543066 8522191 8518888 8512724 8515240 8549837 8542534 8504681 8549770 8539439 8260118 8559956 8536117 8482386 8576265 7754082 8510845 8432230 8481869 8575645 8564168 8588344 8513727 8583581 8541129 8548682 8587968 8594594 8437351 7785756 8548445 8388407 8568487 8529971 8572699 8538079 8587366 8503913 8583907 8590273 8517650 8580479 8583952 8577829 8595604 8569681 8524846 8575938 8576765 7748608 7496233 8585105 8497115 8581125 8473037 8476462 1917496 8561382 8579321 7640031 8498544 First one is another GTX 1060 machine (not using the Ryzen Vega GPU). The second host just crunches CPU tasks. The hosts that do run GPU tasks on the APU do seem a little faster than the Ryzen 3/Vega 8 hosts, though not by an astonishing margin. |
Tom M Send message Joined: 28 Nov 02 Posts: 5124 Credit: 276,046,078 RAC: 462 |
The fourth machine is running some ati*SOG tasks and getting a respectable 156 Gflops processing speed. But the cpu processing isn't all that fast. The dedicated cpu processing machine isn't all that fast. I would swear I saw 30+ Gflops back on a cpu only machine in the 2200G listing. Hmmm... I wonder if this might be an issue with running 1 memory stick vs. 2 since these things are very sensitive to having "fast" ram access. After looking at various gpu tasks, I am wondering if anyone of these people are using anything on the command line and/or with the app_config.xml file? Tom Tom A proud member of the OFA (Old Farts Association). |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.