| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Hardware Chat Room
Thread: AWS Graviton2 (ARM) |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
CurtisNewton
Cruncher Joined: Feb 24, 2008 Post Count: 25 Status: Offline Project Badges: 
|
Just set up a Gravition2 instance in AWS cloud, and the whetstone as caclculated by boinc are looking pretty promising. Floating point MIPS are much better compared to the xeon platinum they are typically offering for computational tasks.
Just wonder if the CPUs will be sold public of it is just for inhouse use. Interesting point here, as far as cloud pricing is concerned, both systems have the same price per hour. However, WCG does not seem to support ARM / Linux since I get a "This project doesn't support computers of type aarch64-unknown-linux-gnu" message. Below ist the boinccmd --get_host_info for an intel c5 instance and a graviton c6 instance. After AMD Ryzen and now ARM, there seems not to be much good news left for Intel. CPU vendor: GenuineIntel CPU model: Intel(R) Xeon(R) Platinum 8124M CPU @ 3.00GHz [Family 6 Model 85 Stepping 4] CPU FP OPS: 4347358946.709541 CPU int OPS: 105077648591.438004 OS name: Linux Ubuntu OS version: Ubuntu 18.04.4 LTS [5.3.0-1032-aws|libc 2.27 (Ubuntu GLIBC 2.27-3ubuntu1.2)] CPU vendor: ARM CPU model: [Impl 0x41 Arch 8 Variant 0x3 Part 0xd0c Rev 1] CPU FP OPS: 5851013403.063561 CPU int OPS: 88330316868.276672 OS name: Linux Debian OS version: Debian GNU/Linux bullseye/sid [5.7.0-1-cloud-arm64|libc 2.31 (Debian GLIBC 2.31-2) |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
WCG does support Linux on ARM for a few sciences. You probably have to add an'alternate_platform' line to cc_comfig.xml. There are various topics discussing this and think armstrdj started a special advisory thread on the subject.
----------------------------------------edit: The armtrdj threat https://www.worldcommunitygrid.org/forums/wcg...ead,42687_offset,0#636106 [Edit 1 times, last edit by Former Member at Aug 2, 2020 2:14:07 PM] |
||
|
|
Umlauf
Advanced Cruncher Joined: Mar 20, 2020 Post Count: 52 Status: Offline Project Badges: 
|
Hello,
today WCG only support OpenPandemics for linux on ARM. A few days ago there was a beta for MCM this also runs on linux on ARM. I think it will be available soon as new version. All you have to do is to add the alternate platform informations in the cc_config.xml file : I have the 3 following infos in my file: arm-unknown-linux-gnueabihf armv7l-unknown-linux-gnueabihf aarch64-unknown-linux-gnu You should find all infos on the link from lavaflow. |
||
|
|
CurtisNewton
Cruncher Joined: Feb 24, 2008 Post Count: 25 Status: Offline Project Badges:
|
Thanks to both of you, I will try this.
|
||
|
|
William Albert
Cruncher Joined: Apr 5, 2020 Post Count: 41 Status: Offline Project Badges:
|
Just set up a Gravition2 instance in AWS cloud, and the whetstone as caclculated by boinc are looking pretty promising. Floating point MIPS are much better compared to the xeon platinum they are typically offering for computational tasks. My old desktop machine was recently retired, and as part of that process, the clock speed was restored to stock settings, and I replaced Windows 10 with Ubuntu 20.04. When comparing the median points/results returned, the machine was about 3% faster when running Windows 10, which is about where I'd expect. (Even though it was clocked higher, I was using Windows 10 as a desktop, so the CPU wasn't 100% dedicated to WCG as it is now.) Despite this, the machine on Ubuntu 20.04 reports a 13% higher floating point score than it did when running Windows 10. So yeah, take that figure with a grain of salt. [Edit 1 times, last edit by William Albert at Aug 2, 2020 9:10:38 PM] |
||
|
|
CurtisNewton
Cruncher Joined: Feb 24, 2008 Post Count: 25 Status: Offline Project Badges:
|
I know, there's a lot of things that effect performance.
After all, for many people running dedicated crunchers the relevant number might be (amount of money spend) divided by (numer of jobs done). Talking about cloud instances, the order from cheap to expensive is typically ARM, AMD, Intel. And Windows machines are always more expensive than linux machines because you have to pay the license, too. Taking this into account, the ratio cost / tasks might be better for slow machines. br, carsten |
||
|
|
|