LHC@home

Can anybody explain how an AMD Ryzen 7 1700 can get this benchmark values?


https://lhcathome.cern.ch/lhcathome/show_user.php?userid=442124
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10579909
Measured floating point speed: 9.27 billion ops/sec
Measured integer speed: 96.27 billion ops/sec



https://lhcathome.cern.ch/lhcathome/show_user.php?userid=94014
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10389189
Measured floating point speed: 5.14 billion ops/sec
Measured integer speed: 59.23 billion ops/sec

Can anybody explain how an AMD Ryzen 7 1700 can get this benchmark values?

You don't have to look far to find discrepancies. Here are the values for my Haswell machines:

i7-4770:
Measured floating point speed 4.57 billion ops/sec
Measured integer speed 117.55 billion ops/sec

i7-4790:
Measured floating point speed 4.49 billion ops/sec
Measured integer speed 59.64 billion ops/sec

The usual explanation (probably on the BOINC forum, or maybe Einstein) is that BOINC gives notoriously unreliable measurements. I see no reason to doubt it.

The usual explanation (probably on the BOINC forum, or maybe Einstein) is that BOINC gives notoriously unreliable measurements. I see no reason to doubt it.

I've seen on couple of my dual boot hosts (Win / Linux) that Linux flavor and/or kernel versions can change those numbers quite a bit. Also different numbers of allowed CPUs while running Boinc benchmarks would change those values. Some time ago I run a few speed tests while I was updating Linux versions on my hosts. These hosts have almost identical hardware, settings and clock speeds (biggest difference is perhaps in their RAM configuration). Windows benchmark results ended up very close to each other, but results in Linux were twisted.

All benchmarks were run with 4 cpus of 12 allowed.

Host 1:

Windows 10
CPU-Z: single 356.7 / multi 1398.6
Boinc: 4878 / 12217

Linux Mint 19.1 Tessa (kernel 5.1-rc4)
Boinc: 4873 / 149970

Host 2:

Windows 10
CPU-Z: single 358.1 / multi 1444.8
Boinc: 4897 / 12197

Linux (Ubuntu 18.10 , kernel 4.18.0-17)
Boinc: 5352 / 105679

Linux (Ubuntu 19.04 , kernel 5.0.0-13)
Boinc: 5720 / 101207

I've seen on couple of my dual boot hosts (Win / Linux) that Linux flavor and/or kernel versions can change those numbers quite a bit. Also different numbers of allowed CPUs while running Boinc benchmarks would change those values.

It is quite possible that Windows is more consistent than Linux.

My i7-4770 is on Ubuntu 18.04, while my i7-4790 is on Ubuntu 16.04.
But they are both on the same Linux kernel (4.15.0-48), though I wonder if it has been updated since the benchmarks were run?
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10594847
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10588304

I allow all cores to run BOINC on the 4770, but I do reserve one core to support a GPU on Folding on the 4790. They are on different projects, if that matters, but BOINC should suspend them to run the benchmarks. I think BOINC just uses a random number generator.

https://lhcathome.cern.ch/lhcathome/results.php?hostid=10455317&offset=0&show_names=0&state=4&appid=13

Maybe we all should switch back to VirtualBox 5.2.16 and run single core tasks (and Required extension pack not installed, remote desktop not enabled)

Maybe it will work with Atlas and CMS of course it won't but maybe a 100,000 credit Sixtrack tasks will be next

And the exact opposite is still running here https://lhcathome.cern.ch/lhcathome/results.php?hostid=10519938&offset=0&show_names=0&state=0&appid=14

... and Required extension pack not installed ...

why so?

Maybe we all should switch back to VirtualBox 5.2.16 and run single core tasks

If only that was just the solution, I've spend some time trying to figure it out but I honestly don't have a clue what is going on. Some observations though:

This host runs VB 6.0.8 and gets 23k for one task: https://lhcathome.cern.ch/lhcathome/result.php?resultid=230825272

I kinda hoped it would indeed be a VB thing, so I installed 6.0.8 and tried, still getting only 500-700 credits per task with similar run times on a very similar system (1950x instead of a 2950x):
https://lhcathome.cern.ch/lhcathome/result.php?resultid=230936322

Given the similar run times there shouldn't be that much of a difference, not consistently at least. The only real difference is the measured Device peak FLOPS, explained here: https://boinc.berkeley.edu/trac/wiki/CreditNew

This has been the case since LHC switched credit systems, explained by Laurence here: https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=5021&postid=38790#38790

And indeed, 680.34/4.51*152.78=23047, so the difference in device peak flops indeed accounts for the difference. While the old system was far from perfect, this system seems to be even worse given how inconsistent it is.

Dont complain, is forbidden here :)

---

I run the old virtualbox and single core on my machines, I find it to more reliable.

CMS is doing quite well on all 8 cores of my i7-3770 with VirtualBox 5.1.38 (Ubuntu 16.04).
https://lhcathome.cern.ch/lhcathome/results.php?hostid=10590338
EDIT: It helps to have 32 GB of memory.

I do native ATLAS and Theory, and don't know about them.

https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10546354
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10596058

Looks like on both hosts a script cancels Theory tasks after 3 h runtime.
This always causes the currently running subtask to get lost.

Not helpful for the project.

Self reporting 10457490 which didn't like the VBox 6.0.10 upgrade (which the 2 other hosts had no problem with and are running fine). It also didn't like the downgrade back to 6.0.8. A series of BSODs of Storage Exception, System Service Exception, numerous restarts, checked Hyper-V disabled, Bios checked for Virtualisation Enabled, and finally Critical Process Died resulting in Boot Failure loops. Original installation disks are long gone so had to download and create installation media on another host, which wouldn't let me overwrite the (presumably) corrupted windows files (or maybe the HDD is past its best) so I'm currently waiting for it to do the full install which is a right royal pain. And then Bionc, VBox etc.
Oh, joy. Several/many hours of my time that I'll never get back but luckily I have only used this host for Boinc and hopefully it might even be faster as it should clear out all the rubbish my sister left on it.

Oh, joy. Several/many hours of my time that I'll never get back but luckily I have only used this host for Boinc and hopefully it might even be faster as it should clear out all the rubbish my sister left on it.

All that work so you can run 1 single core Theory VBox and maybe a Sixtrack simultaneously. That's dedication to the cause.
With just a little more work you could put Linux with a small desktop GUI on that machine and run a single core (2 athena threads) ATLAS plus a (Theory or Sixtrack), like I do on my old X2 with 4 GB. Yes, a bit of a learning curve but you could learn enough in a day to get you up and crunching. Anything to get rid of VBox. It's not just the overhead it's the ongoing PITA.

Hi Bronco
It HAD been running 2 x Single-core Theory tasks on that host quite happily (all my hosts just run singles as I have found that to be most efficient) but it now crashes each time I start a VM, even after all the reinstallation so there's still something not right (something wrong when VM dials out Luckily there is plenty Sixtrack work just now so it can get on with doing that until I have the time (and inclination) to dig further and sort out whatever's wrong.

I tried Ubuntu (running within a VM with the added complication of nested VMs within a VM) ages ago, before the T4T/vLHC/LHC Grand Unification, when we were testing the Christmas Challenge, with some success but was never really comfortable with all the Command Line stuff. If I can't solve the issue, I'll see if I can find the most Windowsy distro then I can join in playing with the Native apps.

https://lhcathome.cern.ch/lhcathome/results.php?hostid=10519938

This one must run itself like HAL 9000
It has been doing this for a long time.

https://lhcathome.cern.ch/lhcathome/results.php?hostid=10519938

This one must run itself like HAL 9000
It has been doing this for a long time.

That host is already down to a max of 1 Theory Native task per day.
To go down to 1 ATLAS task per day it has to crunch another 306 (errors).

Yeah I figured that had to be why it was only one task at a time CP but I know it has been doing this since before March so imagine how many 1000's of these have been wasted and........as I mentioned before have to wait 2 weeks before they are resent to hopefully a pc that has a human checking once in a while and noticing thousands of errors and not ONE single Valid for over 6 months........or is there some award for the most computer error tasks in a row?

Seems to me that the server should finally stop sending to that host after the first 1000 computer errors and zero Valids.
And I imagine this is not the only one doing this.

Maybe they should switch to seti

https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10617647
Attached a week ago and within this week more than 400 Theory native at 100% error rate because CVMFS is not installed:

22:25:43 CEST +02:00 2019-10-14: cranky-0.0.29: [INFO] Checking CVMFS.
22:25:43 CEST +02:00 2019-10-14: cranky-0.0.29: [ERROR] 'which' could not locate the command 'cvmfs_config'.

That pc is one we haven't seen here for close to 10 years

1 -core with less than 2GB ram
https://lhcathome.cern.ch/lhcathome/show_host_detail.php?hostid=10617647

And I think it has a Boinc Version from a couple years ago.

https://cernvm.cern.ch/portal/filesystem/downloads

what am I doing up at 5:50am