Slowly working my way through the various SSee applications to see the impact (good or bad) of them
Thus far I've only done SSE2, which I'm taking as my baseline. Now running SSSE3, which doesn't on first impression look to be significantly quicker, or slower. As ever there will be several days running to make sure all is working well, then 'll try one of the SSxx4.x variants, and finally AVX. All these tests are being run with the stock motherboard settings and no overclocking.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

So I gather you are compiling the apps yourself in Linux? In the past on Seti V7, when we had all the various flavors for the Windows platform, the SSE4.1 and SSE4.2 were always the fastest on the AMD FX processors. Second fastest was the AVX app. Very limited choices today for Seti V8 for the Windows platform. Its either SSE3 or AVX. That's it. No SSE4.X choice. It will be interesting to learn your results and how SSE4.X and AVX compared to each other.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

No, I'm using the Lunatics finest offerings for Linux - this removes any issues I might introduce when compiling my own versions.

It was recalling that test of old that prompted me into this little, very informal, test.


http://lunatics.kwsn.info/index.php?action=downloads;cat=48

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

No, I'm using the Lunatics finest offerings for Linux - this removes any issues I might introduce when compiling my own versions.

It was recalling that test of old that prompted me into this little, very informal, test.


http://lunatics.kwsn.info/index.php?action=downloads;cat=48

Wow, I was completely unaware that there were new V8 CPU apps for Linux. I wonder if Urs could be persuaded to port those apps over to Windows.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

It does look as if the Linux world is ahead of Windows for once.
But I suspect there is just a bit more than a simple re-compilation using the appropriate Windows compilers :-(

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Probably so. As I understand it, the CPU apps developed by Joe Segur were hand optimized, so they probably couldn't just be run through a Windows compiler. Joe's work is still appreciated but we miss him.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Slowly working my way through the various SSee applications to see the impact (good or bad) of them
Thus far I've only done SSE2, which I'm taking as my baseline. Now running SSSE3, which doesn't on first impression look to be significantly quicker, or slower. As ever there will be several days running to make sure all is working well, then 'll try one of the SSxx4.x variants, and finally AVX. All these tests are being run with the stock motherboard settings and no overclocking.

. . That surprises me somewhat. On my Core2 Duo I had been running SSE3.0 and when I changed to Linux I got an SSSE3.0 app which improved things greatly. Under SSE3 normal AR Arecibo tasks would take about 2.5 hours. under SSSE3 that dropped to about 2 hours. I guess it is very much processor dependent.

Stephen

?

I'm running a Ryzen 1800X on my build with (2) GTX 1070 GPUs.

I am very new to SETI@Home and BOINC. I tried running it on my old FX-6300 but the full load was crashing my non-overclocked computer so I ended up uninstalling it until I got my new build set up.

I'm currently only contributing to SETI@Home and so far things have gone really really well. Usually on the 10-hour downtime on Tuesdays I'm able to get through all 300 tasks and my computer gets an hour of downtime or so (haha).

I started up a Ryzen team if any of you are interested. It's kinda fun teaming up with other Ryzen builds, and I was thinking we could share information on our builds, tweaking to get better results, etc.

Finaly got the Corsair AM4 bracket last week and I switched the Ryzen in to the big box yesterday and... WOW.
The 1700 stock cooler is good but the water cooling allow for good overclocking as every one would expect.

The system is now stable at 3.7 GHz and RAM at 2933 MHz. CPU temp never go above 46c.
under full prime load but a little higher at 48 when the GPUs are running and heating up the box a little.
I am crunching Milkyway right now because WU are always the same making it all easy to compare.
with the latest BIOS, The RAM speed boost, and the long awaited bracket allowing for overclocking,
time to complete 16 Milkyway@home cpu tasks went from 01h06m to 00h52m.

I will soon come back to SETI and take some times learning about the app info and config, Lunatics, SSE and AVX, etc.
Keith was very kind to give me a copy of his app config and it is cool, but I need to learn all this.
I see there is a new Ryzen team and many other Happy "From FX to Ryzen upgraders" :)

It really is an Amazing CPU...

Have a nice week end all !

Congrats on getting the AIO back on the chip. Yes, the raw processing power is pretty impressive.
I just had a bit of misfortune though. I didn't get to use my AM4 bracket for very long it seems. Was going to try and see if the system would be stable after dropping the memory from CAS16 >> CAS14. Never made it back to the desktop. I seem to have lost the pump on my AIO ..... again. I have the Corsair H100iV2. This was the warranty replacement one after the original died within 3 months. This one lasted 4 months. I decided that's it for the H100iV2 which is an Asetek cooler manufactured for Corsair. I have been on the fence whether to move to the CooliT made H110i Corsair AIO after the SIV developer has repeatedly urged me to move to that manufacturer since it is 1000% better than the brain-dead Asetek models. So, as soon as I realized the pump had died, I ordered up the H110i which should be here on Tuesday. It apparently just uses the stock socket fixture hold downs for installation so no need for the AM4 bracket now. Just now got the system back online with my temporary Noctua NH-D15 SE-AM4 tower cooler. It doesn't cool as well as the AIO though. Running about 5° C. hotter than the H100iV2 AIO. Now that I have the correct sized case for a 280mm radiator, it will be interesting to find out if it cools any better or not compared to the H100iV2.

The system is running well so far with the drop in CAS latency to CL14. Will be interested to see if any improvement in processing times.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I am sorry to ear that but at the same time tou will have a new 280mm cooler. Mine is working fine so far, the pump is always in quiet mode it might help but if it fails I will too get the h110i.

The FX system power comsumption is not as bad as I first thought. The 8 white led fans + the pump + the rog strix video card (those seems to consume more current than the aero ones of the small box) account for 50% of the power difference I had notice between the two system. The difference between the FX and the Ryzen system is about 60 watts at full load. My appologies to my 8350...
I have tried to bring the 1700 above 3.7 GHz but it became more and more difficult to make it stable so I stopped at 3.7 GHz.
It still scores 16000 on cinebench !!!

Next step here will be Lunatics, app.info and app.config. I am not sure what to select in Lunatics... SSE2-3-4 AVX etc and for the gpu it only propose CUDA app. Not sure what to select and what to do after. will I have to create another app.config ? I wish there was a step by step simple instruction guide somewhere on how to optimize your machine for SETI...

Have to go. Cheers

I'm impatiently awaiting the mailman who has my H110i on the truck. Monday deliveries are always at the very end of the day for some reason. Probably different carrier on Mondays I guess. I digress.

Not sure about what you are seeing in the Lunatics installer regarding your comment about only seeing CUDA. I looked at your task list for the Ryzen system which has a GTX 1070 in it and I see one task completed under Anonymous platform using the r3557 OpenCL SoG app. If you are using the latest 0.45 Lunatics -Beta-06 installer on the page for Nvidia there are tick buttons to select the SoG app. By default that installer will select one of the CUDA apps. You just need to untick the CUDA app and select the SoG app. For the CPU, you need to select the AVX app. The SSE4.X selections are meaningless as it will install the SSE2 app since there are no SSE4.X apps anymore for Windows. The AVX app is a lot faster than the SSE2 app for CPU's.

I also see that the latest MB graphics task completed under the Anonymous platform looks to have been crunched with the r3557 SoG app with the same app_config and tuning parameters I use for my 1070's with the 2048 SBS size. Everything looks really good there. I can't tell whether you have your tuning parameters in an app_config.xml file or in the MB command line file. You can't tell from the stderr.txt output. I find it very easy to put all the tuning parameters into an app_config.xml file since it survives any changes or updates to BOINC or the app_info.xml. A lot easier to edit the app_config compared to the app_info. I just let the Lunatics installer write the default app_info and make any needed changes to the app_config.

You can update the SoG app from the r3557 version the Lunatics installer put in to the latest r3584 SoG app using the MB aistub file that is now included in the r3584 7zip file that is on Mike's World.
mb8_win_x86_sse3_opencl_nv_sog_r3584.7z

Use the aimerge.cmd command file to merge the new r3584 app into your app_info. Or manually edit the app into you app_info via Notepad's Replace All function. Either works. The advantage of the new r3584 SoG app is a reduction in Inconclusives accumulation in your Task list. There really isn't any performance differences between the two versions though as far as I can tell.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Numbskull is closing in on Pipsqueek and Keith-Window7 in the RAC race. Updated Numbskull yesterday to dual 1070's to match the other systems. It is now over 51K RAC, about 12K increase from where its started a month ago before it got a new Ryzen 1700X. I expect it to eclipse my other systems before the end of the month.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Numbskull is closing in on Pipsqueek and Keith-Window7 in the RAC race. Updated Numbskull yesterday to dual 1070's to match the other systems. It is now over 51K RAC, about 12K increase from where its started a month ago before it got a new Ryzen 1700X. I expect it to eclipse my other systems before the end of the month.

must paddle faster...must paddle faster...

Numbskull is closing in on Pipsqueek and Keith-Window7 in the RAC race. Updated Numbskull yesterday to dual 1070's to match the other systems. It is now over 51K RAC, about 12K increase from where its started a month ago before it got a new Ryzen 1700X. I expect it to eclipse my other systems before the end of the month.

must paddle faster...must paddle faster...

. . Gotta be ready for the guppi flood that's coming. Not to mention when Parkes comes online. :)

Stephen

:)

I must have missed the news. So when is Parkes data coming online?

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Shoot, Numbskull is going to pass the other crunchers before the next outage. Maybe before weekend. It is already at 52.5K RAC.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Well, my prognostication was all wet. Numbskull has already matched and passed my original fast crunchers. The trajectory slope was a LOT steeper than I had expected.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I must have missed the news. So when is Parkes data coming online?

. . Well originally I believe it was supposed to be last September as part of BLC, but now the rumour is this September, but as with all other things, expect it when you see it :) If I had an in at Parkes I would ask them what the scuttlebutt is :)

. . The big question to me is what will their WU format be like and how well with they crunch? :)

. . But because there is no certainty that I know of it was really just a throw away gag line ...

Stephen

:)

Well, my prognostication was all wet. Numbskull has already matched and passed my original fast crunchers. The trajectory slope was a LOT steeper than I had expected.

. . Rejoice brother ... and bask in the glow of hard working GPUs ... :)

Stephen

:)