|
Post by mrpaco on Feb 25, 2018 12:54:29 GMT -5
Some help with this. I can not figure out why I get error when trying to run it on GPU. Running it basically @ stock speeds for testing. Any ideas/suggestions?
|
|
|
Post by Mr.Scott on Feb 25, 2018 13:01:38 GMT -5
AMD SDK 2.7 installed?
|
|
|
Post by mrpaco on Feb 25, 2018 13:07:27 GMT -5
HHmm, good question . Gotta check
|
|
|
Post by Mr.Scott on Feb 25, 2018 13:13:51 GMT -5
If you need it I'll link you to my OneDrive.
|
|
|
Post by mrpaco on Feb 25, 2018 13:24:48 GMT -5
I figured it out, ts the 3.2version of GPUpi thats the problem. I can run 2.3.4v with no problem at all. Would you have the 3.1v of GPUpi?
|
|
|
Post by mrpaco on Feb 25, 2018 13:28:29 GMT -5
Never mind. 3.1 dosnt work ether Will just stick with 2.3.4
|
|
|
Post by Mr.Scott on Feb 25, 2018 13:33:02 GMT -5
|
|
|
Post by Shadyreaper on Feb 25, 2018 13:45:01 GMT -5
I just unzipped 3.2 and it asked about HPET clicked yes it restarted my PC and worked perfect without doing anything with my 980ti damn amd stuff LOL always causing issues
|
|
|
Post by mrpaco on Feb 25, 2018 14:25:36 GMT -5
If I read it right Its 3.1 thats required and thats for those running SR-2 setups. On the download page shows 2.3.4 as latest. In the comp you see subs with 3.1, 3.2 & 2.3.4... Anyway will post in the thread and see what happens....
|
|
|
Post by mrpaco on Feb 25, 2018 14:26:54 GMT -5
I just unzipped 3.2 and it asked about HPET clicked yes it restarted my PC and worked perfect without doing anything with my 980ti damn amd stuff LOL always causing issues App reacts differently between ATI & nvidea GPUs and if running AMD or intel cpu and what OS youre running...
|
|
|
Post by dr4g00n on Feb 25, 2018 15:13:48 GMT -5
Your batch & reduction size are too large. Try Batch = 5M , Reduction = 128 or 256. Also, use GPUPI 3.1 or 3.2, either one works but 3.2 is a tiny bit faster I believe.
Play around with the batch & reduction to find what plays best with your hardware.
|
|
|
Post by mrpaco on Feb 25, 2018 15:38:12 GMT -5
Your batch & reduction size are too large. Try Batch = 5M , Reduction = 128 or 256. Also, use GPUPI 3.1 or 3.2, either one works but 3.2 is a tiny bit faster I believe. Play around with the batch & reduction to find what plays best with your hardware. Its not the batch & reduction sizes.. 3.1 or 3.2 will not run.
|
|
|
Post by ksateaaa23 on Feb 25, 2018 16:07:16 GMT -5
i had a similar problem with version 3.1 version but for gpu pi for cpu. removed all amd drivers & software with amd clean up tool. fresh install driver.
|
|
|
Post by Macsbeach98 on Feb 25, 2018 16:07:17 GMT -5
Driver get the latest driver on it Ellesmere dont work properly with older drivers you want around 17.8 or above OpenCL 2.0 is fine with late model AMD Graphics. GPUPI 3.1 / 3.2 is a wobbly boot works good when works but is troublesome to get working 2.3.4 worked with anything but apparently was bugged with SR2. Wait till you play with GPUPI CPU I had to install Intel openCL to get 3.1 working with 8350 2.7SDK wouldnt work.
Also try Batch 20m Reduction 128 worked best with Fury
|
|
|
Post by Mr.Scott on Feb 25, 2018 16:18:14 GMT -5
Driver get the latest driver on it Ellesmere dont work properly with older drivers you want around 17.8 or above OpenCL 2.0 is fine with late model AMD Graphics. GPUPI 3.1 / 3.2 is a wobbly boot works good when works but is troublesome to get working 2.3.4 worked with anything but apparently was bugged with SR2. Wait till you play with GPUPI CPU I had to install Intel openCL to get 3.1 working with 8350 2.7SDK wouldnt work. Also try Batch 20m Reduction 128 worked best with Fury Good info here. Some of this I didn't know. Just saved me some work. Thank you.
|
|
|
Post by Shadyreaper on Feb 25, 2018 17:39:54 GMT -5
I just unzipped 3.2 and it asked about HPET clicked yes it restarted my PC and worked perfect without doing anything with my 980ti damn amd stuff LOL always causing issues App reacts differently between ATI & nvidea GPUs and if running AMD or intel cpu and what OS youre running... ah yea win10 with intel CPU and Nvidia GPUs it doesnt pick up intel CPU cause I dont have openCL installed for the CPU
|
|
|
Post by mrpaco on Feb 26, 2018 11:17:31 GMT -5
Your batch & reduction size are too large. Try Batch = 5M , Reduction = 128 or 256. Also, use GPUPI 3.1 or 3.2, either one works but 3.2 is a tiny bit faster I believe. Play around with the batch & reduction to find what plays best with your hardware. Turns out you were correct about the reduction size Thanx...
|
|
|
Post by Aleslammer on Apr 6, 2018 9:47:59 GMT -5
New 3.3 GPUPI noticeably faster then 3.1 -- 5.5 water beating 7.0+ LN2 with 8700K. Sorta like the Maxmen rev's!!!! Starting to dislike it.
|
|
|
Post by Macsbeach98 on Apr 6, 2018 10:21:29 GMT -5
Thats the same as GPUPI 3.3 legacy that they are using over at the comp on OCN its smashing the tripe out of 3.2
|
|
|
Post by Aleslammer on Apr 6, 2018 10:32:36 GMT -5
Not a problem in real time but sucks for the 294 existing CPU subs I have. Although do have a bunch that suck using the wrong CL packages.
|
|
|
Post by Vinster on Apr 6, 2018 11:03:50 GMT -5
oh that blows... I'd be pissed Ales.
Vin
|
|
|
Post by Shadyreaper on Apr 6, 2018 11:10:43 GMT -5
I found 3.3 to be slower then 3.2 with my 980ti hmmm need more testing I guess
|
|
|
Post by Aleslammer on Apr 6, 2018 11:18:44 GMT -5
I haven't tried it yet won't run on the S1366 I have setup. HWinfo having a problem detecting the RAM. Then again CPUz 1.84 errors out and have to drop back to 1.78 for it to read voltages and report anything for SPD.
|
|
|
Post by Vinster on Apr 6, 2018 11:32:15 GMT -5
I haven't tried it yet won't run on the S1366 I have setup. HWinfo having a problem detecting the RAM. Then again CPUz 1.84 errors out and have to drop back to 1.78 for it to read voltages and report anything for SPD. I always forget to get the latest CPU-Z.. im in the same socket, we'll see if we both have the same issue. Vin
|
|
|
Post by Bones on Apr 6, 2018 11:59:39 GMT -5
1.84 is bugged to Hell..... Still running 1.83 here. 1.84 hasn't worked in anything I've tried it in so far so stick with 1.83 and you'll be fine.
I can't blame Mat for trying to improve the bench itself, it's just the trouble of having to redo all of your GPU PI runs..... But I personally don't mind it so much. Just an excuse to crank it up again and do bear in mind other teams know it too and will be rebenching their stuff for points, esp Team Russia.
I may not be able to do alot but will do what I can - Get busy guys.
|
|
|
Post by Aleslammer on Apr 6, 2018 13:13:39 GMT -5
Have been using 1.83 but doesn't work right on the setup I'm using (No Voltage or RAM Information).
Benches tell us new hardware is faster than the that proceeding it (Generally)as the bench doesn't change. I've noticed some benches stating don't compare the results of version 1 to 2 as improvements changed the meaning of the result. So what I'm seeing is those that are benching the leading big pointer rigs saying it's OK as it looks good in their short term points (tell the generally public catches up), crap take me more time than it's worth to rebench for a new version. Guess it's like HWBOT Prime someone cheats so everyone loses time effort and points, reminds me of being punished in grammar school. Sorry I'm pissed.
|
|
|
Post by Bones on Apr 6, 2018 14:28:34 GMT -5
You make a valid point for sure, the software in itself doesn't change. Another point is where you said about the big-boy rigs jumping all over it. That too is a pain to deal with, not to mention all the time and trouble it took just to run it before, much less again against those guys. Believe me, it's not fun to drag out everything I'd ran it on before just to redo my runs with it. In truth I don't "Mind" it so much except for the reasons you gave which are good points about it. However..... I also knew one day I'd be revisting it across the platforms I have, I tend to do that and that's why it isn't bothering me as much but in your case I understand. Getting penalized for the acts of a very few or even the one isn't fair yet it's happened there before and will happen again I'm afraid.
|
|
|
Post by Aleslammer on Apr 6, 2018 14:56:02 GMT -5
It's all BS, bench more HWM & TPs tired of losing effort.
Noticing most of the dogs using less chilly modes of cooling, nothing like hoping more jump in to pad the future cold runs that turn most of the old LN2 scores into 0.2 pointers. Noticed a pretty nice 7900x score on the Bot's home page but was using 3.2 so recent effort just became a useless point.
|
|
|
Post by Bones on Apr 6, 2018 18:51:17 GMT -5
I'm in the same boat - I'll have to redo some runs myself.
|
|
|
Post by Mr.Scott on Apr 7, 2018 7:56:30 GMT -5
I'm sorry but I do not share the same opinion. Large score changes have been implemented on other benches before this one. CB R15 and XTU for example. Nobody balked at those.
|
|