Posts by QuintLeo

1) Questions and Answers : Wrapper and Client : HOW TO CHOSE a especific application to work on ? im receiving 2 kinds.. and onde just crash my pc... i wish to chose only one.. please help ati_14 and opencl+ati_101 (Message 7875)
Posted 10 days ago by QuintLeo
Post:
The openCL one is for GPU.
The other one I THINK is for CPU.
WHICH ONE is crashing your PC?
2) Questions and Answers : Windows : Hello Admin, can you update the app already? (Message 7874)
Posted 10 days ago by QuintLeo
Post:
I'm pretty sure that the site admin IS the project admin for this particular project.
3) Message boards : Number crunching : Legacy Dual GPUs, BOINC assigned work to both but only one with load (Message 7873)
Posted 10 days ago by QuintLeo
Post:
I believe there is an "ignore intel" option among the GPU ignore commands?
That's an option in config.xml though not app_config.xml
It's wierd that the client would ONLY work on the Intel iGPU though - it should be using both the Intel AND the AMD GPUs in the system if you have allgpus set.
4) Message boards : Number crunching : Vega (Message 7872)
Posted 10 days ago by QuintLeo
Post:
Bumping memory up for Moo Wrapper work is a waste - the entire Dnet client AND data fit into cache memory on anything remotely recent in a GPU.
5) Message boards : Number crunching : How to throttle GPU? (Message 7871)
Posted 10 days ago by QuintLeo
Post:
GPU Driver install or upgrade on windows ALWAYS requires a reboot - often 2 or 3 of them - before the new driver starts working correctly if at all.
That's been the case back AT LEAST to Windows For Workgroups 3.10 (BEFORE 95).
6) Message boards : Status info : Latest ATI/nVidia requirements (Message 7870)
Posted 10 days ago by QuintLeo
Post:
Just had a machine lose 2 days of work because you STILL have not updated the client from the BUGGY 520 version to the current 521 version more than a YEAR after 521 was released to FIX ISSUES with 520.

The FIX is just a drop-in of the 521 client for the 520 client, why has this taken so LONG to fix?
7) Message boards : Number crunching : Vega (Message 7827)
Posted 4 Dec 2017 by QuintLeo
Post:
Have you optimised it for Dnet/Moo?

What I found on my 56 was that it wanted quite a bit of undervolt, along with the usual "drop the ram speed to the minimum" and a bit of set power limit up, to get optimal keyrate while keeping temps fairly low.
8) Questions and Answers : Windows : how to get 1.04 (opencl_nvidia_101) instead of 1.03 (cuda31) (Message 7816)
Posted 25 Nov 2017 by QuintLeo
Post:
The only way I've found to get a NVidia card to use OpenCL on a reliable basis with Moo is to download a copy of the 521 OpenCL client from the Dnet clients download page, rename the .exe file to match the 520 client name in the moo wrapper "projects/moowrap.net" directory, then copy the renamed 521 version client to that directory overwriting the 520 client that is already there.

This will only work as long as BOINC does not get rebooted - if it DOES get rebooted, you'll have to do the copy overwrite again.

I'm not sure how much of a difference the GTX 750 ti has in performance under CUDA vs under OpenCL - the only thing I can remember the numbers on for that comparison is the GTX 1080 ti, which on my Aorus went from around 6-6.5 GKeys/sec to a hair over 10 on the same settings.
9) Message boards : Number crunching : Nvidia OpeCL? (Message 7815)
Posted 25 Nov 2017 by QuintLeo
Post:
For a machine that doesn't game, THE driver version that combines best performance AND best reliability is 373.06 - this comes out of extensive testing of driver versions by the FAH community, but should be applicable here as well.
Unfortunately, if you use a GTX 1080 ti or more recent (like the newest Titan or the 1070 ti) you HAVE to use something more recent - I'd go with the 384.98 (November 2 update) version for LINUX, not sure on the WIndows side as my few remaining WIndows/NVida machines are all running older cards on 373.06

Newer on NVidia drivers are often NOT BETTER for compute usage, and a lot of the versions since 373.06 have been actively BROKEN for compute.


The GTX 1070 is an 8GB card - there is no option for anything else on ANY version of it.
Doesn't really matter though, as the underlying Dnet client only uses perhaps *A* megabyte if that much including data usage.


The only way I've found to date to RELIABLY get a GTX 10xx series card to use OpenCL under Moo is to do the "copy the 521 client in place of the 520 client that Moo uses" trick, which only works 'till BOINC gets rebooted - it's one of the reasons I only run AMD cards on MOO, it's not worth the hassle to get my NVidia farm (almost all 1070/1080/1080i/1070ti cards, though I do have a FEW 950/960 left-overs) to work with Moo as long as Moo continues to use a client that is outdated and where the CURRENT client fixes quite a few issues.
10) Message boards : Number crunching : Problems running wus on AMD R9 380. Suddenly Moo crashes card. (Message 7814)
Posted 25 Nov 2017 by QuintLeo
Post:
Oddly enough, I HAVE to undervolt on my Vega RX 56 to get the best keyrates - and lower temps.

Seems like the "default" GPU voltage on those things is set WAY high.

Not sure how long I'm going to leave the thing "lent" to my cryptocoin mining effort, as the only thing it's worth mining with is Monero and you HAVE to run Win10 and the blockchain driver to get decent hashrate out of it on that (the LINUX version of that driver flat out doesn't WORK with Vega, in fact AMD has had the FE out for MONTHS and it STILL doesn't work under LINUX due to their borked drivers situation on the LINUX side).

On the RC5-72/Dnet/Moo/BOINC side, it's about a 7.5ish GKey/sec card stock and can easily be tweeked to hit 9 reliably - that's a bit more than a PAIR of RX series cards or my old-but-reliable R9 cards, while the power consumption is a little less than a pair of my RX 470 cards.
It also works with 520 under OpenCL as MooWrapper provides, no need to "kludge" it like for a GTX 1080 ti to get it's good hashrate.
11) Message boards : Number crunching : Legacy Dual GPUs, BOINC assigned work to both but only one with load (Message 7790)
Posted 9 Nov 2017 by QuintLeo
Post:
What is the "max concurrent" line for?
I've never had that in any of my multi-GPU systems.
12) Questions and Answers : Windows : 2nd GPU issue (Message 7786)
Posted 8 Nov 2017 by QuintLeo
Post:
Even when running the actual DNet client, 2 GPUs won't "share work" - they are always working on different keyblocks.
What you *should* be seeing is each GPU working on a seperate workunit, and taking about the same time that they did before.
13) Questions and Answers : Windows : Hello Admin, can you update the app already? (Message 7773)
Posted 3 Nov 2017 by QuintLeo
Post:
The NVidia issue is that the Native CUDA cores are a LOT slower then the newer OpenCL cores - but the clients prior to 521 don't SUPPORT OpenCL usage on NVidia.

1080ti for example it's a difference from about 5-6 GKeys/sec to around 10 GKeys/sec - not as bad as your 970 but still a LOT.

It seems like 521 cleaned up some of the interaction with OpenCL driver code so it just WORKS in a lot of cases were the 520 client (or older) don't.

Since the client is literally a drop-in replacement, I don't understand why this small change wasn't made MANY MANY MOONS AGO - shouldn't take much of a change to tell whatever overrwites the client to use the newer client instead?
14) Questions and Answers : Windows : AMD R9 380 - Computation Errors - Distributed.net1.04 (opencl_ati_101) on dnetc_r72_1499529280_13_823_0 (Message 7772)
Posted 3 Nov 2017 by QuintLeo
Post:
One other possible option - grab the 521 OpenCL client from the Distributed.net downloads, rename the exe to match the 520 .exe client in data/projects/Moowrapper.net, then copy the renamed 521 client over the existing 520 client.

I had one machine that just wouldn't run on the 520 client but ran fine on the 521 for some wierd reason I never did figure out.
15) Message boards : Number crunching : Vega (Message 7771)
Posted 3 Nov 2017 by QuintLeo
Post:
Got my Vega 56 in today (Sapphire reference model).

It's WIERD - to get it to get anywhere near the officially specified "1590" core clock, you have to kick the power limit WAY up (+30% at least in Afterburner) AND you have to UNDERVOLT it quite a bit (-100mv ballpark) - and even THEN it won't get to 1590, though I did see mid-1500s after a bunch of settings work.

In stock form, it's not quite a 8 Gigakey/sec card (for perspective, the GTX 1080 ti can hit a little over 10 GKeys/sec with some overclocking).
Maxed out, I saw about 9.5 GKeys/sec though - which is not bad at all for a $460 card compared to the 1080 ti at $700+, and quite a bit more than the GTX 1080 manages in the same price ballpark.

Card also runs AMAZINGLY cool unless you push it very hard - never much past 55C at stock settings with less than 55% fan, and the settings I found to make it a 9GKey/s card was only pushing it to about 62-63C at under 68% fan (I run a straight-line "fan curve" in Afterburner by default that is "all in" at 80C and drops 20% every 10C below that).

I estimate that a stock 56 will exceed 1 million RAC by a bit, a pushed one should get into the 1.3-1.4 million RAC ballpark - in other words, a SINGLE CARD can exceed all but about the top 3 or 4 machines currently working Moo Wrapper.

Given wide reports that the air-cooled Vega 64 clocks higher reliably than the 56, and that it has 64 CU instead of 56 (1/7'th more cores), I am ready to state that the Vea 64 SHOULD be the current Dnet/Moo Wrapper single card performance king - especially if someone wanted to put out the kind of money to get the WATER COOLED version, which supposedly clocks a LOT higher than the air-cooled version.
16) Message boards : Status info : Latest ATI/nVidia requirements (Message 7757)
Posted 27 Oct 2017 by QuintLeo
Post:
The distributed.net client ITSELF does not support NVidia well if at all in versions prior to 521.
Yet another reason MooWrapper NEEDS to get updated to 521 asap - it fixed a TON of issues vs the version 520 client MooWrapper is currently using, and offers a LOT better performance on NVidia cards.
17) Questions and Answers : Windows : AMD R9 380 - Computation Errors - Distributed.net1.04 (opencl_ati_101) on dnetc_r72_1499529280_13_823_0 (Message 7756)
Posted 27 Oct 2017 by QuintLeo
Post:
Try backing out to the 16.9.2 or the 15.12 drivers - the Relive junk has a LOT of issues where it works at all, and a LOT of the newer driver versions flat out don't WORK with R9 series hardware.
18) Questions and Answers : Windows : Hello Admin, can you update the app already? (Message 7755)
Posted 27 Oct 2017 by QuintLeo
Post:
download the Distributed.net 521 client from https://www.distributed.net/Download_clients
Unzip the downloaded file
rename the dnetc.exe file to match the long file name dnetcxxxxx.exe file in the boinc projects/moowrapper.net directory, then copy it in place of the existing file.

Since moowrapper WILL overwrite the file on occasion, keep the renamed copy of the 521 client file around for later recopy.
19) Message boards : Number crunching : Nvidia OpeCL? (Message 7739)
Posted 6 Oct 2017 by QuintLeo
Post:
I'm not sure if OpenCL on NVidia is properly supported by the dnet 520 version of the client Moo Wrapper uses - it's one of the reasons I've been strongly in favor of Moo Wrapper upgrading to use of the CURRENT 521 dnet client, which definitely DOES work correctly with NVidia.

You may also have an issue with the driver version you have not having initiated OpenCL support correctly - there were a couple of NVidia driver versions that had that issue, backing out to around 378 seems to help (and if you're not GAMING on the same machine they run more reliably as well).
20) Message boards : Number crunching : How to throttle GPU? (Message 7738)
Posted 6 Oct 2017 by QuintLeo
Post:
It's just the way mixed GPUs and their drivers interact - or the Collaz application was soaking a lot of CPU cycles as well (like FAH does).


Next 20


 
Copyright © 2011-2018 Moo! Wrapper Project