6700XT distributed computing results

[Endgame124]

Just got my 6700XT installed - I would have gotten a 6900XT had one been available, but you have to work with what the new egg shuffle gives you.

anyone have any requests on what to test?

 

[VirtualLarry]

anyone have any requests on what to test?

Edit: Nevermind, didn't see that this was the DC forum, and not the general GPU forum.

(Sorry Mark, didn't mean to trigger you.)

 

[Markfw]

Just got my 6700XT installed - I would have gotten a 6900XT had one been available, but you have to work with what the new egg shuffle gives you.

anyone have any requests on what to test?

F@H ! forget the nicehash, miners are the reason we have no cards.

 

[Endgame124]

F@H ! forget the nicehash, miners are the reason we have no cards.

F@H is currently running - estimated 570k, but this is my first work unit on this host. Should have better numbers tomorrow.

My first data point: currently running the 6700XT fully stock running F@H, with 5950 in eco mode running 30 threads of Rosetta. Currently pulling 410 watts

 

[Markfw]

F@H is currently running - estimated 570k, but this is my first work unit on this host. Should have better numbers tomorrow.

My first data point: currently running the 6700XT fully stock running F@H, with 5950 in eco mode running 30 threads of Rosetta. Currently pulling 410 watts

I entered the newegg shuffle a couple of times. Once won and got a 3070 for $720 (A little high price) Today, if I win, it will be a 3080 ASUS OC for $1060, probably $200 over MSRP, but way better than ebay. My shopblt order that I placed in Jan is still waiting for my EVGA 3080 at $56 over MSRP (not bad) and another 5950x at $32 over MSRP..

 

[Icecold]

From what I recall Einstein runs well on AMD cards(or at least some) - I'd be interested in seeing that. PPS Sieve, and some of the other various Primegrid projects would be interesting too. I was going to suggest Milkyway@home, but I can't find any double precision specs on it so I don't think it will run MW at all but you could attempt it. Even accounting for OpenCL instead of the higher PPD CUDA F@H app though that seems much lower than I would have expected on F@H so hopefully it evens out after running some more.

 

[Endgame124]

From what I recall Einstein runs well on AMD cards(or at least some) - I'd be interested in seeing that. PPS Sieve, and some of the other various Primegrid projects would be interesting too. I was going to suggest Milkyway@home, but I can't find any double precision specs on it so I don't think it will run MW at all but you could attempt it. Even accounting for OpenCL instead of the higher PPD CUDA F@H app though that seems much lower than I would have expected on F@H so hopefully it evens out after running some more.

For F@H I assume I need to return 10 WUs before I can start getting the quick return bonus. I assume that’s per PC and not per account

 

[Markfw]

For F@H I assume I need to return 10 WUs before I can start getting the quick return bonus. I assume that’s per PC and not per account

I thought it was per account. Give me the unit number(project) and time per frame.

For example, one of my 2080TI's is doing project 13446 in TPF 47 seconds, and the calculator estimates 5251318.46 ppd.

NOW, Make sure the time on your computer is correct. You may get points once the unit is returned, but all estimates are out the door if the time is off, or if your username is not EXACTLY the same (case sensitive) You can look and see a few points for me, when I mis-spelled my name by not capitalizing the F in the middle name.

Edit: this link gets you to what I use: http://www.linuxforge.net/bonuscalc2.php

 

[Endgame124]

looks like 600k ppd... that's not awesome, and much lower than I would expect

 

[Markfw]

looks like 600k ppd... that's not awesome, and much lower than I would expect

So, what project, and how much TPF ? That does sound low.

 

[Endgame124]

1 minute 15 seconds, project 17723

 

[Markfw]

1 minute 15 seconds, project 17723

Yea.... 621k.... But 2060's only do a million. Also was that windows or linux ?

In general, Nvidia owns F@H, and some other projects, AMD OWNs, (mostly the double precision ones)

 

[Endgame124]

This is windows. Picked up a new project -
PROJECT: 17433
TPF: 1 min 49 sec

Just over 2M PPD. Looks like this might just be project specific.

 

[Markfw]

Well, I didn't win the shuffle. and my new 7601 EPYC won't even post.... Probably bios.... Anyway, bad day....

 

[Endgame124]

Still cruising along at 2m ppd. I’ll let it run F@H over the weekend, and then grab a new project

 

[Assimilator1]

From what I recall Einstein runs well on AMD cards(or at least some) - I'd be interested in seeing that. PPS Sieve, and some of the other various Primegrid projects would be interesting too. I was going to suggest Milkyway@home, but I can't find any double precision specs on it so I don't think it will run MW at all but you could attempt it. Even accounting for OpenCL instead of the higher PPD CUDA F@H app though that seems much lower than I would have expected on F@H so hopefully it evens out after running some more.

According to the wiki article for the 6700 XT its DP is 743 GFLOPS, which is comparable to the old HD 7950B/R9 280, which as you know is pretty good at MW.
Assuming the wiki DP numbers for the 6700 is right, and seeing as it's a much more modern architecture it should be much quicker than the Tahiti pro GPU at MW and Einstein?

So yea, my vote goes for running MW and Einstein on the 6700 XT

 

[Endgame124]

Little disappointed in the 6700XT's power target range. I can take it as low as... -6%. If you want to reduce power below that you have to manually reduce voltage and / or reduce clock manually.

Its currently running about 110 watt with a -6% power target and -25mv on my current F@H Project, but the projects vary so much I'm not sure the usage means much. I'm sure with tweaking I can get the power usage lower, it just means more work and more potential for crashing by manually down volting.

 

[Shmee]

Does the latest Phoenixminer work ok on the 6700XT? I would be curious to know how much the 192bit bus hinders ETH hashrate.

 

[TennesseeTony]

We welcome everyone Shmee, but you are off topic and in the wrong place to be asking such a question....

 

[Endgame124]

Does the latest Phoenixminer work ok on the 6700XT? I would be curious to know how much the 192bit bus hinders ETH hashrate.

No idea. Might give it a shot after all the DC projects run out of work units though.

 

[Endgame124]

According to the wiki article for the 6700 XT its DP is 743 GFLOPS, which is comparable to the old HD 7950B/R9 280, which as you know is pretty good at MW.
Assuming the wiki DP numbers for the 6700 is right, and seeing as it's a much more modern architecture it should be much quicker than the Tahiti pro GPU at MW and Einstein?

So yea, my vote goes for running MW and Einstein on the 6700 XT

MW seems to be completing Separation tasks in 1 minute 22 seconds. Has only been running for 5 WU so far though, so the average may change If I let it run a day, is that a large enough sample size?

 

[Skillz]

Is that one task at 1 minute 22 seconds each or are you running multiple tasks at once?

1 minute 22 seconds isn't that great as the good cards can run it in around half a minute, but a minute and a half isn't too bad.

If you would run WUProp@home on your boinc instances your hardware can be benchmarked on the project for others to see what it's capable of doing.

Seems the only 6000 series card so far has been a 6900.

GPU

 

[Endgame124]

Is that one task at 1 minute 22 seconds each or are you running multiple tasks at once?

1 minute 22 seconds isn't that great as the good cards can run it in around half a minute, but a minute and a half isn't too bad.

If you would run WUProp@home on your boinc instances your hardware can be benchmarked on the project for others to see what it's capable of doing.

Seems the only 6000 series card so far has been a 6900.

GPU

Just running one task at a time. I would need a refresher on how to setup multiple WUs at once.
I’ll look into WUProp

 

[Skillz]

WUProp is a NCI project (non-cpu intensive) so you can run it on all your hosts to gather work unit information for all the projects.

 

[biodoc]

Just running one task at a time. I would need a refresher on how to setup multiple WUs at once.

I have this app_config.xml file in my milkyway project directory to run 4 tasks at a time on a RadeonVII.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.05</cpu_usage>
</gpu_versions>
</app>
</app_config>

Change <gpu_usage>0.25</gpu_usage> to <gpu_usage>0.5</gpu_usage> to run 2 tasks at a time.