A walk to the AMD side

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,201,844,931
RAC: 916,332

cecht wrote:When you do,

cecht wrote:
When you do, don't forget to flip the dual BIOS switch to the mining position...

It was on my list of things to try--entirely thanks to your posts here.  Still will be when I get the card inserted into a working machine.

Quote:
$150 is a good price for a new card. I'm in the market for two XFX 570s and have been perusing eBay, but will now check out Uncle Jeff's place.

Jeff prices not only by the day, but by the customer, I hear.  I indeed paid $150 for a March 17 order.  Were I to order another today (March 22), I'd pay Jeff $155.  I paid him $170 for the February 20th order.

koschi
koschi
Joined: 17 Mar 05
Posts: 86
Credit: 1,671,877,555
RAC: 431,641

cecht wrote:Sorry for your

cecht wrote:
Sorry for your down time. Hope you get that new card crunching soon.  When you do, don't forget to flip the dual BIOS switch to the mining position.  As far as I know, XFX are the only 570 cards that offer that golden opportunity for faster tasks and lower power with a pre-loaded mining BIOS.

 

My Sapphire RX580 Nitro also has an alternate BIOS, they call it "Compute", but I guess its just the same thing. Reduced clocks and PowerCap with little to no impact on WU completion times. I rune mine as primary in my workstation, no troubles displaying the desktop, Youtube (which my GTX1060 has Problems with under load), etc...

This card currently completes 2 WUs every ~1050 seconds, resulting in ~570000 credits per day.

Doing so, it consumes 80-90W (I'm still tuning). The mining BIOS uses 2075MHz mclk, I found that to be a bit unstable though, resulting in sporadic computational errors chewing Einstein FGRP. At 2000Hz mclk its very stable, currently I'm trying to up the GPU clock a bit, it seems to pay out...

cecht
cecht
Joined: 7 Mar 18
Posts: 1,519
Credit: 2,858,069,520
RAC: 1,960,121

koschi wrote:[My Sapphire

koschi wrote:

[My Sapphire RX580 Nitro also has an alternate BIOS, they call it "Compute", but I guess its just the same thing. Reduced clocks and PowerCap with little to no impact on WU completion times. I run mine as primary in my workstation, no troubles displaying the desktop, Youtube (which my GTX1060 has Problems with under load), etc...

This card currently completes 2 WUs every ~1050 seconds, resulting in ~570000 credits per day.

Doing so, it consumes 80-90W (I'm still tuning). The mining BIOS uses 2075MHz mclk, I found that to be a bit unstable though, resulting in sporadic computational errors chewing Einstein FGRP. At 2000Hz mclk its very stable, currently I'm trying to up the GPU clock a bit, it seems to pay out...

That's some nice performance and power efficiency!  When you posted earlier about the Sapphire dual bios, I looked it up. Their web site describes the BIOS switch as Quite vs. Performance with respect to gaming performance and no mention of mining, so that's why I had thought XFX was unique with their mining bios. But am glad to hear I was wrong!

Ideas are not fixed, nor should they be; we live in model-dependent reality.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,201,844,931
RAC: 916,332

archae86 wrote:So I'm

archae86 wrote:
So I'm inclined to leap forward to the stripped down minimalist machine "proof of life" check.

It took a while, but:

1. I removed all four cards mounted in the machine.

2. I disconnected all five drives from both SATA data and SATA power

3. I connected only keyboard, mouse, monitor and wall power to the box

And that worked, properly--the first time I had seen pixels on the monitor stimulated by the motherboard monitor connection since this all started.

So I hooked up the two hard drives (but not the SSD used by Intel RST).

I still got monitor pixels, but to move forward I had to instruct RST that I wanted it to quit trying to use the SSD that was not there to accelerate the HD that was.  After that interaction, the machine believed it had a boot drive.

Miraculously, one boot after it told me there had been an improper shutdown but it could find nothing to fix, I had a normal-seeming machine (albeit with no graphics cards, no SSD for acceleration, no optical drives, no sound card, no fax/modem).

At this stage, I had a really jolly thought: the problem was the SSD, which had failed in some way that took out normal system function with it.  I could do without SSD acceleration in a system which mainly serves as an alternate host, a backup target, and a two-card Einstein host.

So, bright with hope I screwed in one of the two graphics cards which had formerly been in the machine.

Oops--no pixels on the monitor. 

Aha, maybe the 1050 card died in a way that stopped the motherboard working.  Maybe the SSD was fine after all.

So I pulled out the 1050 card and put the 1060 card which had been living in the lower PCI-E slot in the place it had lived.

Again no pixels.

So, at the moment I think the following former suspects are not likely primary causes: the SSD drive, the 1050 card.

I think it likely that something on the motherboard has failed in a way which kills the system when at least some subset of working cards are plugged into some subset of slots.

With that diagnosis, I'm currently back on a short fuse to ordering parts to build a largely new replacement system, using the old optical drives, a fan, the power supply, and not much else.

As I typed "power supply", I could hear Gary's voice whispering "but maybe the power supply is the problem, but it takes above some threshold of loading to trigger it".  To which I reply, the current lower-end Nvidia cards take something like 5 watts of idle power.  I just don't think that likely enough use my time to take apart a system I am still decommissioning in order to extract a "known good" supply for verification purposes.

Still, I'll wait about 16 hours before placing any orders, in case I have another thought on my own, or one of you suggests something I think worth trying first.

Just in case it is of any interest, the more interesting of my "cheapish but modern and not low-end" choices at the moment are:

case: Antec Three Hundred Two (lots of fan positions, enough room for a longish graphics card)

CPU: Intel Coffee Lake i5-9400F  (hex core, no graphics, lowish power, not slow, no HT, cheap by Intel standards)

motherboard: Asrock B360 Pro4 (cheap because no overclocking support, slowish maximum RAM speeds)

Boot drive: Intel 660p 1TB SSD drive (very, very cheap for the capacity, fast enough for my purpose)

Yes, I understand that the SSD drive won't help Einstein speeds.  I do like the prospect of consistent fast boot times and application launch.  And the 1TB size should help keep at bay my "not big enough anxiety" which kept me away from SSD boot drives after my first try four systems ago in 2011, when an 80Gb drive turned out to be not enough for Windows 7 to be comfortable for me long term.

If all goes extremely well, I'll have the new system built and running Einstein on my second RX 570 by next week, though it will be weeks after that before I finish with configuring, and re-acquiring use of or abandoning hope of applications.

 

 

 

 

 

mikey
mikey
Joined: 22 Jan 05
Posts: 12,634
Credit: 1,839,015,786
RAC: 6,138

archae86 wrote:archae86

archae86 wrote:
archae86 wrote:
So I'm inclined to leap forward to the stripped down minimalist machine "proof of life" check.

 

I still got monitor pixels, but to move forward I had to instruct RST that I wanted it to quit trying to use the SSD that was not there to accelerate the HD that was.  After that interaction, the machine believed it had a boot drive.

Miraculously, one boot after it told me there had been an improper shutdown but it could find nothing to fix, I had a normal-seeming machine (albeit with no graphics cards, no SSD for acceleration, no optical drives, no sound card, no fax/modem).

At this stage, I had a really jolly thought: the problem was the SSD, which had failed in some way that took out normal system function with it.  I could do without SSD acceleration in a system which mainly serves as an alternate host, a backup target, and a two-card Einstein host.

So, bright with hope I screwed in one of the two graphics cards which had formerly been in the machine.

Oops--no pixels on the monitor.

I have the same problem whenever I put in a gpu that is higher than the bios can handle. I ALSO have the same problem for no reason whatsoever sometimes and even replacing the psi, the gpu, the memory and the hard drive have done nothing to fix it. I'm down to the motherboard being bad so I replaced it one time and even that didn't work for me!! Now I just put the parts in the parts bin and use them as replacements when others go bad, seems some boxes just aren't meant to crunch for years and years.  And yes I even took the parts out of the case and tried that way too, no joy for me. Sometimes computers are just frustrating!!

koschi
koschi
Joined: 17 Mar 05
Posts: 86
Credit: 1,671,877,555
RAC: 431,641

cecht wrote:That's some nice

cecht wrote:
That's some nice performance and power efficiency!  When you posted earlier about the Sapphire dual bios, I looked it up. Their web site describes the BIOS switch as Quite vs. Performance with respect to gaming performance and no mention of mining, so that's why I had thought XFX was unique with their mining bios. But am glad to hear I was wrong!

 I believe they changed the naming from Silent to Compute mode around autumn last year. On the box it now says "Optimized for Compute."

RX580 compute mode

 

I am currently trying to get WU completion times down to 997 seconds, so that the card would be able to produce 600000 points per day. Consumption rises quickly though, I'm already at 102W for a run time of 1015-1020 seconds (2 WUs). Long term I'll most likely operate it at 82W and 1050-1070 seconds.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,157
Credit: 7,201,844,931
RAC: 916,332

archae86 wrote:If all goes

archae86 wrote:
If all goes extremely well, I'll have the new system built and running Einstein on my second RX 570 by next week, though it will be weeks after that before I finish with configuring, and re-acquiring use of or abandoning hope of applications.

I'm typing these words on the new machine I built to accept the RX 570 which I was just at the point of installing in the previous machine when it died on me.

One thing I tried for the first time seems to have worked out nicely.  I copied the complete BOINC directory from the old machine to a USB drive before shutting down for the last time, and then copied that under ProgramData on the new machine before running the BOINC install.

When I ran the BOINC install, things started right up.  It knew I was signed up to Einstein, and got me a bit of work and started running it.  It renamed my historic records for the machine to my new machine name as I see it on the network (Stoll10 in lieue of Stoll7), but kept the previous Einstein number computer ID 10706295--no merge request needed.

I've plenty of configuration and installation details yet to go, but no great hurry to do so, as it is up and running Einstein, and has even had  a couple of validations.

I'm not sure I did it the specific way he suggests, but the thought of jump-starting BOINC by pre-populating the directory tree from the previous machine came to me from something Gary Roberts has commented on in the past, and for me, so far, it has worked a treat.

BoincTasks from another machine could even see this machine right away once I thought to ask it to look.  the annoying task of setting up the permission was already done by files that were part of the implant.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,870
Credit: 116,848,082,939
RAC: 36,273,526

archae86 wrote:I'm typing

archae86 wrote:
I'm typing these words on the new machine I built to accept the RX 570 which I was just at the point of installing in the previous machine when it died on me.

Sincere congratulations on the safe 'birth' of your new 'baby' :-).

archae86 wrote:
I'm not sure I did it the specific way he suggests, but the thought of jump-starting BOINC by pre-populating the directory tree from the previous machine ... has worked a treat.

Well done!!  There's no 'specific' way - if it worked, it was the 'correct' way :-).  I like it because it saves a lot of fiddling around trying to get things 'right' and then all the potential mistakes avoided because of some 'senior's moment' :-).

I hope all goes well with the new RX570!

 

Cheers,
Gary.

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 1,055,935,564
RAC: 0

archae86 wrote:I'm not sure I

archae86 wrote:

I'm not sure I did it the specific way he suggests, but the thought of jump-starting BOINC by pre-populating the directory tree from the previous machine came to me from something Gary Roberts has commented on in the past, and for me, so far, it has worked a treat.

BoincTasks from another machine could even see this machine right away once I thought to ask it to look.  the annoying task of setting up the permission was already done by files that were part of the implant.

When retiring a machine and replacing it, that's the way I do it. Copy the whole Boinc data directory from the old to the new and placing it where I want it. Then I run the Boinc installer and using the advanced install I make sure the path to the data directory corresponds with where I copied it to.
This way everything just picks up where it was unless one makes a fundamental change of operating system, like switching between Linux - Windows - macOS.

kb9skw
kb9skw
Joined: 25 Feb 05
Posts: 21
Credit: 376,184,005
RAC: 9,364

So are we in agreement that

So are we in agreement that lower GPU clock speed and higher memory clock speed produce faster crunch times?

I just checked my XFX, it has been in the mining bios so I switched it to the gaming bios just to see what the difference would be. I should think the RAC will drop a little bit, currently at 911,186.32. I really need a quad core CPU to run the four two cards, but they are in an old E2200 dual core Pentium. That is probably reducing my RAC a bit I suspect.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.