A walk to the AMD side

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7023314931
RAC: 1824054

I am terminating my

I am terminating my experiment to see whether daily rebooting would make my Windows RX 570 machines 3X friendly.  While I really did observe again a quite inviting almost 10% productivity improvement over the same machine at 2X (unlike Gary and cecht), my second crash to sloth mode came less than four hours after today's reboot, on just the third day after the minor Windows update which I speculated might have been a forgivable reason for the first crash to sloth node.

Regarding mode, on reviewing the machine state a little more closely before terminating things, I realized that since the collapse the machine had been running two tasks in sloth mode, with a very repeatable elapsed time of about 1:12:24 and CPU time of just 0:00:17 (!?), but the third task was in catatonic mode.  So the productivity loss to sloth mode was even worse than I thought when I imagined it had been running three tasks at the sloth rate.

Oh, one last thing.  When I woke up the PC which had two tasks running in sloth mode and one in catatonic mode, my monitor just showed black and while hash (a little like old TV sets showed when no station was being received).  I've seen this or something similar a few times with both of my RX 570s, and so far each time just turning off the monitor and immediately powering it back up has fixed things, as it did this time.  I suspect this to be a separate problem from the sloth/catatonic mode problems.  I don't like it either, but now that both my wife and I know a so-far-successful workaround it is tolerable.

[Edit on May 19 to add that this second try at 3X running on this machine produced two "Validate Errors".  That is the bad kind that flunks a sanity test before ever actually being compared with the result from a quorum partner.  One of those was a unit that was running at the moment it dropped to sloth mode, but the other was returned hours earlier when things appeared to be running normally, and time since most recent reboot was well under a day.]

 

 

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109386166811
RAC: 35921809

I'm sorry that you weren't

I'm sorry that you weren't able to find a reliable way to take advantage of the extra productivity of 3x.

The three RX 570 hosts I set up to use 3x continue to produce good results with little variation in crunch times.  The averages remain essentially unchanged from what I listed previously.  There have been no compute errors and the ratio of invalids to valids has moved up and down at times but on average has effectively not changed much overall.  I looked at several other machines with older CPUs from the 2008-2010 period and, somewhat surprisingly, they had marginally better crunch times of around 583-587 secs/task when crunching at 2x.  I didn't try them at 3x and I will soon revert the three that I did change, back to 2x.

I also 're-discovered' two machines in which I had installed a single Asus RX 580.  One had a Q6600 quad core CPU and the other a Q8400 quad core.  I put these two on 3x to see what would happen.  Before changing, I had noted that the 2x crunch times were also very tightly grouped around an average, with little variation from that average.  The averages for these two with RX 580s are shown in the table below.

Immediately after setting 3x, there was an obvious change in that the crunching rates for individual tasks changed markedly - quite in contrast with the very stable crunching rates for the 570s at 3x.  Both showed very much the same type of behaviour.  I let them run that way for about 15 hours before putting them back on 2x. Individual task times varied between about 1150 secs and 2200 secs, with many in the 1500-1900 range.

I went to the trouble of taking 40+ consecutive tasks as listed in the online database (validated or not) in order to work out some sort of average.  Because of the huge variability of individual times, I have doubts about how sensible the numbers (shown below) might be.  Despite the weird behaviour, there were no computation errors and no validate errors for any results so far.  There are lots of validations because I was using a 2.5 day cache size and relatively 'old' tasks were being crunched.  2.5 days was a precaution I'd put in place before the Pentathlon event got started and I haven't wound it back yet :-).

       CPU      2x Time - Per Task   3x Time - Per Task    Invalid   Valid   Ratio   Uptime
       ===      ==================    ==================    =======   =====   =====   ======
       Q6600      1144  -   572         1724  -   575            15     927    1.6%    67 d.
       Q8400      1128  -   564         1727  -   576            14     808    1.7%    67 d.

Cheers,
Gary.

QuantumHelos
QuantumHelos
Joined: 5 Nov 17
Posts: 190
Credit: 64239858
RAC: 0

archae86 wrote:I am

archae86 wrote:

I am terminating my experiment to see whether daily rebooting would make my Windows RX 570 machines 3X friendly.  While I really did observe again a quite inviting almost 10% productivity improvement over the same machine at 2X (unlike Gary and cecht), my second crash to sloth mode came less than four hours after today's reboot, on just the third day after the minor Windows update which I speculated might have been a forgivable reason for the first crash to sloth node.

Regarding mode, on reviewing the machine state a little more closely before terminating things, I realized that since the collapse the machine had been running two tasks in sloth mode, with a very repeatable elapsed time of about 1:12:24 and CPU time of just 0:00:17 (!?), but the third task was in catatonic mode.  So the productivity loss to sloth mode was even worse than I thought when I imagined it had been running three tasks at the sloth rate.

Oh, one last thing.  When I woke up the PC which had two tasks running in sloth mode and one in catatonic mode, my monitor just showed black and while hash (a little like old TV sets showed when no station was being received).  I've seen this or something similar a few times with both of my RX 570s, and so far each time just turning off the monitor and immediately powering it back up has fixed things, as it did this time.  I suspect this to be a separate problem from the sloth/catatonic mode problems.  I don't like it either, but now that both my wife and I know a so-far-successful workaround it is tolerable.

[Edit on May 19 to add that this second try at 3X running on this machine produced two "Validate Errors".  That is the bad kind that flunks a sanity test before ever actually being compared with the result from a quorum partner.  One of those was a unit that was running at the moment it dropped to sloth mode, but the other was returned hours earlier when things appeared to be running normally, and time since most recent reboot was well under a day.]

 

In regard to sloth mode the X280 does cap at 500Mhz some times when overheating..

Perhaps the X570 does the same thing ? Bios upgrade!

Stefan Ledwina
Stefan Ledwina
Joined: 23 Oct 05
Posts: 17
Credit: 2410973169
RAC: 28607

Are the problems with "sloth

Are the problems with "sloth mode" and "catatonic mode" only with windows or also with linux? I got a Radeon VII a few days ago, put it in a Win 10 PC and I got the same problems.

If it´s only with windows I would consider to install a dual-boot with Linux just for crunching... 

mikey
mikey
Joined: 22 Jan 05
Posts: 11888
Credit: 1828067059
RAC: 206328

Stefan Ledwina wrote:Are the

Stefan Ledwina wrote:

Are the problems with "sloth mode" and "catatonic mode" only with windows or also with linux? I got a Radeon VII a few days ago, put it in a Win 10 PC and I got the same problems.

If it´s only with windows I would consider to install a dual-boot with Linux just for crunching... 

I don't know but in Windows the only way to get back to full speed is to reboot the pc as Windows doesn't provide a way to do it. Their drivers don't crunch so don't have the problem.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109386166811
RAC: 35921809

Stefan Ledwina wrote:Are the

Stefan Ledwina wrote:
Are the problems with "sloth mode" and "catatonic mode" only with windows or also with linux?

I'm using Linux exclusively.  The two 'modes' that have been mentioned seem to show up for RX 570 GPUs when attempting to run three concurrent tasks under Windows.  I don't see the same behaviour under Linux when running 3x.  I just see no improvement in output - so no incentive to go beyond 2x.

I also see a difference in behaviour between an RX 570 and an RX 580.  A Radeon VII is bound to be 'different' again.  I doubt that the observations reported for Polaris GPUs will have any real guaranteed bearing on how a Radeon VII might behave under the different OS's.  You'll probably just have to suck it and see :-).

Apart from his RX 570s, Archae86 has a Radeon VII running under Windows.  There is a separate Radeon VII thread where he has posted extensively about his experiences with that GPU.  I don't really recall him talking about 'sloth' and 'catatonic' there but, since I don't own a Radeon VII, I haven't been paying really close attention.

Cheers,
Gary.

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7023314931
RAC: 1824054

archae86 wrote:Gary Roberts

archae86 wrote:
Gary Roberts wrote:
I don't know what the 2nd hand market is like in your neck of the woods, but would it be possible to get enough for the 1060 + 1050 to pay for a RX 570 - particularly if you could get one for $140?

Your comment motivated me to browse around on eBay auctions.  It appears likely that I could get appreciable money for my RTX 2080, some money for my Founders Edition GTX 1070.  Even the 1050 and 1060 cards seem to have a pretty active auction market for sales of used cards by individuals likely to get $50 or a bit more.

Long after Gary made his suggestion in late February, I'm shipping out today the last of the six reasonably modern cards that I sold.  All went on U.S. eBay, for the following amounts:

$80  1050 2GB
$52  1050 2GB
$136 1060 6GB
$86   1060 3GB
$213 1070 Founders Edition
$583 2080 8GB 

Prices for the lower-end Pascal cards dropped quite a bit in the three months between my first batch of sales and my second.

As it happens, I almost exactly funded my three new AMD cards (Radeon VII and two of RX 570) for which I paid $1170, with my sales realizations from five Pascal and one Turing card of $1151.  I'd have been well ahead had I paid retail for my Radeon VII rather than paying a scalper extra when I feared the stock was limited.  The rumor that the Radeon VII cards were a one-time vanity build in extremely limited number turned out to be false.

My revised flotilla of three AMD cards produces a great deal more Einstein Gamma-Ray Pulsar work than my previous setups of four or five of these Nvidia cards.  The power productivity is also considerably improved.  The prices I paid for the two RX 570 cards ($150 and $170) were low enough that if the soon-to-be announced AMD Navi cards turn out to do something wonderful at Einstein, I'll not suffer much buyer's remorse in replacing them.

[Edit later to confess--in claiming near break-even I neglected to consider the appreciable eBay and PayPal fees, which probably dropped my net realization to barely over $1000.]

solling2
solling2
Joined: 20 Nov 14
Posts: 219
Credit: 1563256630
RAC: 48358

A memory tweak tool for your

A memory tweak tool for your AMD GPU in both Windows and Linux version:

https://github.com/Eliovp/amdmemorytweak

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109386166811
RAC: 35921809

archae86 wrote:Long after

archae86 wrote:
Long after Gary made his suggestion in late February, I'm shipping out today the last of the six reasonably modern cards that I sold  ....


I'm really pleased (and quite relieved) that this whole exercise seems to have worked out well for you.  I always worry about being 'too enthusiastic' in pushing my own particular point of view.  I tend to 'rabbit on' quite a bit, sometimes even when not in full possession of all the relevant facts, so the worry is that it all might turn into a disaster for some poor soul who is persuaded to follow the advice.

However, I was very confident that the RX 570 would deliver for you in at least two of your key criteria - more output for your hosts and an overall improved power metric.  I wasn't confident about the noise factor since with my forced ventilation system pretty much drowning out any individual host noise, I'm not really in a position to assess that :-).

archae86 wrote:
I'd have been well ahead had I paid retail for my Radeon VII rather than paying a scalper extra when I feared the stock was limited.  The rumor that the Radeon VII cards were a one-time vanity build in extremely limited number turned out to be false.


This is a feature of the modern world that drives me insane.  Maybe I was more immune to it when I was much younger and just didn't notice it but these days it seems you are constantly bombarded with ever more sophisticated methods of ever more dubious morality for parting you and your money in the absolute shortest possible time.  The whole supply chain conspires with ultimate cunning to convince you that you WILL miss out if you don't buy right now.

This annoys me sufficiently to feel compelled to comment further about it.  The comments are not at all directed at archae86 but rather at the general readership.  I know he is more than capable to decide these matters for himself without any assistance from me. They are essentially a plea to everybody to not let misinformation, hype and highly dubious 'facts' unduly sway your decisions about this stuff.

I can understand why people are at the mercy of the scalpers for something like a once-in-a-lifetime opportunity to attend something very important to you where the venue size is fixed and it's likely never to happen again.  In that sort of case, if you didn't get in early, there really is no alternative.

I'm astounded that these mongrels are cheeky enough to use this tactic with something like a new GPU.  Sure, if enough rumours can be created about how good the device will be and that the demand will far exceed the supply, and that the supply is truly limited, then it will probably become a self-fulfilling prophesy.  Even if people are convinced there will not be further fab time immediately available to allow more units to be produced, you have to counter that with the fact that a manufacturer will ALWAYS find a way to produce more whilst the demand rages on.  You might have to wait a bit but you will be able to get what you want at the proper price.

For this particular case you have to ask the question as to how the scalpers could be so smart as to know that here was a product on which good money could be made without any risk of being stuck with it.  My theory goes something like this.  It has become almost the norm for there to be extreme hype accompanied by some sort of restricted availability at launch time.  After all, why wait longer to build up full inventory levels and have a later launch date if you can launch earlier and create an artificial shortage with the associated frenzy to 'get in quick'?

So, somewhere in the supply chain, an opportunity was seized.  A decision was made to siphon off part of the much touted 'limited availability' for the standard retail outlet point and thereby create an even worse availability picture.  Then, this 'siphoned off' portion was made available at much higher price through 'alternative outlet channels', eg. something like ebay.  If the worst comes to the worst and few people take up the scalper's offers, just put the excess back in the normal supply chain at the normal price.  It's a 'win/no lose' for people able to manipulate the supply chain in this way.  Something like this did seem to happen, if you think about it.  After a period of virtually no availability, suddenly there was availability, ie. the 'leftovers' were returned to the normal outlets.

It's probably a lot more complicated than this simple picture but it really seems to me to be a manipulation.  I certainly don't know of any way to fight this other than to resolve not to participate.

archae86 wrote:
... if the soon-to-be announced AMD Navi cards turn out to do something wonderful at Einstein, I'll not suffer much buyer's remorse in replacing them.

Navi has been touted as something spectacular for quite a long time.  I'd be surprised if its performance, availability and price isn't manipulated in a similar way as well.  My advice to anyone interested in Navi is to wait until reputable sites have properly tested it and also until someone has reported results on Einstein - if that's what you intend to use it for.

What I see as the real opportunity, is the potential 'clearing of the decks' that might happen with Vega (and perhaps even Radeon VII).  Vega was always too expensive for me but now that Navi seems likely to be something 'real', I might look at what Vega is doing.  Maybe Vega56 will have a sell-off of RX 570 proportions :-).  I guess it might depend on how successful it was as a mining card and how much unsold inventory still exists.  I don't know those answers (the RX 570 sell off has met my needs) but I guess the price trends will tell the story.

Cheers,
Gary.

cecht
cecht
Joined: 7 Mar 18
Posts: 1421
Credit: 2444695657
RAC: 1502366

solling2 wrote:A memory tweak

solling2 wrote:

A memory tweak tool for your AMD GPU in both Windows and Linux version:

https://github.com/Eliovp/amdmemorytweak

Okay, that's cool. For anyone wanting to tweak their RX570 memory settings to match those of my cards switched to the factory pre-set mining bios, I used amdmemorytweak to list their current settings while running tasks in Linux. The specifics are in this public file . Two cards are listed, one with Elpida GDDR5, the other with Hynix GDDR5. Don't ask me what any of it means - you are on your own. Laughing

Edit: In the Google Doc file I added the mV values to the clock speeds for the p-state that the cards were running (state 6), because clocks and voltages at the various performance states are different in the mining bios than those in the standard gaming bios. From my limited understanding, memory timings and clocks need to be matched to GPU clocks and voltages for best computing (mining) efficiency.

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

Comment viewing options

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