All things Navi 10

Gavin
Gavin
Joined: 21 Sep 10
Posts: 191
Credit: 35,037,153,310
RAC: 789,134

Looking very good Archae86

Looking very good Archae86 and I'm glad you've dipped your toe into the puddle of Navi owners here!

cecht wrote:

Has anybody had success running gravity wave work on the  5700-series?

Prior to the January 2020 driver release I had been running GW work since last October. During that time every task validated successfully with the exception of one and I suspect that was caused by my meddling. I ran tasks at x1, x2, x3 and finally x4 without issue :-)

Gav.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,071
Credit: 6,022,763,617
RAC: 2,441,706

cecht wrote:Let us know how

cecht wrote:
Let us know how 3X works.  And 4X!

I think I'll skip 4X.

Here are approximate Elapsed time and nominal daily GRP credit rate for the first three:

1X  6:12 804,774
2X 10:46 926,861
3X 15:24 972,000

As to box level power consumption from the wall socket, I observed 71.8 watts idle (vs. 69.0 for the same box with an RX 570)

1X 208 watts
2X 226.4 watts
3X 232.2 watts

The 1X numbers are suspect--they come from a single task, so should be considered "for indication only" as the mandatory signs on the non-traceable voltmeters in the lab used to say.  But those voltmeters were generally quite close to the truth.

As I truly hope to run this box with two cards (initially this 5700 and my "good" 570), and it has only four physical cores with no hyperthreading, the practical interest for me of 3X is not so high, so I intend to put it back after a day, which should be enough to get an initial stability and validity indication.

Speaking of validity, with 19 hours on the box, so over 200 tasks returned, and working on tasks with about 5 days of queue "seasoning", I've yet to post an invalid.  This just means it is not awful, as  I don't know how many of the 12 inconclusives in progress are from the 5700, and how many from the 570 card before the change.  My actual guess is that the invalid rate at the 2X operating point is similar to my 570 results and that it is too soon to say about the 3X work.  Lest we forget, I had a serious invalidity rate increase when I went from 2X to 3X on my Radeon VII, months ago when it was otherwise working well.

Next, I plan to try turning on MSIAfterburner, as I have liked the fan noise variability for my VII and my 570s better under Afterburner fan control than under AMD's Radeon software direct control.  I also have trouble finding my way around the Radeon 2020 edition control pages, which seem rather differently configured than last year's version.

Bottom line: of Gavin's, Mumak's, and my 5700/5700 XT results hold up, these cards are worth serious consideration for Einstein GRP work where the user is hungry for more Einstein production per PCI-e slot than the much-loved RX 570 cards produce.  I suspect of the two, that this 5700 may be more price-effective and power-efficient than the XT cards.

Gavin
Gavin
Joined: 21 Sep 10
Posts: 191
Credit: 35,037,153,310
RAC: 789,134

archae86 wrote: Bottom line:

archae86 wrote:

Bottom line: of Gavin's, Mumak's, and my 5700/5700 XT results hold up, these cards are worth serious consideration for Einstein GRP work where the user is hungry for more Einstein production per PCI-e slot than the much-loved RX 570 cards produce.  I suspect of the two, that this 5700 may be more price-effective and power-efficient than the XT cards.

I agree with your judgment, these cards could be a serious contender for the best 'bang for buck' GPU here at Einstein.

As always, time will tell and hopefully future driver improvements will further improve this series of cards worth. For the moment I am overjoyed with the performance v's power consumption :-)

archae86
archae86
Joined: 6 Dec 05
Posts: 3,071
Credit: 6,022,763,617
RAC: 2,441,706

As I hoped, launching

As I hoped, launching MSIAfterburner altered my 5700 fan speeds in a favorable way.  I did not change the fan curve I had last used with my RX 570 cards, though it was not fine-tuned to them.  So the initial results were somewhat accidental, while, to me, quite favorable.

That happened to mean appreciably higher initial fan speed, which quickly brought the reported GPU temperature down from about 75C to about 67C.  Once there, the audible fan noise has been far less "lumpy", and it only takes GPU fan speed of 41% average (1464 rpm) to keep it there.  It was also a relief to see the reported GPU memory temperature come down from the low 90s to the low 80s.  Overall this makes me more optimistic about the prospects for successfully adding another 570 (at first), or (5700) to the box with this one.  

Somewhere in there, the GPU voltage dropped to .943, and the average wall socket power for the box dropped to 221 watts.  As Afterburner was just set to a power limit of 0%, this was a surprise.

I suspect there is a bit more power efficiency available if I push down the power limit parameter far enough (the XFX 570s responded little until one hit about -30%, but by -35% there was enough power reduction and little enough performance impact to get a nice little box level efficiency adder).

But for the near term, I need to gather some stability evidence by just letting it run, probably overnight still at 3X, then a while at 2X.  If all seems well, I may put the "good" 570 card in the second PCI-e slot Tuesday.

archae86
archae86
Joined: 6 Dec 05
Posts: 3,071
Credit: 6,022,763,617
RAC: 2,441,706

My first invalid running on

My first invalid running on the 5700 has turned up, and it was on a 3X job.  Not only that, but a plod through the "All" list sorted by run time showed three more 3X tasks currently displayed as inconclusive on their first quorum partner comparison.

There are many dozens of validations, so the rate is far short of catastrophic, but I suspect my card under current conditions may have a slightly elevated invalid rate at 3X.   I've reverted to 2X.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 458
Credit: 233,122,964
RAC: 917

My RX 570 (Win7 64-bit) does

My RX 570 (Win7 64-bit) does the Gamma-ray in 700 seconds, at about 90 watts the best I can figure out.  As I see the numbers, that still compares reasonably well with the RX 5700.  I have just ordered a second one to try it out on Ubuntu.

EDIT: The power could be even a bit more.  I will check it out in a day or two, as the weather gets cooler here.  I have to take the GPU-Z power, and add at least 15 watts to it, maybe 20 to get the full board power.

 

cecht
cecht
Joined: 7 Mar 18
Posts: 1,084
Credit: 1,608,973,036
RAC: 2,416,228

Archae86 wrote:The card has a

Archae86 wrote:
The card has a tiny switch resembling the famous "mining mode BIOS" switch on the XFX RX 570 cards I've used.  A little Internet searching suggested that also on this 5700 card XFX might have configured the forward (toward the I/O panel) switch position for a power-efficient mode of operation.  I took a chance and moved that switch to that position on first install.  Thus all results I post here may be non-representative of typical 5700 card behavior.

I just read a web review of your XFX RX 5700 DD that mentioned that the dual BIOS switch is also found on XFX 5700 XT models, so crunchers with those cards may want to check out possible compute enhancements by flipping the switch. Faster crunch times have been realized by flipping the BIOS switch to the mining/power-saving/quite position on RX 570 (and 580?) cards from XFX and Sapphire, so the 5700-series BIOS switch would be expected to provide similar benefit.

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

archae86
archae86
Joined: 6 Dec 05
Posts: 3,071
Credit: 6,022,763,617
RAC: 2,441,706

Today I tried adjusting the

Today I tried adjusting the Power Limit % parameter while running GRP tasks at 2X on my new 5700 card.  Bear in mind that this is an XFX model with what appears to be a "BIOS mining mode" switch and that all my work has been with that switch forward (toward the I/O panel), which is not how it was shipped.   Probably that switch position changes the power/frequency/elapsed time relationship for some of the power limitation settings.

Somewhat as I think I remember with my XFX 570 cards, adjusting the lower limit down from 0 had no detectable effect until I hit -30%.  My procedure was to move it down in increments of 5 while watching live graphs of GPU Core clock, GPU Voltage, and GPU Power consumption displayed in real-time by MSI Afterburner. 

While an effect at -30% was detectable, it was slight, but the steps down from there were quite distinct.  To my surprise, the system level power efficiency improved at each step down, terminated by the limit of the slider at -50%.  While the productivity loss at the lower steps was quite large, the power savings was even larger, giving an improved overall power efficiency metric all the way down.

For each step, my table shows a  measured average wall socket power for the system, calculated nominal daily credit productivity running Einstein GRP, and my power efficiency parameter.  The power efficiency parameter as I calculate it has units of credit/day/watt and is just the ratio of the previous two numbers.  The higher it is, the more bang you are getting per buck, if bang is deemed to be Einstein GRP credit, and "buck" considers system level power consumption only.

All of these numbers were logged within a few hours of each other on the same system, with control being exercised by MSIAfterburner, with a fan curve that gave reported GPU temperatures ranging from about 67C for the 0 limitation line to about 63C for the most limited cases.  I generally observed half a dozen task completions at each step shown in the table.

Lim Power Credit_Rate Power_Efficiency
+0% 215.9 925,428 4286
-30% 214.6 924,000 4305
-35% 210.5 911,342 4330
-40% 202.7 894,996 4416
-45% 192.2 872,816 4542
-50% 182.1 836,246 4592

I should emphasize that all power numbers in this table refer to system total power coming out of the wall socket, so comparison with purported card-level power consumption reported by GPU-Z is problematic.  The system power consumption I observed on this system after installing this card with a system idle of running any BOINC tasks, and browsers shut down, etc. but no attempt to stifle normal background tasks was 71.8 watts.  By no means is all the power above 71.8 watts consumed in the graphics card, as there is additional system activity to support the GPU giving added power consumption in the CPU, memory, motherboard logic components, motherboard power conversion electronics, and additional power dissipated in the system power supply (which is less than 100% in conversion efficiency).

 

archae86
archae86
Joined: 6 Dec 05
Posts: 3,071
Credit: 6,022,763,617
RAC: 2,441,706

I tried adding an XFX 570

I tried adding an XFX 570 card into the box with my new XFX RX 5700 DD Ultra 8G Model: RX-57XL8LB36.  As I thought this 5700 card had impressively effective cooling, my first try put the 570 below it.  

I was a bit alarmed to see that this particular card design carries so much extra (beyond two slot) width that the remaining space for air to get into the main card fans between it and the next card was pretty slender.  So I expected the card to run much warmer than before.  But it was worse than "much warmer".  The reported GPU temperature rose within a few minutes into the  90s, despite a fan curve that had the fan screaming much louder than I will tolerate on my main daily use machine.

So I tried putting this 5700 below this 570.  While I expected the 5700 to do fine in that position, I thought the 570 would be almost as challenged as the 5700 had been when on top.  Actually it did much better.  Right now with several hours running in a 71F room, the 5700 is reading 66C while running the fan at 44% (1574 rpm) while the 570 is reading 78C while running the fan at 56%, 2526.

I've got no power limitation in place at all yet, though I am running with MSIAfterburner active.  The approximate nominal credit production rate is a bit over 1.4 Million/day, with box power consumption at the wall socket running about 348 watts.

Stability is the first important question, and I don't plan to fiddle with settings until tomorrow afternoon.  But I have good hope I can get enough power reduction out of both cards to improve the thermal and fan noise situation, while still getting a pretty substantial credit rate.  I'll take more accurate measurements when I think I've got a suitable operating point.

I had really thought that a pair of 5700 cards in a capable case with a capable supply was a completely viable proposition.  But unless I missed something really basic, this particular 5700 card is not suitable for "top card" use with a motherboard that spaces the PCI-e slots of interest three apart, which I think is the norm.

Quite likely there is another 5700 card out there which is not as wide as this one.  There may also be a card with a cooling solution more capable of dealing with this sort of configuration.  But I'd suggest considering this particular 5700 only as the "bottom card" in a two card box.

cecht
cecht
Joined: 7 Mar 18
Posts: 1,084
Credit: 1,608,973,036
RAC: 2,416,228

Archae86 wrote:Stability is

Archae86 wrote:
Stability is the first important question, and I don't plan to fiddle with settings until tomorrow afternoon.  But I have good hope I can get enough power reduction out of both cards to improve the thermal and fan noise situation, while still getting a pretty substantial credit rate.  I'll take more accurate measurements when I think I've got a suitable operating point.

If you don't have a monitor connected to your crunching cards, a surprisingly effective cooling solution I've found for my dual GPU hosts (not 5700s, unfortunately) is to attach a 120 mm case fan externally over the GPU ports. On one host, the fan is drawing hot air out from the GPUs; on the other, it's blowing room temp air in over them. (The optimal airflow direction coincides with the direction of the main case fans, which I previously determined by trial and error.) The external fans are very quiet and cost ~$25 USD each. Just two zip ties did the trick to firmly attach a fan over the ports. It only took a few reboots to find suitable BIOS/UEFI fan speed settings.
By the way, thanks for the detailed reporting of getting your Navi card set up and tuned. I hope you get your dual card system humming along stably and coolly.

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.