Performance difference from one brand of reference r9 290 to another?

[hodger]

Hi All,

Did some searching, couldn't find anything specific so thought I'd ask.

I have 2 rigs, both 3 x r9 290.  Both have identical CPU, mobo, RAM, PSU same version of Ubuntu, same version of drivers, hardware all purchased at the same time...You get the idea...

The only difference is one rig's 290's are XFX, and the others are VTX3D.

Machines are sat near to each other (so environment is essentially the same), and both mining the same pool(s)....Yet I'm seeing roughly a 10% drop in performance on the VTX system compared to the XFX (measured this over a number of days so enough time for fluctuations to average out)....  I thought all reference cards were the same, was wondering if I'm missing something?  I even copied the cgminer.conf from one system to the other, so even that's identical.

Was hoping someone could shed a little light on this??  Thanks in advance for any thoughts...

P.S I'm aware I'm not pushing the cards quite as hard as I could, but I wanted to stay at roughly stock clock and mem till I'm really comfortable with the setup(s).

"intensity" : "20,20,20",
"vectors" : "1,1,1",
"worksize" : "256,256,256",
"kernel" : "scrypt,scrypt,scrypt",
"lookup-gap" : "2,2,2",
"thread-concurrency" : "32765,32765,32765",
"shaders" : "0,0,0",
"gpu-engine" : "938-938,938-938,938-938",
"gpu-fan" : "0-65,0-65,0-65",
"gpu-memclock" : "1250,1250,1250",
"gpu-memdiff" : "0,0,0",
"gpu-powertune" : "0,0,0",
"gpu-vddc" : "0.000,0.000,0.000",
"temp-cutoff" : "95,95,95",
"temp-overheat" : "85,85,85",
"temp-target" : "75,75,75",
"api-mcast-port" : "4028",
"api-port" : "4028",
"expiry" : "120",
"gpu-dyninterval" : "7",
"gpu-platform" : "0",
"gpu-threads" : "1",
"hotplug" : "5",
"log" : "5",
"no-pool-disable" : true,
"queue" : "1",
"scan-time" : "30",
"scrypt" : true,
"temp-hysteresis" : "3",
"shares" : "0",
"kernel-path" : "/usr/local/bin"


XFX SYSTEM
--------------------------------------------------------------------------------
 (5s):2.428M (avg):2.408Mh/s | A:3868160  R:125952  HW:0  WU:2220.2/m
 ST: 2  SS: 0  NB: 2958  LW: 29143  GF: 0  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  83.0C 3542RPM | 802.4K/801.0Kh/s | A:1273856 R:40448 HW:0 WU: 731.9/m I:20
 GPU 1:  84.0C 3626RPM | 804.6K/799.8Kh/s | A:1269248 R:38912 HW:0 WU: 725.9/m I:20
 GPU 2:  79.0C 3581RPM | 809.6K/808.0Kh/s | A:1325056 R:46592 HW:0 WU: 762.5/m I:20
--------------------------------------------------------------------------------


VTX3D SYSTEM
--------------------------------------------------------------------------------
 (5s):2.240M (avg):2.234Mh/s | A:3526656  R:112640  HW:0  WU:2021.1/m
 ST: 2  SS: 7  NB: 2975  LW: 29087  GF: 1  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  81.0C 3631RPM | 743.1K/744.4Kh/s | A:1160192 R:41472 HW:0 WU: 668.7/m I:20
 GPU 1:  82.0C 3489RPM | 743.2K/745.0Kh/s | A:1158144 R:37888 HW:0 WU: 665.8/m I:20
 GPU 2:  71.0C 3396RPM | 747.3K/745.0Kh/s | A:1208320 R:33280 HW:0 WU: 686.6/m I:20
--------------------------------------------------------------------------------

[kenshirothefist]

The first thing that you have to do is to force gpus at a specific Powertune level. You won't hurt the cards, you will just draw some more power. Some cards are more efficient the the others at powertune level 0, but when you rise powertune to 10 or 20 the get equal. So, first thing to do is:

"gpu-powertune" : "10,10,10",

see what happens, then try also

"gpu-powertune" : "20,20,20",

And don't be afraid to do some overclocking ... E=1000 and M=1500 should run just fine at 880 kh/s.

[hodger]

The first thing that you have to do is to force gpus at a specific Powertune level. You won't hurt the cards, you will just draw some more power. Some cards are more efficient the the others at powertune level 0, but when you rise powertune to 10 or 20 the get equal. So, first thing to do is:

"gpu-powertune" : "10,10,10",

see what happens, then try also

"gpu-powertune" : "20,20,20",

And don't be afraid to do some overclocking ... E=1000 and M=1500 should run just fine at 880 kh/s.

Noted, I'll have a play with the powertune some more tonight.  I did dabble with it a little early days when setting these rigs up, but decided not to set powertune till I had a reliable baseline.

Regarding E and M, again I did play a little, and I found I could get the XFX cards up around 875-900khash each, but while I wasn't getting any noticeable HW errors etc. I did see the WU value either not really increasing, or perhaps even decreasing a touch.  That, combined with a couple of forum threads suggesting the R9 290 struggles to push WU up despite khash increase when you increase these, caused me to leave near stock initially.

Thanks.

[crazyates]

I got a bunch of PowerColor R9 290 cards, and noticed they run best at a engine/mem ratio of ~0.74. You could probably get higher than 800 if you kept that ratio, but overclocked a little bit.

All my settings run at Powertune @ +20
1105/1500 gets me 855
1075/1450 gets me 835
1050/1400 gets me 810
1025/1350 gets me 795

I would love to run them at 850+ each, but the temps are too high for my liking right now, so I have to downclock. We're going to move them in a couple of weeks and install some more AC units. Hopefully between that and 2 weeks of burn in will allow us to overclock even further.

[SlimePuppy]

I've only one R9 290, a HIS with stock BIOS and Hynix memory, that's 24/7 stable crunching ~880 Kh/s and ~860 WU/m in a Win 7/64 system with 8GB RAM and latest 13.12 driver.  (It doesn't seem to mind the pair of 7870's it's living with!) Temp stays at 79°C/4285 RPM.

Overclocking to 1000/1500 was very unstable and didn't increase performance; neither did increasing TC.

Try:
TC 24550 w256 g1 engine 1015 mem 1250 power +20 I19

Good hunting!

[hodger]

I got a bunch of PowerColor R9 290 cards, and noticed they run best at a engine/mem ratio of ~0.74. You could probably get higher than 800 if you kept that ratio, but overclocked a little bit.

All my settings run at Powertune @ +20
1105/1500 gets me 855
1075/1450 gets me 835
1050/1400 gets me 810
1025/1350 gets me 795

I would love to run them at 850+ each, but the temps are too high for my liking right now, so I have to downclock. We're going to move them in a couple of weeks and install some more AC units. Hopefully between that and 2 weeks of burn in will allow us to overclock even further.

Quick question....Is this at stock voltage?

[hodger]

I got a bunch of PowerColor R9 290 cards, and noticed they run best at a engine/mem ratio of ~0.74. You could probably get higher than 800 if you kept that ratio, but overclocked a little bit.

All my settings run at Powertune @ +20
1105/1500 gets me 855
1075/1450 gets me 835
1050/1400 gets me 810
1025/1350 gets me 795

I would love to run them at 850+ each, but the temps are too high for my liking right now, so I have to downclock. We're going to move them in a couple of weeks and install some more AC units. Hopefully between that and 2 weeks of burn in will allow us to overclock even further.

Quick question....Is this at stock voltage?

OK, thought I'd updated posters so far:


WALL 950w powertune0 938/1250 (START POINT)
--------------------------------------------------------------------------------
 (5s):2.287M (avg):2.234Mh/s | A:4506112  R:156672  HW:0  WU:2018.6/m
 ST: 2  SS: 16  NB: 3838  LW: 37536  GF: 3  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  80.0C 3622RPM | 746.6K/744.4Kh/s | A:1483776 R:56832 HW:0 WU: 668.5/m I:20
 GPU 1:  81.0C 3481RPM | 743.2K/744.8Kh/s | A:1495552 R:52736 HW:0 WU: 671.7/m I:20
 GPU 2:  70.0C 3390RPM | 745.4K/744.9Kh/s | A:1526784 R:47104 HW:0 WU: 678.4/m I:20
--------------------------------------------------------------------------------


935w powertune20 938/1250 (POWER USAGE DROPPED SLIGHTLY AND WU DROPPED A FAIR BIT???)
--------------------------------------------------------------------------------
 (5s):2.194M (avg):2.228Mh/s | A:16384  R:0  HW:0  WU:1860.7/m
 ST: 2  SS: 0  NB: 13  LW: 126  GF: 0  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  79.0C 3620RPM | 743.5K/746.2Kh/s | A:7168 R:0 HW:0 WU: 609.0/m I:20
 GPU 1:  80.0C 3481RPM | 744.1K/744.2Kh/s | A:5632 R:0 HW:0 WU: 612.3/m I:20
 GPU 2:  70.0C 3391RPM | 745.7K/746.2Kh/s | A:3584 R:0 HW:0 WU: 647.1/m I:20
--------------------------------------------------------------------------------


1040w powertune10 1025/1350 (BIT MORE WU BUT 100W MORE AT THE WALL)
--------------------------------------------------------------------------------
 (5s):2.414M (avg):2.349Mh/s | A:9216  R:0  HW:0  WU:1979.5/m
 ST: 2  SS: 0  NB: 9  LW: 103  GF: 0  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  82.0C 3612RPM | 788.2K/784.0Kh/s | A:1536 R:0 HW:0 WU: 619.6/m I:20
 GPU 1:  87.0C 3473RPM | 804.5K/801.0Kh/s | A:5632 R:0 HW:0 WU: 752.7/m I:20
 GPU 2:  75.0C 3379RPM | 792.0K/789.6Kh/s | A:2560 R:0 HW:0 WU: 663.1/m I:20
--------------------------------------------------------------------------------


1040w powertune20 1025/1350 (SAME AT WALL AS TUNE10 BUT WU UP ABOUT 10% TOO FROM WHERE I STARTED)
--------------------------------------------------------------------------------
 (5s):2.346M (avg):2.351Mh/s | A:16896  R:1024  HW:0  WU:2288.4/m
 ST: 2  SS: 0  NB: 16  LW: 144  GF: 0  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  83.0C 3616RPM | 803.0K/781.6Kh/s | A:6144 R:512 HW:0 WU: 728.1/m I:20
 GPU 1:  89.0C 3476RPM | 804.5K/800.1Kh/s | A:5632 R:  0 HW:0 WU: 830.4/m I:20
 GPU 2:  77.0C 3381RPM | 785.5K/788.5Kh/s | A:5120 R:512 HW:0 WU: 760.3/m I:20
--------------------------------------------------------------------------------


1060w powertune20 1050/1400 (MORE WALL POWER, LESS WU)
--------------------------------------------------------------------------------
 (5s):2.381M (avg):2.364Mh/s | A:6656  R:1536  HW:0  WU:2125.6/m
 ST: 2  SS: 0  NB: 10  LW: 89  GF: 1  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  78.0C 4148RPM | 768.8K/789.5Kh/s | A:4096 R:   0 HW:0 WU: 670.5/m I:20
 GPU 1:  80.0C 4080RPM | 811.4K/811.6Kh/s | A:1536 R:1024 HW:0 WU: 749.7/m I:20
 GPU 2:  69.0C 3957RPM | 783.2K/793.2Kh/s | A:1024 R: 512 HW:0 WU: 725.3/m I:20
--------------------------------------------------------------------------------



So for now I've settled on powertune20 1025/1350

At this point, I have two questions...

- Based on my cgminer.conf, is it normal that my wall usage and WU dropped when I turned on powertune without increasing clocks?

- Am I right in thinking that WU is ultimately more important than hashrate?  So if hashrate goes up a bit, but WU drops, that's worse right?  Just checking the things I've read previously are correct?

thanks again all.....

[crazyates]

I got a bunch of PowerColor R9 290 cards, and noticed they run best at a engine/mem ratio of ~0.74. You could probably get higher than 800 if you kept that ratio, but overclocked a little bit.

All my settings run at Powertune @ +20
1105/1500 gets me 855
1075/1450 gets me 835
1050/1400 gets me 810
1025/1350 gets me 795

I would love to run them at 850+ each, but the temps are too high for my liking right now, so I have to downclock. We're going to move them in a couple of weeks and install some more AC units. Hopefully between that and 2 weeks of burn in will allow us to overclock even further.

Quick question....Is this at stock voltage?

Yes, stock voltage. I'm in linux, so I can't adjust that at all.

[hodger]

I got a bunch of PowerColor R9 290 cards, and noticed they run best at a engine/mem ratio of ~0.74. You could probably get higher than 800 if you kept that ratio, but overclocked a little bit.

All my settings run at Powertune @ +20
1105/1500 gets me 855
1075/1450 gets me 835
1050/1400 gets me 810
1025/1350 gets me 795

I would love to run them at 850+ each, but the temps are too high for my liking right now, so I have to downclock. We're going to move them in a couple of weeks and install some more AC units. Hopefully between that and 2 weeks of burn in will allow us to overclock even further.

Quick question....Is this at stock voltage?

Yes, stock voltage. I'm in linux, so I can't adjust that at all.

Thanks for confirming, wanted to check as I'm on Ubuntu too, and while it's theoretically possibly to change firmware and change voltage, it's something I'd rather avoid.

One thing that did occur to me once I'd found a happy setting last night, is whether the extra 30watts per card is likely to have a negative effect on the mother board.  This rig is cased up for numerous reasons, so I'm not using powered risers...It's a decent 1300w EVGA PSU so I have no concern with the PSU itself as at most the system is pulling around 1050w at the wall.  I gather 6 pin power is rated at 75w and 8 pin at 150w, but wasn't sure whether "rated" meant "maximum" or whether a decent PSU would supply more, or whether anything over 225w will definitely be coming from the motherboard?

So far the system seems happy, but it was just something I curious about:

--------------------------------------------------------------------------------
 (5s):2.421M (avg):2.370Mh/s | A:737280  R:19968  HW:0  WU:2166.7/m
 ST: 2  SS: 6  NB: 498  LW: 5323  GF: 1  RF: 0
--------------------------------------------------------------------------------
 GPU 0:  87.0C 3928RPM | 792.9K/785.4Kh/s | A:233984 R:6656 HW:0 WU: 689.0/m I:20
 GPU 1:  89.0C 3803RPM | 804.2K/797.9Kh/s | A:264192 R:4608 HW:0 WU: 759.9/m I:20
 GPU 2:  78.0C 3703RPM | 776.8K/787.6Kh/s | A:239104 R:8704 HW:0 WU: 717.8/m I:20
--------------------------------------------------------------------------------

[SlimePuppy]

<snips>
At this point, I have two questions...

- Am I right in thinking that WU is ultimately more important than hashrate?  So if hashrate goes up a bit, but WU drops, that's worse right?  Just checking the things I've read previously are correct?

thanks again all.....

Ultimately tuning for best WU is the way to go as we don't get paid for the hash rate.  I'm finding that it's much easier (takes much less time) to adjust clocks and other settings when tuning for max has rate as the response is fast.  Tuning for max WU rate is slower as it's a long-term average.

Spend some time in CGMiner changing clocks 1 MHz up and down at a time.  I noticed trends up or down in some ranges, with spikes up or down 1 MHz farther.  It'll be worth your time to chart 20 MHz either side of your chosen frequencies (one at a time).  What you find will be 'normal' for your system - trust your findings.

I don't have enough experience yet with the R9 290, but spent some 'quality time' tuning four 7870s and those puppies just ain't normal.   Whoever wrote the BIOS for those was smoking something they really should have shared with the rest of us...

[hodger]

Sorry I've not replied in a while.  Been putting all your suggestions to good use, and have managed to improve GPU performance a fair bit.

However, having installed another rig, I've hit a new problem...Available power.  In that I'm now tripping the breaker out on my electricity.

I'm going to get this sorted by having another mains feed installed, but in the mean time, I'd like to change the config of my rigs to drop the power usage, while still giving best hashrate possible.  In otherwords the most efficient power to hash ratio.

Any thoughts on the best way to achieve this?  Wans't sure whether to just drop intensity, or maybe GPU or memclock settings.  Bear in mind I'm using Ubuntu, and don't fancy tackling undervolting yet.

Each rig is currently pulling 980w at the wall, I'd like to get it down to around 700 short term if possible.

Thanks in advance for any suggestions.