Unofficial "High MHash"/"Best Settings" GPU Comparision Thread

[BinaryMage]

Looks like the 5830 info only got half-updated in the original post.

The only reason I suggest some overclock, is that this affects ideal memory speed; ideal memory clock will follow core GPU clock in a non-linear way. ATI cards can typically go from stock 800MHz to 950+MHz given cooling and voltage. The fans are programmed to be quiet to mask their huge power draw, and this let the chips go over 90C. You might look at https://en.bitcoin.it/wiki/Mining_hardware_comparison, ignore the stock speed posts, and take the smallest or second-smallest overclock that a card has rounded to a nice number.

Fixed, thanks for pointing that out.

Instead of limiting overclock, what if I limit temperature and fan speed? After all, overclocking itself doesn't kill your card; overvolting and/or running at high temperatures does. (And running at high fan will kill the fan quickly) "Stock speed" varies card-to-card, and looking at the wiki page, the average overclock is extremely variant. How does 70C and 60% fan speed (in addition to the already-implemented stock voltage) sound for limitations?

[1715033282]

1715033282

[1715033282]

1715033282

[1715033282]

1715033282

[deepceleron]

Instead of limiting overclock, what if I limit temperature and fan speed? After all, overclocking itself doesn't kill your card; overvolting and/or running at high temperatures does. (And running at high fan will kill the fan quickly) "Stock speed" varies card-to-card, and looking at the wiki page, the average overclock is extremely variant. How does 70C and 60% fan speed (in addition to the already-implemented stock voltage) sound for limitations?

That will only show variations in the product itself and make for completely non-replicable results. Two cards out of the box with serial number right next to each other might run 5C different due to manufacturing and assembly differences. One user might have an open frame rig, one might put it in his Dell with only one system fan. Those two cards, though, will get identical mhash results if run with the same software settings at the same clocks.

[BinaryMage]

Instead of limiting overclock, what if I limit temperature and fan speed? After all, overclocking itself doesn't kill your card; overvolting and/or running at high temperatures does. (And running at high fan will kill the fan quickly) "Stock speed" varies card-to-card, and looking at the wiki page, the average overclock is extremely variant. How does 70C and 60% fan speed (in addition to the already-implemented stock voltage) sound for limitations?

That will only show variations in the product itself and make for completely non-replicable results. Two cards out of the box with serial number right next to each other might run 5C different due to manufacturing and assembly differences. One user might have an open frame rig, one might put it in his Dell with only one system fan. Those two cards, though, will get identical mhash results if run with the same software settings at the same clocks.

Fair point, strike that. I suppose I'll look over the wiki and see what I come up with.

[Vbs]

Card Manufacturer: Sapphire
Card Model Number: 5850 (Refurbished)
Card OEM: ATI Reference
MHash per Second: 400
Core Clock: 970MHz
Memory Clock: 150MHz
Voltage: stock
Average Operating Temperature: 70ºC
Ambient Temperature: 18~20ºC
Fan Speed: 56-59%
Host OS: Win7 x64
Driver Version: 12.1
SDK Runtime Version: 2.5.793.1 (from driver 11.11)
Mining Program: cgminer 2.3.1
Command Line Flags/GUIminer Settings: phatk kernel, vectors 2, worksize 256, intensity 9

Card Manufacturer: Sapphire
Card Model Number: 5850 (Refurbished)
Card OEM: ATI Reference
MHash per Second: 404
Core Clock: 980MHz
Memory Clock: 150MHz
Voltage: stock
Average Operating Temperature: 68ºC
Ambient Temperature: 18~20ºC
Fan Speed: 55%
Host OS: Win7 x64
Driver Version: 12.1
SDK Runtime Version: 2.5.793.1 (from driver 11.11)
Mining Program: cgminer 2.3.1
Command Line Flags/GUIminer Settings: phatk kernel, vectors 2, worksize 256, intensity 9

[check_status]

Thanks for making this thread BinaryMage. This is exactly what I was looking for, https://bitcointalk.org/index.php?topic=62071.msg743905#msg743905 , results that could be trusted to teach n00b5 like me some OC 5k1llz.
I don't trust that wiki, if there are too few results it's impossible for new people, like when I first started reading, to determine credible results. Instead of teaching skills they could be teaching kills, which is not so good. At least someone with a little more knowledge has taken on the task for safe guiding ideas.

[bulanula]

Thanks for making this thread BinaryMage. This is exactly what I was looking for, https://bitcointalk.org/index.php?topic=62071.msg743905#msg743905 , results that could be trusted to teach n00b5 like me some OC 5k1llz.
I don't trust that wiki, if there are too few results it's impossible for new people, like when I first started reading, to determine credible results. Instead of teaching skills they could be teaching kills, which is not so good. At least someone with a little more knowledge has taken on the task for safe guiding ideas.

Not relevant.

Each GPU performs differently.

One may do 1000 without voltage increase and another ( of same model ) might only do 800 on stock volts ...

YMMV

[deepceleron]

Thanks for making this thread BinaryMage. This is exactly what I was looking for, https://bitcointalk.org/index.php?topic=62071.msg743905#msg743905 , results that could be trusted to teach n00b5 like me some OC 5k1llz.
I don't trust that wiki, if there are too few results it's impossible for new people, like when I first started reading, to determine credible results. Instead of teaching skills they could be teaching kills, which is not so good. At least someone with a little more knowledge has taken on the task for safe guiding ideas.

Not relevant.

Each GPU performs differently.

One may do 1000 without voltage increase and another ( of same model ) might only do 800 on stock volts ...

YMMV

Not relevant.

This thread is for finding the fastest software miner settings.

One may overclock more than another, but the competition is to find the fastest hashing miner at a particular overclock.

[check_status]

This thread is for finding the fastest software miner settings.

One may overclock more than another, but the competition is to find the fastest hashing miner at a particular overclock.

Absolutely. That's why I chose the term "skill". It will require effort, gong, over time, fu, to see what you can achieve.

[BinaryMage]

This thread is for finding the fastest software miner settings.

One may overclock more than another, but the competition is to find the fastest hashing miner at a particular overclock.

Absolutely. That's why I chose the term "skill". It will require effort, gong, over time, fu, to see what you can achieve.

Sorry I haven't updated this thread in a while, I was taking a few days break to clear my head. I'll update it momentarily.

Sounds like two different threads are needed. One for software settings, and one for a general overclocking guide. I'm no overclocking expert, but I probably could write a guide like that if no one else steps up. Thoughts?

[BinaryMage]

Alright. Main post has been updated, with new regulations limiting core clock to 100 MHz or 20%, whichever is lower, above reference stock. That seemed a fair way to limit clocks that could be applied to any card, wouldn't allow insane overclocks, and requires no judgement on my part. As always, suggestions, corrections, etc. are always welcome.

[deepceleron]

Alright. Main post has been updated, with new regulations limiting core clock to 100 MHz or 20%, whichever is lower, above reference stock. That seemed a fair way to limit clocks that could be applied to any card, wouldn't allow insane overclocks, and requires no judgement on my part. As always, suggestions, corrections, etc. are always welcome.

The smaller of +100Mhz or +20% is +100Mhz in all cases, you would need core clock under 500MHz for the 20% rule to take effect. "Running stable for 24 hrs" etc doesn't matter because these are easy overclocks, you just want to run the miner until it has results at least xxx.xMhash/s stable accuracy.

Note that running a lower GPU core overclock for benchmarking will also affect the "sweet spot" for memory underclock on 2.1-2.5 SDKs. My own research shows that changing a 5770 from 1000->850MHz will change the best memory speed from 295MHz to 270Mhz. This is important to note both for this competition, and for trying these results on your own miner when you want to go back to maximum overclock.

Here are the official AMD stock core clocks of common mining GPUs, and their "competition" overclock of +100Mhz (which is significantly lower than typically attainable, but should be achievable by all [except perhaps 7970 @ 1025MHz]):

Model	Shaders	Stock MHz	Overclock
5770	800	850	950
5830	1120	800	900
5850	1440	725	825
5870	1600	850	950
5970	3200	725	825
Model	Shaders	Stock	Overclock
6850	960	775	875
6870	1120	900	1000
6950	1408	800	900
6970	1536	880	980
6990	3072	830	930
Model	Shaders	Stock	Overclock
7950	1792	800	900
7970	2048	925	1025

(Note the table above highlights the numbering sham that is the 6xxx series; a 5830 gets about the same hashrate as a 6870 when pushed to the similar max overclock, as they have the same number of vertex shader cores.)

[superfastkyle]

Just wanted to say the 5970 at 775mhz core 300 mhz memory and 1.02v works well, thanks for the suggestion, sapphire 5850 runs at these clocks fine as well. This definitely looks like this may be my summer clocks. Am pulling 911 watts right now with two 5970 and two 5850. 2102mhash using bamt

[Isokivi]

Card Manufacturer: AMD
Card Model Number: HD 5870
Card OEM: ATI
MHash per Second: 458MHS
Core Clock: 1010MHZ
Memory Clock: 294MHZ
Average Operating Temperature: 62C
Ambient Temperature: 22C
Fan Speed: 50%
Host OS: Bamt 0.5
Driver Version: 11.6 and SDK 2.4
Mining Program: Phoenix
Command Line Flags/GUIminer Settings:  BFI_INT VECTORS, FASTLOOP=false AGGRESSION=10
Watts Drawn While Mining: Unknown
Other Information: kernel  phatk2, core voltage 1.7
Record stable uptime at current settings 8d (and counting).

[deepceleron]

Card Manufacturer: AMD
Card Model Number: HD 5870
Card OEM: ATI
MHash per Second: 458MHS
Core Clock: 1010MHZ
Memory Clock: 294MHZ
Average Operating Temperature: 62C
Ambient Temperature: 22C
Fan Speed: 50%
Host OS: Bamt 0.5
Driver Version: 11.6 and SDK 2.4
Mining Program: Phoenix
Command Line Flags/GUIminer Settings:  BFI_INT VECTORS, FASTLOOP=false AGGRESSION=10
Watts Drawn While Mining: Unknown
Other Information: kernel  phatk2, core voltage 1.7
Record stable uptime at current settings 8d (and counting).

That can't be your GPU voltage, or it would be a smoky pile of smoke. Plus, see edited original post and above for new GPU MHz to benchmark at, so we can measure who's the best at tweaking their mining software.

[check_status]

Maybe he was just following BinaryMages example and hadn't read the "Edit" paragraph.

[ssateneth]

Card Manufacturer: ATI
Card Model Number: 5870
Card OEM: Asus
MHash per Second: 504.3-504.5 MH/s
Core Clock: 1090 Mhz
Memory Clock: 205 Mhz
Voltage: 1.200v
Average Operating Temperature: Varies. Usually Ambient + 38c
Ambient Temperature: Varies. Sits outside. Currently 8c as I type this.
Fan Speed: Dedicated 60mm delta FFB0612EHE @ 100%
Host OS: Win7 x64
Driver Version: 11.12 CAT
SDK Runtime Version: 2.1
Mining Program: Phoenix 2.0 Beta RC1
Command Line Flags/GUIminer Settings: phatk2, AGGRESSION=14, VECTORS=true, BFI_INT=true, WORKSIZE=128
Other Information: Stable for 1 month so far. Will reduce speed/volts as spring temperatures increase.

[Isokivi]

Card Manufacturer: AMD
Card Model Number: HD 5870
Card OEM: ATI
MHash per Second: 458MHS
Core Clock: 1010MHZ
Memory Clock: 294MHZ
Average Operating Temperature: 62C
Ambient Temperature: 22C
Fan Speed: 50%
Host OS: Bamt 0.5
Driver Version: 11.6 and SDK 2.4
Mining Program: Phoenix
Command Line Flags/GUIminer Settings:  BFI_INT VECTORS, FASTLOOP=false AGGRESSION=10
Watts Drawn While Mining: Unknown
Other Information: kernel  phatk2, core voltage 1.7
Record stable uptime at current settings 8d (and counting).

That can't be your GPU voltage, or it would be a smoky pile of smoke. Plus, see edited original post and above for new GPU MHz to benchmark at, so we can measure who's the best at tweaking their mining software.

You are correct, theres a typo in there the correct core voltage is 1.17
I also should propably mention that the stock fans have been replaced with Arctic coolings Accelero Extreme.

[BinaryMage]

Alright. Main post has been updated, with new regulations limiting core clock to 100 MHz or 20%, whichever is lower, above reference stock. That seemed a fair way to limit clocks that could be applied to any card, wouldn't allow insane overclocks, and requires no judgement on my part. As always, suggestions, corrections, etc. are always welcome.

The smaller of +100Mhz or +20% is +100Mhz in all cases, you would need core clock under 500MHz for the 20% rule to take effect. "Running stable for 24 hrs" etc doesn't matter because these are easy overclocks, you just want to run the miner until it has results at least xxx.xMhash/s stable accuracy.

100 MHz was the intent, yes. I simply included the 20% to ward off the inevitable obscure low-clocked (perhaps laptop) GPU for which 100 MHz is not an easily obtainable overclock. Running 24hrs stable is in there for the same reason, even though I agree it is somewhat redundant.

Isokivi, that is an excellent overclock, but it does not meet the guidelines. (Core clock too high) I do apologize if my example misled you; it was created before the edit and has now been updated.

[cmg5461]

Just wanted to say the 5970 at 775mhz core 300 mhz memory and 1.02v works well, thanks for the suggestion, sapphire 5850 runs at these clocks fine as well. This definitely looks like this may be my summer clocks. Am pulling 911 watts right now with two 5970 and two 5850. 2102mhash using bamt

Glad I could help

[twoBitBasher]

As I read these 500+ Mhash with 1090Mhz numbers I gotta say "man I was stupid to buy that xfx card back in the day. Even more so when I knew I'd try to oc the hell out of it eventually". I got 955mhz core, 300mhz ram with bamt 0.5 @ 438Mhash/s. btw the power draw delta from idle is 143w(wall). My PSU is under this load around 90-91% efficient.