Pages:
Author

Topic: Good Speeds on Sapphire 5830's 311.4MHash/s - page 2. (Read 24167 times)

sr. member
Activity: 282
Merit: 250
330mhash/sec with my HIS 5830 card  @ 1040/340

300mhash/sec with my 2 Sapphire 5830 cards (they are in my game rig so i'm not pushing them hard) @ 950/340

All stock voltages and stable.
member
Activity: 68
Merit: 10
High Desert Dweller-Where Space and Time Meet $
312Mh/s @ 975Mhz core??? Thats impressive.

Thank you. I'm probably a little closer to 313.5, but I don't want to have people get their panties in if they clone my setup and can't match it. Looking to push it higher, but as always, taking the rigs offline for tweaking costs more then it's sometimes worth with difficulty changes.
newbie
Activity: 51
Merit: 0

Update:

GPUs:
2x Gigabyte 5830's 1GB
Core: 975Mhz
Mem: 340Mhz
Fan: 85%
Temp: ~69.0C (top card sometimes 1C higher)

System
OS: Ubunturd 11.04 - Linux [REDACTED] 2.6.38-8-generic #42-Ubuntu SMP Mon Apr 11 03:31:24 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux
SDK: ati-stream-sdk-v2.1-lnx64
CPU: AMD Phenom(tm) II X6 1090T Processor
RAM: 4G

(tempted to consider trying out 2.4, but meh)

Latest poclbm from June 27th, 2011 (https://github.com/m0mchil/poclbm): poclbm/2011.b7

./poclbm.py -u [REDACTED] --pass=[REDACTED] -o [REDACTED] -p [REDACTED] --verbose -d 1 -a 2 -w 256 --backup=[REDACTED] --tolerance=4 --failback=2 -r 5 -f 1 -v

Stable at 312.5Mhps.
312Mh/s @ 975Mhz core??? Thats impressive.
member
Activity: 112
Merit: 10
Just to update on my above post, I did this.

"Same observation with 5830. The only difference here is worksize=256 peak @ mem clock 354 is about 5 Mh/s faster than the worksize=128 peak at mem clock 186."

and ended up with the 312 Mhashes/sec on my 5830 Win7 box.  Tongue

*EDIT* Now I made this change #define Ma(x, y, z) amd_bytealign( (z^x), (y), (x) )

to phatk kernel.cl and am now getting 316 - 319 Mhashes/sec Tongue Tongue

 
member
Activity: 68
Merit: 10
High Desert Dweller-Where Space and Time Meet $
Update:

GPUs:
2x Gigabyte 5830's 1GB
Core: 975Mhz
Mem: 340Mhz
Fan: 85%
Temp: ~69.0C (top card sometimes 1C higher)

System
OS: Ubunturd 11.04 - Linux [REDACTED] 2.6.38-8-generic #42-Ubuntu SMP Mon Apr 11 03:31:24 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux
SDK: ati-stream-sdk-v2.1-lnx64
CPU: AMD Phenom(tm) II X6 1090T Processor
RAM: 4G

(tempted to consider trying out 2.4, but meh)

Latest poclbm from June 27th, 2011 (https://github.com/m0mchil/poclbm): poclbm/2011.b7

./poclbm.py -u [REDACTED] --pass=[REDACTED] -o [REDACTED] -p [REDACTED] --verbose -d 1 -a 2 -w 256 --backup=[REDACTED] --tolerance=4 --failback=2 -r 5 -f 1 -v

Stable at 312.5Mhps.

Previously was around 304 with either kernels optimized in Phoenix. Has been up about 3ish weeks with these overclock settings.

GPU / System Stats:
Code:
Adapter 0 - ATI Radeon HD 5800 Series 
            Sensor 0: Temperature - 69.00 C

Adapter 1 - ATI Radeon HD 5800 Series 
            Sensor 0: Temperature - 68.50 C
*** Clocks ***

Adapter 0 - ATI Radeon HD 5800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    975           340
             Current Peak :    975           340
  Configurable Peak Range : [600-875]     [900-1200]
                 GPU load :    99%

Adapter 1 - ATI Radeon HD 5800 Series 
                            Core (MHz)    Memory (MHz)
           Current Clocks :    975           340
             Current Peak :    975           340
  Configurable Peak Range : [600-875]     [900-1200]
                 GPU load :    99%
*** GPU #0 Stats ***
Fan speed query:
Query Index: 0, Speed in percent
Result: Fan Speed: 85%

Temperature for thermal controller 0 is 69.000000
Current Activity is Core Clock: 975MHZ
Memory Clock: 340MHZ
VDDC: 1163
Activity: 99 percent
Performance Level: 2
Bus Speed: 5000
Bus Lanes: 16
Maximum Bus Lanes: 16

*** GPU #1 Stats ***
Fan speed query:
Query Index: 0, Speed in percent
Result: Fan Speed: 85%

Temperature for thermal controller 0 is 68.500000
Current Activity is Core Clock: 975MHZ
Memory Clock: 340MHZ
VDDC: 1163
Activity: 98 percent
Performance Level: 2
Bus Speed: 5000
Bus Lanes: 16
Maximum Bus Lanes: 16

*** CPU and Misc Sensors ***
atk0110-acpi-0
Adapter: ACPI interface
Vcore Voltage:         +1.44 V  (min =  +0.80 V, max =  +1.60 V)
CPU/NB Voltage:        +1.15 V  (min =  +0.80 V, max =  +1.60 V)
CPU VDDA Voltage:      +2.49 V  (min =  +2.00 V, max =  +3.00 V)
DRAM Voltage:          +1.55 V  (min =  +1.40 V, max =  +1.90 V)
HT Voltage:            +1.20 V  (min =  +0.80 V, max =  +1.50 V)
NB Voltage:            +1.10 V  (min =  +0.90 V, max =  +1.35 V)
SB Voltage:            +1.11 V  (min =  +0.80 V, max =  +1.50 V)
+3.3V Voltage:         +3.36 V  (min =  +2.97 V, max =  +3.63 V)
+5V Voltage:           +5.06 V  (min =  +4.50 V, max =  +5.50 V)
+12V Voltage:         +12.00 V  (min = +10.20 V, max = +13.80 V)
CPU Temperature:       +38.0°C  (high = +40.0°C, crit = +90.0°C) 
MB Temperature:        +40.0°C  (high = +35.0°C, crit = +95.0°C) 
NB Temperature:        +49.0°C  (high = +65.0°C, crit = +95.0°C) 
SB Temperature:        +44.0°C  (high = +35.0°C, crit = +75.0°C) 
OPT_TEMP1 Temperature:  +0.0°C  (high =  +0.0°C, crit = +90.0°C) 
OPT_TEMP2 Temperature:  +0.0°C  (high =  +0.0°C, crit = +90.0°C) 
OPT_TEMP3 Temperature:  +0.0°C  (high =  +0.0°C, crit = +90.0°C) 

k10temp-pci-00c3
Adapter: PCI adapter
temp1:        +0.0°C  (high = +70.0°C, crit = +90.0°C) 
hero member
Activity: 700
Merit: 500
Has anybody else had an issue w\ invalid/stale shares?

I noticed that my 3x5830 rig had a 1/7 ratio w\ my cards oc'd to 975/300 ( no voltage increase ).  I had around 7000 valid shares per card and then 1000-1200 stales :X.

Went back to stock bios and stale ratio went way down.   

Nope, i have 1 running and im only getting <0.05% stale. 1015/300.
hero member
Activity: 1148
Merit: 501
Has anybody else had an issue w\ invalid/stale shares?

I noticed that my 3x5830 rig had a 1/7 ratio w\ my cards oc'd to 975/300 ( no voltage increase ).  I had around 7000 valid shares per card and then 1000-1200 stales :X.

Went back to stock bios and stale ratio went way down.   
member
Activity: 112
Merit: 10
Windows 7 Areo off with only one Sapphire 5830 ATM.


Using latest ATI driver 11.6 Phoenix phatk AOCLBF front end and I'm getting 305 Mh/s

Trixx Settings

Core Clock 1010

Memory Clock 625 (cant seem to go lower)

GPU Voltage 1200 (Mhs were all over the map before tweaking the voltage that made things settle to 305 although I was hesitant to tweak)

Temp 68c

AOCLBF 1.72 settings

(check) Vectors

(check) HD5xxx

(check) Fastloop

(check) phatk

Agression 12

Worksize 128

GPU - Z

Shows 99 - 100% GPU Load

Fan Speed 65%

Does not show the tweaked Trixx voltage



sr. member
Activity: 458
Merit: 250
beast at work

Edit: This has been brought up before in other threads (or so I've been told). While TriXX lets you adjust the voltage of the Sapphire 5830 cards, many programs that monitor your video card may NOT actually be reading the correct voltage.

For example, pushing the voltage from ~1.16 default to ~1.20 in TriXX may still show up as ~1.16 in GPU-Z, GPU Caps Viewer, HWMonitor, etc, etc - which gives you the impression that TriXX didn't work.

There is one program I know of that reads the voltage correctly. HWiNFO  http://www.hwinfo.com/

On load, it displays the voltage of 1.156v (1163 in TriXX). When I set it to "1200" in TriXX, HWiNFO then reads 1.195v. So I know TriXX is working.


i can confirm this, thank you Xenomorph
newbie
Activity: 32
Merit: 0
Sapphire 5830, model 100297L ($110 shipped)

I'm using the latest Catalyst 2.4 drivers (which include OpenCL), and using Sapphire TriXX to control settings.

1000/300, 1.2v, 65% fan.

phoenix miner:
-k phatk vectors bfi_int aggression=12

~310 Mhash

When I push it to 1033/300, I get ~320 Mhash.

----------

Edit: This has been brought up before in other threads (or so I've been told). While TriXX lets you adjust the voltage of the Sapphire 5830 cards, many programs that monitor your video card may NOT actually be reading the correct voltage.

For example, pushing the voltage from ~1.16 default to ~1.20 in TriXX may still show up as ~1.16 in GPU-Z, GPU Caps Viewer, HWMonitor, etc, etc - which gives you the impression that TriXX didn't work.

There is one program I know of that reads the voltage correctly. HWiNFO  http://www.hwinfo.com/

On load, it displays the voltage of 1.156v (1163 in TriXX). When I set it to "1200" in TriXX, HWiNFO then reads 1.195v. So I know TriXX is working.
sr. member
Activity: 254
Merit: 250
Will one of you guys who are running at 300 mem clock and worksize=256 give 225 mem clock and worksize=128 a try, and let me know your before and after results.
I don't have any 5830s, but all 58xx cards should behave similarly. On my 5870s I noticed that below ~250 mem clock worksize=128 works better. In fact, the optimal settings I've found we're at mem=186, worksize=128. That gives a few Mhash/s more than worksize=256 and mem anywhere between 250 and 400.

The conclusion I came to with the 6870 was that the sweetspot was between 180 and 240. I was thinking the 5800s should be similar. There seemed to be a nice spike between those values, then below 180 the hash rates fell off rapidly. Did you find a similar pattern?
Pretty much spot-on. At worksize=128, between 180 and 250 the hashrate varies by a couple Mh/s, below 180 it drops fast and above 250 it starts decreasing slowly. By mem clock 300, worksize=256 is ~10 Mh/s faster and peaks at mem clock 354. But the worksize=256 peak @ mem clock 354 is still about 5 Mh/s lower than the worksize=128 peak at mem clock 186.

If this is not a dedicated rig though, then there is one issue with using mem clocks as low as ~200. Any load on the video card, such as playing video or flash, will lower the hashrate a lot more than at higher mem clocks.

Same observation with 5830. The only difference here is worksize=256 peak @ mem clock 354 is about 5 Mh/s faster than the worksize=128 peak at mem clock 186.
legendary
Activity: 826
Merit: 1004
Pretty much spot-on. At worksize=128, between 180 and 250 the hashrate varies by a couple Mh/s, below 180 it drops fast and above 250 it starts decreasing slowly. By mem clock 300, worksize=256 is ~10 Mh/s faster and peaks at mem clock 354. But the worksize=256 peak @ mem clock 354 is still about 5 Mh/s lower than the worksize=128 peak at mem clock 186.

I wonder why the hashrate is spiking at those values? It seems extremely odd to me that there should be performance spikes at such low clocks. I'd expect them to be at the high end.

If this is not a dedicated rig though, then there is one issue with using mem clocks as low as ~200. Any load on the video card, such as playing video or flash, will lower the hashrate a lot more than at higher mem clocks.

Yeah, I noticed that. I got around the flash problem by disabling hardware acceleration for flash. I also noticed that avi files had no effect on the hash rate but mkv's dropped it to about 50 Mh/s. I'm only downloading xvids from now on, they seem to play ok with AGRESSION set to 6 and I'm trying to avoid or disable anything that uses hardware acceleration.

newbie
Activity: 31
Merit: 0
Will one of you guys who are running at 300 mem clock and worksize=256 give 225 mem clock and worksize=128 a try, and let me know your before and after results.
I don't have any 5830s, but all 58xx cards should behave similarly. On my 5870s I noticed that below ~250 mem clock worksize=128 works better. In fact, the optimal settings I've found we're at mem=186, worksize=128. That gives a few Mhash/s more than worksize=256 and mem anywhere between 250 and 400.

The conclusion I came to with the 6870 was that the sweetspot was between 180 and 240. I was thinking the 5800s should be similar. There seemed to be a nice spike between those values, then below 180 the hash rates fell off rapidly. Did you find a similar pattern?
Pretty much spot-on. At worksize=128, between 180 and 250 the hashrate varies by a couple Mh/s, below 180 it drops fast and above 250 it starts decreasing slowly. By mem clock 300, worksize=256 is ~10 Mh/s faster and peaks at mem clock 354. But the worksize=256 peak @ mem clock 354 is still about 5 Mh/s lower than the worksize=128 peak at mem clock 186.

If this is not a dedicated rig though, then there is one issue with using mem clocks as low as ~200. Any load on the video card, such as playing video or flash, will lower the hashrate a lot more than at higher mem clocks.
hero member
Activity: 700
Merit: 500
2x 5830 Sapphire with GUIMiner "-k phatk VECTORS AGGRESSION=12 FASTLOOP WORKSIZE=256 BFI_INT" getting ~290Mhash/s per card 930/300 with open case and a fan blowing on it, staying at 60c.

quoted so I can find this later. Wink
Trixx software version 4.1 let us change voltage at 5830 ...

(that is what the software says ...=) i don't know if it really change it !

where did you find 4.1, because on the official page i can only see 4.0.1

you right!

sorry was 4.0.1 official on saphire web page ....
[/quoge]

Yep, 4.0.1 lets me change the voltage aswell, although if i go over 1030 and try to mine, gpu cracks a shit.
member
Activity: 103
Merit: 10
2x 5830 Sapphire with GUIMiner "-k phatk VECTORS AGGRESSION=12 FASTLOOP WORKSIZE=256 BFI_INT" getting ~290Mhash/s per card 930/300 with open case and a fan blowing on it, staying at 60c.

quoted so I can find this later. Wink
Trixx software version 4.1 let us change voltage at 5830 ...

(that is what the software says ...=) i don't know if it really change it !

where did you find 4.1, because on the official page i can only see 4.0.1

you right!

sorry was 4.0.1 official on saphire web page ....
sr. member
Activity: 458
Merit: 250
beast at work
i`m doing 301.xx Mh/s with my Sapphire 5830 set at 970/300 stable 24/7
hero member
Activity: 700
Merit: 500
Running my Sapphire 5830, 317.2mh/s 1025/300.
sr. member
Activity: 254
Merit: 250
Will one of you guys who are running at 300 mem clock and worksize=256 give 225 mem clock and worksize=128 a try, and let me know your before and after results.
I don't have any 5830s, but all 58xx cards should behave similarly. On my 5870s I noticed that below ~250 mem clock worksize=128 works better. In fact, the optimal settings I've found we're at mem=186, worksize=128. That gives a few Mhash/s more than worksize=256 and mem anywhere between 250 and 400.

The conclusion I came to with the 6870 was that the sweetspot was between 180 and 240. I was thinking the 5800s should be similar. There seemed to be a nice spike between those values, then below 180 the hash rates fell off rapidly. Did you find a similar pattern?

Only work for worksize=128.

My best setting (5830) is memclcok = 300 with default worksize (SDK 2.1 + poclbm), which gave me 6 mhs more. Of course, setting memclock to 200 will lower the electr bill. Smiley
legendary
Activity: 826
Merit: 1004
Will one of you guys who are running at 300 mem clock and worksize=256 give 225 mem clock and worksize=128 a try, and let me know your before and after results.
I don't have any 5830s, but all 58xx cards should behave similarly. On my 5870s I noticed that below ~250 mem clock worksize=128 works better. In fact, the optimal settings I've found we're at mem=186, worksize=128. That gives a few Mhash/s more than worksize=256 and mem anywhere between 250 and 400.

The conclusion I came to with the 6870 was that the sweetspot was between 180 and 240. I was thinking the 5800s should be similar. There seemed to be a nice spike between those values, then below 180 the hash rates fell off rapidly. Did you find a similar pattern?
newbie
Activity: 31
Merit: 0
Will one of you guys who are running at 300 mem clock and worksize=256 give 225 mem clock and worksize=128 a try, and let me know your before and after results.
I don't have any 5830s, but all 58xx cards should behave similarly. On my 5870s I noticed that below ~250 mem clock worksize=128 works better. In fact, the optimal settings I've found we're at mem=186, worksize=128. That gives a few Mhash/s more than worksize=256 and mem anywhere between 250 and 400.
Pages:
Jump to: