Topic: Faster than Phoenix (Read 4474 times)
I'm getting 351 Mhash/s under Windows 7 from an unlocked hd 6950 using the latest poclbm -f 60 (97% GPU usage). From your datas 702 Mhash/s vs 708 Mhash/s. An 8,6% higher hash rate for over $700 in not a so good deal. Buying a new card is a far better investment for most of the people.
With Phoenix on a straight stock 6990 I get 670 mhps.
Try pushing the clock up to 850-900.. should run no problem
With Phoenix on a straight stock 6990 I get 670 mhps.
yeah Phoenix miner has been the slowest on my 5970
Phoenix minder nets just under 600 (593)
poclbm is doing 660-670
GUI miner doesn't even launch cus it doesn't see my 5970 lulz
I'm buying a 6990 soon though, so I may have to start testing more again
Phoenix minder nets just under 600 (593)
poclbm is doing 660-670
GUI miner doesn't even launch cus it doesn't see my 5970 lulz
I'm buying a 6990 soon though, so I may have to start testing more again
mrb - I had posted early in the phoenix thread questioning how close this was to your hdminer. Seems you are still ahead of the pack on a 6990?!?
On a 5970, with Phoenix 1.3, 815/300 and stock voltage I am at 725 (362.5 per). I know there has been talk that the hash rate reported by Phoenix is "questionable". FWIW I get the same results (725) with the latest poclbm (with BFI_INT).
On a 5970, with Phoenix 1.3, 815/300 and stock voltage I am at 725 (362.5 per). I know there has been talk that the hash rate reported by Phoenix is "questionable". FWIW I get the same results (725) with the latest poclbm (with BFI_INT).
Shader clock can't be changed separately and GPU voltage doesn't influence speed. I have voltage at 1.02 for 830 MHz and 1.1 for 880 MHz. Memory speed is set to 150 MHz.
That's what I was talking about... On a stock card you will not observe the same Phoenix performance.
Voltage does influence speed. It changes power consumption, which is detected by PowerTune, which in turns dynamically adjusts the clock to make the speed vary by +/- 3-4% in my experience. Temperature also influences power consumption. And voltage influences temperature. It's all intertwined and makes benchmarking very difficult, especially when arguing about a few percents of speed difference.
Reducing memory speed also often modifies (either decreases or increases) performance, depending on the compute shader.
Good. I want to collect accurate data, so: stock shader clock as well as stock memory clock? No changes of the PowerTune settings? No GPU voltage mod? No mem voltage mod?
Shader clock can't be changed separately and GPU voltage doesn't influence speed. I have voltage at 1.02 for 830 MHz and 1.1 for 880 MHz. Memory speed is set to 150 MHz. I'm not sure what you're saying no to, I get slightly above 340 Mhash at 830 MHz with -f 1.
("no" to the 720 number).
Good. I want to collect accurate data, so: stock shader clock as well as stock memory clock? No changes of the PowerTune settings? No GPU voltage mod? No mem voltage mod?
Raulo: I was selling my miner privately in early Jan. And I was using it since Dec. Hence 4 months. Not sure about poclbm. On the contrary, most users seemed to have trouble finding the magic catalyst/sdk/flag combo to reach acceptable speeds. The simplicity of hdminer bypassing the whole OpenCL layer is part of the value I sell. And again, many users who report perf numbers at "stock GPU clock" don't tell you they tweak the mem clock or voltages which influence performance (edit: case in point: Grinder in this thread). Also, some 5970 models are factory-overclocked to 735MHz up from 725MHz. Users may forget this detail when comparing numbers between their 5970s.
Not 720 but 679: 2*830/880.*355 = 679 Mhash/s theoretical. However the best I have seen reported by a Phoenix user on 6990 at 830 MHz is 650 Mhash/s. 708/650 = 9%.
I'm not sure what you're saying no to, I get slightly above 340 Mhash at 830 MHz with -f 1. Well, it was not fairly quick. It took 4 months, from January to April, for an open source miner to catch up.
Not really. It took 3 months to be ahead (late January to late April is 3 months on my calendar) but poclbm caught up in early February (4th to be pricise), i.e. in 10 days from your first hdminer post. For 5970, I got 570 MH/s at stock clock with the February version and there were a lot of reports of 565-570 MH/s at stock clock depending on SDK/system/driver/options. Earlier poclbm versions were a few percent slower so hdminer was an improvement at that time. But only for 10 days.
Anyway, 9% seems like an exaggeration. I get ~360 Mhash at 880 MHz on my unlocked 6950 with poclbm, so a 6990 should give 720 at the same speed.
Not 720 but 679: 2*830/880.*355 = 679 Mhash/s theoretical. However the best I have seen reported by a Phoenix user on 6990 at 830 MHz is 650 Mhash/s. 708/650 = 9%.
But I agree that quoting a percentage of performance improvement is not ideal... there seems to be such a wide variation across all Phoenix users (due to OpenCL's complexity and CPU usage bug) that it is hard to give a % number that everyone will agree upon. Some Phoenix users even measure much less than 650 Mhash/s on the 6990. What I can promise though is the exact perf numbers I quoted in the hdminer thread: 708 Mhash/s at 830MHz and 746 Mhash/s at 880MHz.
Assuming Phoenix (or any other open program) will not catch up. For 5xxx series, this assumption has proven false fairly quickly.
Well, it was not fairly quick. It took 4 months, from January to April, for an open source miner to catch up.
I am glad your price is so high. If it was lower it would distort the mining market and I would be
Forced to buy it to keep up. At that price so few people buy it there is little impact.
And of course, if anybody has purchased it and found that it works really well it would be stupid to tell others. Anyway, 9% seems like an exaggeration. I get ~360 Mhash at 880 MHz on my unlocked 6950 with poclbm, so a 6990 should give 720 at the same speed. That's even without "cheating" by using dynamic speed. Forced to buy it to keep up. At that price so few people buy it there is little impact.
You would need to be running 8 6990's for that to pay off vs buying another card.
Assuming Phoenix (or any other open program) will not catch up. For 5xxx series, this assumption has proven false fairly quickly.
You would need to be running 8 6990's for that to pay off vs buying another card.
I am glad your price is so high. If it was lower it would distort the mining market and I would be
Forced to buy it to keep up. At that price so few people buy it there is little impact.
I am glad your price is so high. If it was lower it would distort the mining market and I would be
Forced to buy it to keep up. At that price so few people buy it there is little impact.
A current mtgox exchange rate value of $836.50 US... Ouch, I've never paid that much for software. Is there 24/7 support?
I operate on a best effort basis via email or skype. I have asked customers to post testimonials in the hdminer thread to give you an idea of how good/bad my support is. (Don't trust me. Trust my users.)
A current mtgox exchange rate value of $836.50 US... Ouch, I've never paid that much for software. Is there 24/7 support?
I apologize for the shameless plug. Apparently some in the Mining subforum don't read the Marketplace subforum, so I would like to say:
There is a miner faster than Phoenix (+9%) on the HD 6990, it's called hdminer.
(It is not free, so do the math to determine whether its price is worth it or not for you.)
There is a miner faster than Phoenix (+9%) on the HD 6990, it's called hdminer.
(It is not free, so do the math to determine whether its price is worth it or not for you.)
Jump to: