AMD Polaris GPU

[TheDarkBGM]

since the new drivers are out i thought id ask for those that have the 480 has it affected your power draws using f@h or you general ppd it would be great to see reports of what the 480 is doing f@h wise but it seems very few people have this card yet or arnt using it for f@h at least.

[Sven]

Test of Hardware-Sites show, that the performance doesn't change in standard configuration. The Power-Draw goes away from the PCI-E Slot to the 6pin power supply cable, which is less critical. The overall power draw doesn't change.

There is however an option in the driver where you can select the "compatibility mode". This mode puts down the overall power draw down to 150 Watts, so with this option there's probably a decrease in performance.

[foldy]

Sn1ken reported a rx 480 gpu power use of total 131 watts with old driver. So even with new driver compatibility mode limit of 150 Watts - it is still not hit with folding.

[foldy]

AMD RX 460 with 125k PPD at 75 watts for $120
AMD RX 470 with 275k PPD at 120 watts for $200

PPD is just a guess and watts are TDP and price is from newegg.

[Jejking]

AMD RX 460 with 125k PPD at 75 watts for $120
AMD RX 470 with 275k PPD at 120 watts for $200

PPD is just a guess and watts are TDP and price is from newegg.

That's just a guess, but results are now popping in from RX 480 owners, around 310-320k PPD.

That would mean something like this for the rest of the line-up, based on calculations (*) and real life prices:

RX 480 - 310k PPD, 130w, 230-280 dollar.
RX 470 - 280k PPD, 105w, 200-230 dollars. (around 88-90 per cent of 480 based on chip and clocks.)
RX 460 - 143k PPD*, 60w, 110-140 dollars. (43 per cent of SPs, TUs, 117 per cent clocks = around 50 per cent of 470 performance).

nVidia:

Titan X Pascal - 1400k PPD, ~250w, 1300 dollar.
Titan X - 700-850k PPD, ~230w, 1200 dollar (old price, now 1100-1200?)
GTX 1080 - 700-900k PPD, ~160w, 700 dollar
GTX 1070 - 600-750k PPD, ~120w, 430-470 dollar
GTX 1060 - 330-420k PPD*, ~90w, 290 dollar (around 47% of 1080, based on clocks and chip)

Nvidia is the way to go if you want low power consumption and solid folding performance, AMD is what you want if you go the cheaper route while still maintaining a fairly good PPD for MUCH lower power consumption compared to their old 28nm GPU range.

[Jejking]

I just got an RX 480, and on Windows 10, with Radeon Crimson driver 16.6.2(?)(whatever the initial release driver was) I see 270-300K points per day. This is on projects 9643 and 11402.

It runs continuously at the 1266Mhz boost clock, and a temperature of 78-81C.

I'm seriously wondering how you get the measurements of 270-300k PPD, everything the client throws in my face are messages about 28k PPD (installed it last week). Now at Project 9181, FahCore 0x21. Yesterday I went all undervolt on it and it saves me a good bundle of power on my RX 470. But the score is seriously project-depending. How on earth are we supposed to measure performance between cards? FAHBench? I don't see any translation between PPD and the scores posted over there, only vague assessments of unknown people on forums.

[foldy]

Only 28k PPD seams too low. Do you have a passkey and folded more than 10 work units and run folding 24/7 ?
Maybe you want to make a new thread with that your problem and post the logfile there so we can check if everything is setup right?

FahBench shows the raw power comparison only in ns/day. The PPD are a different not linear but more exponential curve which means if one gpu has double the FahBench value than another gpu then the PPD could be 3 times as high just as an example. That is because folding@home gives more points for a work unit if you return it faster.

Because of this users in the forum post their PPD for a GPU in a range like 250-300k PPD depending on work unit. So they monitor e.g. 10 work units and get the min and max or they take the average value, e.g. for your RX470 the post says around 280k PPD which means e.g. 250-300k PPD

[Jejking]

Only 28k PPD seams too low. Do you have a passkey and folded more than 10 work units and run folding 24/7 ?
Maybe you want to make a new thread with that your problem and post the logfile there so we can check if everything is setup right?

FahBench shows the raw power comparison only in ns/day. The PPD are a different not linear but more exponential curve which means if one gpu has double the FahBench value than another gpu then the PPD could be 3 times as high just as an example. That is because folding@home gives more points for a work unit if you return it faster.

Because of this users in the forum post their PPD for a GPU in a range like 250-300k PPD depending on work unit. So they monitor e.g. 10 work units and get the min and max or they take the average value, e.g. for your RX470 the post says around 280k PPD which means e.g. 250-300k PPD

For the first time I put a passkey on it, so that might help! I'll start a new topic (now I'm at 42k f.e.) soon.

[Jejking]

Update:

239653 Points per day. 11707 (64, 21, 0) 7 hours 37 mins 76833
Work Unit (PRCG) Work Unit (ETA) Estimated Points

Holy shit cannon balls... was it really that setting?

[foldy]

Looks good!

[MARSTG]

So anyone in the community has any hard numbers for the RX 470 and RX 460?

[davidcoton]

was it really that setting?

Yes. Of course it isn't doing any more work than before, it's just that more points are given for units returned more quickly, but only if you have a passkey. This is the Quick Return Bonus (QRB). Points are intended to recognise science value, and a quick return is valuable to the project, so more points. However the exponential system used is being rather swamped by the newer, faster GPUs claiming vast scores. It will be politically difficult, but I would not be surprised to see a different QRB scheme soon.

[antropofob]

So anyone in the community has any hard numbers for the RX 470 and RX 460?

RX 480 - 310k PPD, 130w, 230-280 dollar.
RX 470 - 280k PPD, 105w, 200-230 dollars. (around 88-90 per cent of 480 based on chip and clocks.)
RX 460 - 143k PPD*, 60w, 110-140 dollars. (43 per cent of SPs, TUs, 117 per cent clocks = around 50 per cent of 470 performance).

nVidia:

Titan X Pascal - 1400k PPD, ~250w, 1300 dollar.
Titan X - 700-850k PPD, ~230w, 1200 dollar (old price, now 1100-1200?)
GTX 1080 - 700-900k PPD, ~160w, 700 dollar
GTX 1070 - 600-750k PPD, ~120w, 430-470 dollar
GTX 1060 - 330-420k PPD*, ~90w, 290 dollar (around 47% of 1080, based on clocks and chip)

Nice chart for those wondering who the current king is:
Thanks to foldinger@reddit!

https://docs.google.com/spreadsheets/d/ ... s=57f3e0fa
- - - - - ppd/price - ppd/watt
Titan XP - 1.1 - - - - - 5.6
GTX 1080 1.3 - - - - - 5.0
GTX 1070 1.7 - - - - - 5.6
GTX 1060 1.5 - - - - - 4.2
with 3 GB 1.7 - - - - - 4.2
- -RX 480 1.3 - - - - - 2.4
- -RX 470 1.3 - - - - - 2.4
- -RX 460 1.3 - - - - - 2.4

[Jejking]

was it really that setting?

Yes. Of course it isn't doing any more work than before, it's just that more points are given for units returned more quickly, but only if you have a passkey. This is the Quick Return Bonus (QRB). Points are intended to recognise science value, and a quick return is valuable to the project, so more points. However the exponential system used is being rather swamped by the newer, faster GPUs claiming vast scores. It will be politically difficult, but I would not be surprised to see a different QRB scheme soon.

Fascinating. Obviously it's about the work, but the rewards are also nice to have

[Duce H_K_]

Does anyone have an RX 580 to bench/fold & show us results?