Next POTM...
August: bill1024  August 20-23rd PG GCW-LLR Solar Eclipse Challenge.  Can crunch a 4th day, clean up if you want.

 
Anyone got suggestions for September... I have plenty, but someone else needs a turn..

bill1024
Also, any dates in mind to do this POTM?

According to the first post in this thread, Moo is the October POTM. Specific dates to be determined.

bill1024
I forget, does Moo use DP, is that why AMD cards do so well?
If so my GTX Titan 6gb should do very well with DP enabled.
 
Also, any dates in mind to do this POTM?

Nope, Tested faster with OC on the Titans with DP off.
MilkyWay@home is the only Project that I have tested DP with and works well.
I run 10 GPU Tasks GPU with 0.2 CPU and takes about 58% CPU Usage and 100% GPU Usage.

bcavnaugh

bill1024
I forget, does Moo use DP, is that why AMD cards do so well?
If so my GTX Titan 6gb should do very well with DP enabled.
 
Also, any dates in mind to do this POTM?

Nope, Tested faster with OC on the Titans with DP off.
MilkyWay@home is the only Project that I have tested DP with and works well.
I run 10 GPU Tasks GPU with 0.2 CPU and takes about 58% CPU Usage and 100% GPU Usage.

Thanks Bill, the other DP that I can think of is PG genfer 20 and 21 sub projects.

bill1024

bcavnaugh

bill1024
I forget, does Moo use DP, is that why AMD cards do so well?
If so my GTX Titan 6gb should do very well with DP enabled.
 
Also, any dates in mind to do this POTM?

Nope, Tested faster with OC on the Titans with DP off.
MilkyWay@home is the only Project that I have tested DP with and works well.
I run 10 GPU Tasks GPU with 0.2 CPU and takes about 58% CPU Usage and 100% GPU Usage.

Thanks Bill, the other DP that I can think of is PG genfer 20 and 21 sub projects.

Cool I will have to run some PG Genfer 20/21 Tasks both are Very Long Talks.
Never have ran any PG GPU Tasks on my Titans that I can Recall with DP Enabled.

Moo (RC5-72 DNet) is strictly integer operations - mostly rotates, adds, and "move the data into and out of memory" work.
 Current Pascal-based NVidia cards are comparable on raw keyrate, but their much higher "normal" pricing tended to make them have a much worse $/keyrate ratio prior to the "cryptocoin GPU shortage" and resulting gouge pricing on the high-end RX series cards.
 Only reason I have keyrates on the GTX 1070 and 1080 was curiosity - it doesn't take long to set up the actual DNet client that Moo "wraps" to test speeds with.
 I keep thinking I should do that sometime on my 1080ti based rig, but just haven't gotten around to it (yet).
 
Milkyway is largely FP64 (which is why most of the "top computers" in that project run the Tahiti chip AMD cards, they're only topped on FP64 performance by VERY VERY EXPEN$IVE workstation-specific cards, so you see a LOT of 7950/7970/R9 280/R9 280x cards used in that project.
 R9 380/380x were almost the ONLY AMD 3xx cards that weren't the same GPU as the 2xx varients - they were based on the chip used in the R9 285 instead IIRC (which was the LAST 2xx model released).
 
 The Hawaii cards (R9 290/290x/390/390x) are close on FP64 to the Tahiti cards, but use more power for a little less performance.

QuintLeo
Moo (RC5-72 DNet) is strictly integer operations - mostly rotates, adds, and "move the data into and out of memory" work.
 Current Pascal-based NVidia cards are comparable on raw keyrate, but their much higher "normal" pricing tended to make them have a much worse $/keyrate ratio prior to the "cryptocoin GPU shortage" and resulting gouge pricing on the high-end RX series cards.
 Only reason I have keyrates on the GTX 1070 and 1080 was curiosity - it doesn't take long to set up the actual DNet client that Moo "wraps" to test speeds with.
 I keep thinking I should do that sometime on my 1080ti based rig, but just haven't gotten around to it (yet).
 
Milkyway is largely FP64 (which is why most of the "top computers" in that project run the Tahiti chip AMD cards, they're only topped on FP64 performance by VERY VERY EXPEN$IVE workstation-specific cards, so you see a LOT of 7950/7970/R9 280/R9 280x cards used in that project.
 R9 380/380x were almost the ONLY AMD 3xx cards that weren't the same GPU as the 2xx varients - they were based on the chip used in the R9 285 instead IIRC (which was the LAST 2xx model released).
 
 The Hawaii cards (R9 290/290x/390/390x) are close on FP64 to the Tahiti cards, but use more power for a little less performance.

Great info once again QuintLeo.
Please do, if you get some time to test the 1080ti's and their performance in comparison...would be much appreciated..
 
Cheers
Steve
 

Aorus GTX 1080ti wasn't impressive - 4.8 Mkeys/sec after filtering through ALL of the possible options and with a mild overclock - but that was on the CUDA client, for some reason the openCL client wasn't even identifying the card, and the openCL client USUALLY runs faster.
 
I'm pretty sure the numbers I have from the 1070 and 1080 were on the OpenCL client, and I was expecting a LOT more out of a 1080 ti given the much higher core count vs the slightly lower core clock.
 
 Odd part is that GPU-Z says OpenCL *IS* enabled on the GPU....
 
 

Update - reminder that the previous numbers were running the DNet CUDA client.
 
Got drivers updated so they were WORKING on OpenCL and updated that machine to the 521 version from the 520 version - not sure if that was the fix or the drivers were as I forgot to test it right after the drivers update.
 
 Running 378.92 drivers, and Dnet OpenCL client 521, the Aorus was pushing about 9.5 Megakeys/sec (I also have a Gigabyte Windforce GTX1080 ti in that same machine, it was pulling a hair under 9).
 I didn't switch the block size on it though, and it was going through a 32 block workunit so fast I suspect it would have a somewhat faster total net keyrate if I upgraded the client to using "128" or "256" block workunits.
 For perspective, Moo Wrapper runs "64" block workunits for most of it's work, with an occasional smaller block - that's what the end of the work unit naming is about, the number of workunits and the total number of "standard Dnet RC5-72 blocks" in those workunits.
 
 THIS is more like what I was expecting out of the 1080 ti.
 
 The scary part - I expect the Vega RX 64 to be pull over 10 Megakeys/sec (the Frontier Edition should almost match that) and the Vega RX 56 to pull 9+, if they scale properly to core count and core clocks.
 
 Now if we can just talk the Master of Moo to update the client to 521 from 520....
 
 
 For perspective - my R9 290 cards pull around 4.5 Mkeys/sec (overclocked to 1100 core thanks to a BIOS mod by TheStilt), my RX 470 cards pull right around 4 depending on the overclock (hair under stock, hair over with a mild overclock on core).
 

Running 378.92 drivers, this is the best Driver ATM for any GTX 1080 Ti & Folding as Well the best driver for the 1080 Ti Cards.