So do I understand this correctly that a Radeon VII is about 1.7x more productive than a 3090 in terms of tasks/day? The 300W seems pretty bad considering the Radeon VIIs drawing ~170W of power at full compute 2x tasks... What's the deal? (buys another Radeon VII)
the deal is that the apps here are not well optimized for nvidia cards. and part (only part) of the high power draw from the 3090 specifically seems to come from the 24x GDDR6X memory modules.
the efficiency of the 3080 should be a bit better since it has less modules sucking up power but not really being utilized. but nvidia will be at a disadvantage without a CUDA app.
but it seems all the folks running Radeon VIIs always only run the higher credit paying GR tasks. I wonder what the GW performance is like on them? 15 hosts in the top 50 crunchers with Radeon VIIs and not a single one running GW tasks.
the deal is that the apps here are not well optimized for nvidia cards. and part (only part) of the high power draw from the 3090 specifically seems to come from the 24x GDDR6X memory modules.
Not well optimized seems like an understatement! The ampere cards are beasts in terms of computing, however the einstein apps can only access a fraction of this. Seems like a waste of money to buy NVIDIA to crunch for einstein as the cards are quite expensive. We need a CUDA app like we had years ago... Is this something that is planned for or totally ruled out?
but it seems all the folks running Radeon VIIs always only run the higher credit paying GR tasks. I wonder what the GW performance is like on them? 15 hosts in the top 50 crunchers with Radeon VIIs and not a single one running GW tasks.
No surprise, only with GW-WUs you are not under the top 50 hosts. A single VII with FGRPB1G generates about 1.5-1.6 million per day.
My VII with 6x O2MDF Spotlight tasks needs between 12-15 min. per task, about 0.6-0.7 Mio per day.
GPU: 980 mV ,Powerlimit 180 W,
A3700x in Ecomode, Wallpower ~310 W
the 3080Ti is about 70% faster than a 2080Ti. Ignore previous comments in this thread from last year regarding poor Ampere performance. A new Gamma Ray application has been released that greatly improves performance of all Nvidia cards back to the Maxwell generation, but Ampere sees the best gains from it.
really depends on the clock speeds and power limits you enforce, so there's some variance depending on exact config. I run my 3080Ti at 300W, and that's a good balance between power draw and performance in my opinion. Einstein Gamma Ray performance is sensitive to memory latency and speed, so memory overclocks have a good impact. i run a +50MHz core OC, and a +1000MHz memory OC at the 300W power limit. be sure to test your stability if you want to overclock.
remember to only run 1 task at a time for best performance.
So do I understand this
)
So do I understand this correctly that a Radeon VII is about 1.7x more productive than a 3090 in terms of tasks/day? The 300W seems pretty bad considering the Radeon VIIs drawing ~170W of power at full compute 2x tasks... What's the deal? (buys another Radeon VII)
the deal is that the apps
)
the deal is that the apps here are not well optimized for nvidia cards. and part (only part) of the high power draw from the 3090 specifically seems to come from the 24x GDDR6X memory modules.
the efficiency of the 3080 should be a bit better since it has less modules sucking up power but not really being utilized. but nvidia will be at a disadvantage without a CUDA app.
but it seems all the folks running Radeon VIIs always only run the higher credit paying GR tasks. I wonder what the GW performance is like on them? 15 hosts in the top 50 crunchers with Radeon VIIs and not a single one running GW tasks.
_________________________________________________________________________
Ian&Steve C. wrote: the deal
)
Not well optimized seems like an understatement! The ampere cards are beasts in terms of computing, however the einstein apps can only access a fraction of this. Seems like a waste of money to buy NVIDIA to crunch for einstein as the cards are quite expensive. We need a CUDA app like we had years ago... Is this something that is planned for or totally ruled out?
Ian&Steve C. schrieb: but it
)
No surprise, only with GW-WUs you are not under the top 50 hosts. A single VII with FGRPB1G generates about 1.5-1.6 million per day.
My VII with 6x O2MDF Spotlight tasks needs between 12-15 min. per task, about 0.6-0.7 Mio per day.
GPU: 980 mV ,Powerlimit 180 W,
A3700x in Ecomode, Wallpower ~310 W
bumping this back up. I just
)
bumping this back up. I just got a 3080Ti not too long ago. hoping to see more Ampere on the project :)
_________________________________________________________________________
To get more petformance out
)
To get more petformance out of NVIDIA cards would require constant twiddling of the code.
I’m constantly twiddling
)
I’m constantly twiddling
_________________________________________________________________________
My new Water Cooled 3080ti
)
My new Water Cooled 3080ti should arrive on Dec 13th. Will be interesting to see how it does as compared to my existing 2080TI....
You can look at Ian's hosts
)
You can look at Ian's hosts to see the comparison between his 3080 Ti and his 2080 Ti's.
the 3080Ti is about 70%
)
the 3080Ti is about 70% faster than a 2080Ti. Ignore previous comments in this thread from last year regarding poor Ampere performance. A new Gamma Ray application has been released that greatly improves performance of all Nvidia cards back to the Maxwell generation, but Ampere sees the best gains from it.
really depends on the clock speeds and power limits you enforce, so there's some variance depending on exact config. I run my 3080Ti at 300W, and that's a good balance between power draw and performance in my opinion. Einstein Gamma Ray performance is sensitive to memory latency and speed, so memory overclocks have a good impact. i run a +50MHz core OC, and a +1000MHz memory OC at the 300W power limit. be sure to test your stability if you want to overclock.
remember to only run 1 task at a time for best performance.
_________________________________________________________________________