mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing

Reply
 
Thread Tools
Old 2022-05-12, 06:04   #1
AwesomeMachine
 
Apr 2018
USA

11012 Posts
Default Is it worth it to get a nvidia tesla k80 these days.

In the chart of gpu benchmarks on mersenne.ca, it says the tesla K80 will produce 110GhzDays/day.


Does that mean what I think, it's many times faster than any 'c'pu. That would get the same amount of credit in 5 days that my cpu gets in 90 days. This just doesn't sound realistic. Could someone explain what is going on with that?


Thanks for all the help.


I do want to obtain a gpu board for factoring with primenet, but I need to figure out what hardware I actually want. So, help me get that far, and I'll dive right in! I already know pretty much how to set up a tesla gpu on a desktop. It's just finding a reference for the benchmarks, so I can compare cpus to gpus, so I can determine the value of the hardware, that's thrown me for a loop.
AwesomeMachine is offline   Reply With Quote
Old 2022-05-12, 10:20   #2
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

3D616 Posts
Default

Because the kind of work that TF is fits GPUs much more than CPUs, GPUs are so much faster than CPUs in this case. For example, if you take a recent card like the 3060 Ti, which is definitely not the highest end card and its prices are getting lower at the moment, you would look at 2,267 GHz·d/d! That's more than a magnitude more!

So: If you do not have to pay for electricity and you can get the Tesla K80 for one twentieth (or less) of the price of an 3060 Ti, go for it (this might also apply in other cases where you need some features the Tesla cards might have that the 30xx series does not have). Otherwise, there might be better choices.
kruoli is offline   Reply With Quote
Old 2022-05-12, 12:25   #3
axn
 
axn's Avatar
 
Jun 2003

5×29×37 Posts
Default

K80 produces 110 GHz D/d on PRP tests. It produces 800 GHz D/d on TF.

If factoring is what you want to do, GTX 16xx, RTX 20xx and RTX 30xx are the things to get. For example, 1650 produces 800 Gd/d on TF while consuming much less power compared to K80.
axn is offline   Reply With Quote
Old 2022-05-12, 13:31   #4
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

11×71 Posts
Default

Tesla boards usually require different Nvidia video drivers than consumer boards.
This usually prevents any mixing boards within the same machine..

Heat and power issues could be a problem.

The K80 is a 2 GPU board. These GPUs work separately. Performance figures might be quoted combining both GPUs but your application might only use 1 GPU. This could disappoint you when you start running your application to see only half the performance you expected.

Deciding on an approach to factoring ( TF ? PRP ? P-1? ) would make it easier to give better advice.

Last fiddled with by tServo on 2022-05-12 at 13:36
tServo is offline   Reply With Quote
Old 2022-05-12, 14:55   #5
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·7·463 Posts
Default

Something else to consider is server type GPUs such as Tesla K80 usually have no fans of their own, expecting high pressure blowers in rack mount server cases to push air through them lengthwise. Putting those in a home computer, they will lack cooling, and performance, longevity, or both may suffer. There are third-party blower assemblies that can be attached to the front end of such GPUs, if space is available, & which MIGHT provide adequate cooling. Consumer GPUs have their own fans built in.

Also, for best power efficiency (computations/sec/watt), operate a GPU at a well chosen midrange power level.
I just recently went through a test cycle in mfaktc on a GTX 1080 Ti which can be set from 125 to 250W power limit.
(nvidia-smi -i 0 -pl <watts>) Highest indicated GHD/day/watt (6.46) in mfaktc was at 163 watts. (Minimum power efficiency was at minimum power, 4.8.) That's after tuning for mfaktc.ini parameters first, & running single instance, on Windows 10.

Last fiddled with by kriesel on 2022-05-12 at 14:56
kriesel is offline   Reply With Quote
Old 2022-05-16, 20:33   #6
AwesomeMachine
 
Apr 2018
USA

D16 Posts
Smile From what I've learned by the replies ...

I guess then GHzDays/d directly correlates between CPUs and GPUs, GPUs being much more productive, because they are much more adept at factoring type work, and the memory is faster. I would like to begin with TF work on the GPU. But, depending on the software, I might change that. It all seems more finicky than mprime for CPus! But, everything changes all the time.



Money isn't really much of an issue, except all other things being equal, the less I have to spend the better. People who spend extravagantly without evaluating the options just sink into debt. I know a married couple who are both Doctors. They always complain about being short of money!


So I'll analyze the GeForce boards mentioned. It's just that the Tesla boards are in much less demand (no monitor connections), so prices are very low. But cooling solutions are a pain, and power consumption is astronomical! There's no way to mount a heatsink/fan directly on the GPUs.


I've only found 2 solutions I would even try. I guess I could use thermal-transfer epoxy to affix a decent cooler to the GPUs. But then that's it. NO MORE TRIES!


i'LL check back in a while. I'm going to purchase hardware. This is a hands on type of endeavor.



Thanks for all the help!
AwesomeMachine is offline   Reply With Quote
Old 2022-05-18, 15:31   #7
Magellan3s
 
Mar 2022

61 Posts
Default

Quote:
Originally Posted by AwesomeMachine View Post
I guess then GHzDays/d directly correlates between CPUs and GPUs, GPUs being much more productive, because they are much more adept at factoring type work, and the memory is faster. I would like to begin with TF work on the GPU. But, depending on the software, I might change that. It all seems more finicky than mprime for CPus! But, everything changes all the time.



Money isn't really much of an issue, except all other things being equal, the less I have to spend the better. People who spend extravagantly without evaluating the options just sink into debt. I know a married couple who are both Doctors. They always complain about being short of money!


So I'll analyze the GeForce boards mentioned. It's just that the Tesla boards are in much less demand (no monitor connections), so prices are very low. But cooling solutions are a pain, and power consumption is astronomical! There's no way to mount a heatsink/fan directly on the GPUs.


I've only found 2 solutions I would even try. I guess I could use thermal-transfer epoxy to affix a decent cooler to the GPUs. But then that's it. NO MORE TRIES!


i'LL check back in a while. I'm going to purchase hardware. This is a hands on type of endeavor.



Thanks for all the help!
If all you plan on doing is trial factoring type of work, I definitely recommend the Nvidia 3xxx series of graphics cards.

Last fiddled with by Magellan3s on 2022-05-18 at 15:31
Magellan3s is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Safe Temperature for a Tesla K20xm? EdH GPU Computing 6 2022-03-01 22:43
Help needed - needing a Nvidia Tesla K80 in a Dell 7920 tower workstation drkirkby GPU Computing 22 2021-07-24 09:15
Just some fun playing with a Tesla P100 plus a question... JonRussell Hardware 9 2017-04-27 11:46
Tesla P100 — 5.4 DP TeraFLOPS — Pascal Mark Rose GPU Computing 52 2016-07-02 12:11
Running on /--Prime95--/ Lucas-Lehmer on Nvidia Tesla S1070 optimus GPU Computing 4 2013-08-31 03:01

All times are UTC. The time now is 05:34.


Tue May 24 05:34:59 UTC 2022 up 40 days, 3:36, 0 users, load averages: 0.90, 1.17, 1.17

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔