mersenneforum.org  

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

Reply
 
Thread Tools
Old 2019-11-09, 03:00   #1
philbo0042
 
Oct 2019

2×7 Posts
Default New PC build

Greetings. Brand new shiny noob here.

This is my first forum post, so please forgive me if I break any rules or etiquette. I have several questions that I am combining into one post. Many of my concerns have already been answered by looking through this forum’s threads, but I still have more I would like to know.

My current build is an i5-4690k on the Z97 chipset with 16GB RAM running at 2.4 GHz. I am running an LL test on #332,220,437. The test is estimated to take about 105 days. My best iteration time is about 25.5 ms. My best throughput is about 39 iterations per ms.

I would like to build a PC that will test much faster. My budget is $900 to $1,000. I don’t play games currently, but I do like to overclock and tinker with settings. This build is not specifically for Prime95 but I thought, “Hey, since I am building this anyways I may as well build it for what I am currently using it for.”

For now, I am leaning towards the following:
- Ryzen 3700X (8 cores/16 threads with 32 MB cache)
- MSI X570 Gaming Edge
- 16 GB 2x8 G.Skill Ripjaws V Series 3200 #F4-3200C16D-16VGB (dual-rank)
- Seasonic FOCUS 450 watt 80+ PSU
- Reusing previous SSD (either SATA 960 Evo or M.2 970 Evo)
- Least expensive, but still decent video card

1. What are your iteration times/throughput for 100,000,000+ digits and what are your specs?

2. What would you change on my proposed build?
a. Should I choose a comparable Intel build for better throughput?

3. Can I run more than one instance of Prime95 to run multiple LL tests?

4. How would I set up workers? I don’t yet understand how workers relate to cores/threads.

5. Should I fill all four DIMM slots?

Thank you for taking the time to read this.
philbo0042 is offline   Reply With Quote
Old 2019-11-09, 03:18   #2
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

157718 Posts
Default

You may not like my answer.

Keep your present PC and puchase a Radeon VII GPU for $700. This particular GPU will have roughly 6 to 8 times the throughput of the CPUs that fit your budget. See https://www.mersenne.ca/cudalucas.php
Prime95 is offline   Reply With Quote
Old 2019-11-09, 04:12   #3
GP2
 
GP2's Avatar
 
Sep 2003

22×3×5×43 Posts
Default

Are they still available? I thought they discontinued them.

How noisy are they, by the way.
GP2 is offline   Reply With Quote
Old 2019-11-09, 04:23   #4
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

5108 Posts
Default

Quote:
Originally Posted by Prime95 View Post
Keep your present PC and puchase a Radeon VII GPU for $700. This particular GPU will have roughly 6 to 8 times the throughput of the CPUs that fit your budget. See https://www.mersenne.ca/cudalucas.php
Seconded, if really thinking about large PRP performance. The Zen 2 chips, while great at smaller FFT sizes due to the large L3 cache size, won't be much of an improvement at 18432K. As a comparison, the Ryzen 3600 will do 25.4 ms/iter (throughput 39.3 iters/sec) so it's pretty much equal to what you have now. Having eight cores instead of six won't change it, since by that point, the calculation speed is limited by RAM bandwidth.

As a comparison, in the 100M digit exponent benchmarh thread the Radeon VII went at about 3.66 ms/iter with reasonable power settings and 3.39 ms/iter with unreasonable (fan speed/noise) settings.

A few relevant datapoints from smaller FFT sizes for the Ryzen 3600:
2688K 1.78 ms/iter (will drop a bit with eight cores, there's still some L3 bandwidth left)
2880K 1.93 ms/iter (ditto)
5120K 5.01 ms/iter (mostly DRAM limited by now, won't change much with 8 cores anymore)
nomead is offline   Reply With Quote
Old 2019-11-09, 04:26   #5
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

3·7·11·31 Posts
Default

https://www.newegg.com/xfx-radeon-vi...ion=radeon+vii

https://www.newegg.com/asrock-radeon...ion=radeon+vii

Alas, a $130 off sale just ended
Prime95 is offline   Reply With Quote
Old 2019-11-09, 05:13   #6
kracker
ἀβουλία
 
kracker's Avatar
 
"Mr. Meeseeks"
Jan 2012
California, USA

5·433 Posts
Default

Quote:
Originally Posted by Prime95 View Post
You may not like my answer.

Keep your present PC and puchase a Radeon VII GPU for $700. This particular GPU will have roughly 6 to 8 times the throughput of the CPUs that fit your budget. See https://www.mersenne.ca/cudalucas.php
Ditto. If you really wanted to upgrade your core, a system with a quad channel memory system is the only way you'll get any decent speed boost, which means you'll need a new cpu/mobo/ram... But if you can get a Radeon VII, it'll blow most if not all cpu's out of the water - funny thing is if you can't get a VII a system upgrade might be the way to go.

Also you(or mod can sensor?) probably should not post the full number that you're working on...
kracker is offline   Reply With Quote
Old 2019-11-09, 06:40   #7
nomead
 
nomead's Avatar
 
"Sam Laur"
Dec 2018
Turku, Finland

5108 Posts
Default

OK, here's a reply to some other points in the first post:
Quote:
Originally Posted by philbo0042 View Post
Greetings. Brand new shiny noob here.
Welcome
Quote:
Originally Posted by philbo0042 View Post
2. What would you change on my proposed build?
Well if you decide not to go the Radeon VII route, consider running trial factoring on the GPU. Something like a GTX 1660 is really sweet for running mfaktc, bang for the buck. For complicated reasons (that I'm not going into here), the NVidia Turing cards (GTX16/RTX20) are really good for trial factoring, but really abysmal for any primality testing.
Quote:
Originally Posted by philbo0042 View Post
3. Can I run more than one instance of Prime95 to run multiple LL tests?
I think that is not recommended. You can, in principle, run multiple workers on one instance of Prime95 but since even one test will saturate the available RAM bandwidth, it won't improve anything. The throughput will likely be worse than with just one worker.
Quote:
Originally Posted by philbo0042 View Post
4. How would I set up workers? I don’t yet understand how workers relate to cores/threads.
Prime95 just divides the number of available cores by the number of workers and distributes work accordingly. So two workers on 8 cores would give 4 cores to each assignment. But like I said earlier, these big tests are RAM bandwidth limited, so having more than one worker doesn't make sense. I just got an i5-8400 box at work, and on that processor, it seems to be preferable to run one worker, four cores, even though six are available; the timings are slightly better that way. On the previous generation Zen+ since the FPU implementation sucked a bit, there was a slight throughput advantage in running one worker per core, as the RAM wouldn't get saturated. (Used to run a Ryzen 3 2200G with four workers for quite a while) But that's changed in the current generation.
Quote:
Originally Posted by philbo0042 View Post
5. Should I fill all four DIMM slots?
I don't think it makes a difference with regards to primality testing. There are two memory channels on the processor, so as long as both channels are populated with one DIMM, everything is fine.

Side note: I seem to get the most joy from getting new hardware and fiddling around with it, trying to find the best combination of work by benchmarking it and tuning small details in both the hardware and the software. Just running the tests after that stage is relatively tedious...
nomead is offline   Reply With Quote
Old 2019-11-09, 15:20   #8
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

23·79 Posts
Default

If you were going for pure Prime95 Radeon VII would be the answer. As you're going for a PC build with Prime95 potential and light gaming here's a baseline: https://pcpartpicker.com/list/gGByk6

The M.2 drive is just for OS or for very light storage, you'd likely want a 2.5" SSD for 512GB-1TB of additional storage, beyond that a 3TB-4TB 3.5" HDD instead. The 3600 is great for budget gaming/everything, in terms of CPUs for Prime95 it's also a good budget option. The RX 570 is the best bang for buck you can get for low-midrange gaming. 80+ gold supply is the minimum I normally consider, YMMV but at least go for bronze. Unless you have a specific need for X570 features it's a waste going for an enthusiast chipset. If it'll be a few months until your build B550 might be out by then, that's the chipset to go for if it's priced comparably to current B450's.

With a bigger budget the first thing I'd do is get a 580 instead of a 570. If it's going to be a few months until the build the low-end 7nm GPU's from AMD might be released. If an RX5600 is priced comparably to a 580 on release it's probably the way to go. The next step up from there is a Vega 56, then an RX 5700, then an RX 5700 XT. nvidia is not price-competitive in the low to midrange, for the top end they are the only option. The minimum nvidia GPU to consider is the RTX 2060 Super and only if it's way cheaper than the 5700 XT (AMD/nvidia price differentials vary worldwide). The 5700 XT is only slightly slower than a 2070 Super for gaming ( https://youtu.be/pcGWIJW3iNU?t=446 ), so beyond the 5700 XT you're looking at the 2070 Super, then 2080 Super, then 2080 TI (nvidia has many super/non-super/ti/current-gen/last-gen variants that slot between these in some way that might make sense at the right price). Raytracing in games is a gimmick for the next few years so IMO don't bother unless you have a non-gaming use for it.

The next upgrade I'd pick if you can make good use of the CPU (or are planning to keep the build for a long time with no upgrades, IMO 3+ years) is to a 3700/X. If you have a specific need for more than 16GB of RAM (most people don't) then you could double up to 4x8GB. Peripherals are very subjective but if you've never considered them important you'd be surprised how much difference a $30 keyboard makes over cheap tat.
M344587487 is offline   Reply With Quote
Old 2019-11-09, 15:45   #9
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22·11·101 Posts
Default

Quote:
Originally Posted by M344587487 View Post
If you have a specific need for more than 16GB of RAM (most people don't) then you could double up to 4x8GB.
12+GB can be used up in P-1 on midrange exponents for days at a time, per worker running P-1 in prime95/mprime. GPU apps that do the P-1 gcd on the cpu will need an additional similar amount of system ram for it.

Last fiddled with by kriesel on 2019-11-09 at 15:46
kriesel is offline   Reply With Quote
Old 2019-11-09, 18:02   #10
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

22·5·7·11 Posts
Default

Quote:
Originally Posted by nomead
Something like a GTX 1660 is really sweet for running mfaktc, bang for the buck.
Seconded here too. My son has a 1660. It really goes.
storm5510 is offline   Reply With Quote
Old 2019-11-09, 18:15   #11
M344587487
 
M344587487's Avatar
 
"Composite as Heck"
Oct 2017

63210 Posts
Default

That's true, in some markets the 1660 could be competitive. Looks like the 1660 Super is the latest refresh that sits between the 580 and Vega 56 in my list, at least here the non-super 1660 is not worth getting: https://youtu.be/Y1Rr1e99eBQ?t=803
M344587487 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
New build help Prime95 Hardware 147 2018-11-10 00:58
Weird build? Xyzzy Hardware 122 2016-01-14 10:32
32-bit CUDA build? f1pokerspeed Msieve 2 2013-12-30 01:14
New build with ECC ram PageFault Hardware 1 2012-08-23 03:44
New build PageFault Hardware 28 2011-09-19 07:15

All times are UTC. The time now is 13:27.

Mon Sep 28 13:27:22 UTC 2020 up 18 days, 10:38, 1 user, load averages: 1.93, 1.95, 2.25

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, 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.