mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-04-16, 18:31   #1
Viliam Furik
 
Jul 2018
Martin, Slovakia

23·33 Posts
Default Threadripper 3990X vs. Threadripper 3970X

How much performance difference (expressed as a rough percentual benefit of Config. 2 above Config. 1) can I expect between the following two configurations, namely in current double-check (50M), first-time test (90M-95M) and 100M digits (332M) ranges:

Config. 1)
- AMD RYZEN Threadripper 3970X, running at 4,1 GHz, having 32 cores and 128 MB of L3 cache
- 64 GB of 4400 MHz RAM

Config. 2)
- AMD RYZEN Threadripper 3990X, running at 3,8 GHz, having 64 cores and 256 MB of L3 cache
- 64 GB of 4400 MHz RAM

If somebody can give a rough estimate of timings, that would be great!
Viliam Furik is offline   Reply With Quote
Old 2020-04-17, 02:52   #2
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

22×3×739 Posts
Default

In theory, let x be the performance of (1), then the performance of 2 will be about 1.85x. This is done by multiplying by two (double number of cores) and scaling the clocks. So it is optimistic, and a bit dumb. In practice, you may have heat issues, memory limitations and all the stuff... It will heavily depend of your mobo/chipset (memory speed, channels). Here the numbers are not so much different, so according with that, the pessimistic side is that you may get a boost from x to 1.28x (done by dividing the two numbers there).

But don't take me seriously, I know this calculus is dumb, and am an intel guy, not much knowledge about rippers. On the benchmark page (as well as here on the forum, there is an "eternal benchmarks" tread, you can search for it) you can find better info (and a lot of it).

Last fiddled with by LaurV on 2020-04-17 at 02:55
LaurV is offline   Reply With Quote
Old 2020-04-17, 13:29   #3
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

17×31 Posts
Default

I realize your question is one concerning the performance of two specific processors but here is a thought:

I was reading about the new AMD Epyc processors a few months ago and was intrigued by their 8 channel memory. This is double the specified Threadrippers. Since Prime95's performance is very closely tied to memory accesses, I wanted to see if an Epyc machine could be configured for a reasonable amount of money and I found it could.

I found two Epyc processors, one for 750 dollars and a slightly faster one for 1000. A motherboard for these that has one processor slot costs 550 - 650 dollars and has 8 memory dimm slots; enough for the 8 channel access. This cost surprised me since Epyc is considered an "enterprise" processor for servers.
The Threadripper 3990x costs about 4,000 dollars.

It would be fascinating to compare these 2 systems.
tServo is offline   Reply With Quote
Old 2020-04-17, 14:51   #4
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

17×31 Posts
Default Or how about this beast?

If you REALLY want LOTS of threads, how about an HP blade server that has 32 8 core Xeons with 64GB ram per pair of processors for 3,325 US dollars !

The only problems are:
No disk included.
The memory is somewhat slow.
How to cool this thing.
Affording the nuclear reactor to run it ( it has 6 2.5 kw power supplies ) !!!

https://www.ebay.com/itm/HP-Blade-Se...3/202960056632
tServo is offline   Reply With Quote
Old 2020-04-17, 17:51   #5
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

112816 Posts
Default

Quote:
Originally Posted by tServo View Post
If you REALLY want LOTS of threads, how about an HP blade server that has 32 8 core Xeons with 64GB ram per pair of processors for 3,325 US dollars !
Translated to MF units: 1/3rd of a RyanP.

The strongest yet reminder that our hobby is limited by power costs rather than hardware costs. That's an incredible machine for the price!

Hrmm.. I *do* have a 240V/50A circuit in my garage... I could wire up some regular plugs and fire this up! Nobody's using the electric-car-charger anyway these days... 16 blades, each blade 250-300W with all cores running.... 5kW should cover it?
VBCurtis is offline   Reply With Quote
Old 2020-04-17, 18:20   #6
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2·4,643 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
Hrmm.. I *do* have a 240V/50A circuit in my garage... I could wire up some regular plugs and fire this up! Nobody's using the electric-car-charger anyway these days... 16 blades, each blade 250-300W with all cores running.... 5kW should cover it?
I know you're (half) joking, but... Remember that a 240V (more likely 208V) circuit will be two joined phases. In order to use that run at 120V, you'd split out the phases to be two different outlets. This can actually be done using a single wall outlet (if appropriately rated) by breaking the copper bridge on the outlet's wiring terminals.

Many 120V servers provide the option for redundant power supplies. It's advisable to feed each from a different circuit (which you'd have) to order to balance the load, and provide redundancy.
chalsall is offline   Reply With Quote
Old 2020-04-17, 19:53   #7
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

23×32×61 Posts
Default

I thought most computer power supplies have a switch (or auto-detect?) to run 240 vs 120V; is it not valid that in this dreamworld I could just wire a regular-looking USA plug (well, six) to the 240V circuit, and run computer power supplies to said plugs without splitting the phase/voltage down to 120+ground?

I figured 15kW of power supplies had to include redundancy; I'm not server-literate enough to know if it's more like "7500W rating to run machine, all supplies redundant", or some sort of management scheme where, say, 4 power supplies is enough for the entire machine but 6 exist for redundancy purposes.

My garage lacks a second separate circuit at 240, so I guess I'd have to do without that level of redundancy if physics allows me to run the servers at 240.

ETA: My solar-inverter claims output line voltage of 236-241 Volts, so I think it's actually 240 and not 208?

Last fiddled with by VBCurtis on 2020-04-17 at 19:54
VBCurtis is offline   Reply With Quote
Old 2020-04-17, 20:07   #8
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

37·157 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
ETA: My solar-inverter claims output line voltage of 236-241 Volts, so I think it's actually 240 and not 208?
208V comes from the voltage between two phases of 120V three phase.

sqrt(3)*120 = 207.85...

In the same way 240V three phase gives 415V between each two phases.
retina is online now   Reply With Quote
Old 2020-04-17, 20:52   #9
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

23·5·17 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
I thought most computer power supplies have a switch (or auto-detect?) to run 240 vs 120V; is it not valid that in this dreamworld I could just wire a regular-looking USA plug (well, six) to the 240V circuit, and run computer power supplies to said plugs without splitting the phase/voltage down to 120+ground?
Or you could be really tricky and wire each outlet of a duplex outlet to a different leg of the split phase, then combine them in a modified power cable. The power cable would have 2 plugs to the wall (one for each socket) with only live and ground pins (one plug's live pin would be rewired to the old neutral line in the cord). That way the outlets would still be 120V individually (and meet code), with the odd three headed power cable would deliver 240V to the PSU. (What? Yes I have considered doing this for increased efficiency, and no I haven't actually tried it )

Last fiddled with by sdbardwick on 2020-04-17 at 20:53
sdbardwick is offline   Reply With Quote
Old 2020-04-17, 20:58   #10
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

37×157 Posts
Default

Quote:
Originally Posted by sdbardwick View Post
Or you could be really tricky and wire each outlet of a duplex outlet to a different leg of the split phase, then combine them in a modified power cable. The power cable would have 2 plugs to the wall (one for each socket) with only live and ground pins (one plug's live pin would be rewired to the old neutral line in the cord). That way the outlets would still be 120V individually (and meet code), with the odd three headed power cable would deliver 240V to the PSU. (What? Yes I have considered doing this for increased efficiency, and no I haven't actually tried it )
You don't get double the voltage from using different phases. See my post above. You only get sqrt(3) times the single phase voltage.
retina is online now   Reply With Quote
Old 2020-04-17, 21:06   #11
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

23·32·61 Posts
Default

I looked up HP specs for the C7000 chassis, and the power specs include this:
"Two 3.05 m (10 ft) power cables with NEMA L15-30p connectors" with 200-208V "line to line" voltage. This makes sense- so I would be wiring a different 208V-specific outlet to the big EV-charger circuit.

Edit: Nope, not true. 3-phase power is 4 wires and that plug is 4-post, and my EV charger is 3 wires/3 posts.

But, it seems Retina is getting industrial 3-phase power confused with residential 2-phase 240V. In 3-phase (4 wire plug) circuits, voltage is indeed 120 * sqrt3. But at home, with the two live wires exactly out of phase with each other (3rd wire = ground), where does the sqrt3 term come from? Why, if they're perfectly out of phase, isn't it 120 - (-120) = 240?

Reference page, which I suppose could also be wrong: http://www.oempanels.com/what-does-s...ase-power-mean

Last fiddled with by VBCurtis on 2020-04-17 at 21:14
VBCurtis is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Has anyone tried linear algebra on a Threadripper yet? fivemack Hardware 3 2017-10-03 03:11

All times are UTC. The time now is 11:40.

Sun Oct 25 11:40:08 UTC 2020 up 45 days, 8:51, 0 users, load averages: 1.34, 1.68, 1.56

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.