mersenneforum.org  

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

Reply
 
Thread Tools
Old 2010-01-11, 21:27   #1
stars10250
 
stars10250's Avatar
 
Jul 2008
San Francisco, CA

3·67 Posts
Default Intel 6 core Gulftown memory saturation

I guess I'll fire the first salvo on gulftown and ask for comments on possible memory saturation running prime95 on all 6 cores. I don't fully understand all the bandwidth numbers, so I'm throwing this out to the group.

I do, however, have one practical data point from my i7 920. For the first few days operating my i7 I had a dead memory slot on the motherboard. Interestingly, after I switched out the motherboard and verified all three memory channels were working, I didn't really notice any improvement in performance. Some users commented on i7 bandwidth and seemed to indicate that 2 memory channels could just handle the bandwidth of the 4 cores. If this still holds, then it would appear that three memory channels should be able to handle the 6 cores.

I'd be interested in numbers for stock 1066 memory, as well as how bandwidth scales when overclocking (we gotta overclock this thing).
stars10250 is offline   Reply With Quote
Old 2010-01-11, 22:13   #2
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

7×647 Posts
Default

Isn't the i7 truly 4 cores with hyperthreading to give the "illusion" of 8 cores.

The general consensus on this forum seems to be that hyperthreading ranges between a small detriment and a small improvement depending on the Chip.
petrw1 is offline   Reply With Quote
Old 2010-01-11, 22:32   #3
stars10250
 
stars10250's Avatar
 
Jul 2008
San Francisco, CA

110010012 Posts
Default

The Gulftown, or Core i7 980X, is due to be released around March 16 2010 and it has 6 cores (12 threads). It shares some things in common with the earlier i7s, has some differences, but retains the same memory controler (as far as I can tell). So my question is, will prime95 over-saturate the memory bandwidth. From my estimates and experience with a failed memory slot, I believe it will be able to run all 6 cores without any slowdown...but I'd like to hear others weigh in on the numbers.

stars10250 is offline   Reply With Quote
Old 2010-01-12, 10:36   #4
garo
 
garo's Avatar
 
Aug 2002
Termonfeckin, IE

1010110001012 Posts
Default

Yes I should think so. I have a Core i5-750 which has dual channel memory and it is sufficient for four cores.
garo is offline   Reply With Quote
Old 2010-01-15, 12:51   #5
nucleon
 
nucleon's Avatar
 
Mar 2003
Melbourne

5·103 Posts
Default

I have a corei7 920 (not over clocked, memory @1066MHz) that runs a NAS VM (amoungst other things), and runs mprime in the host OS.

Doing smb/cifs file transfers from the NAS VM to the host OS yields 15MB/sec (pretty consistant) when mprime is full throttle. (running 3x LL-Ds and 1x P-1)

With mprime turned off the same transfer increases to 20-25Mb/sec. In my testing process, I've made sure not to time cached transfers. (As a side issue - I've seen Linux cache cifs/smb transfers which is pretty cool. I haven't seen WinXP do the same.)

In my past experience, mprime (or is that an OS task?) is really good at throttling cpu cycles, but not memory bandwidth.

To me, this is a great example of memory saturation. So I'm leaning towards memory saturation at 6 cores with default clock speeds (memory @ 1066MHz-ish). But, hey if testing proves otherwise, then fair enough, I'll accept defeat :)

The new 8x real cores coming will defiantely hit memory saturation with memory @ 1066MHz-ish.

6 real cores at 1333HHz - subject to testing, I'm leaning away from memory saturation. It may just be aok.
8 real cores at 1333HHz - subject to testing, but I'm leaning towards memory saturation.

-- Craig
nucleon is offline   Reply With Quote
Old 2010-01-15, 18:19   #6
R.D. Silverman
 
R.D. Silverman's Avatar
 
Nov 2003

22×5×373 Posts
Smile

Quote:
Originally Posted by nucleon View Post
I have a corei7 920 (not over clocked, memory @1066MHz) that runs a NAS VM (amoungst other things), and runs mprime in the host OS.

Doing smb/cifs file transfers from the NAS VM to the host OS yields 15MB/sec (pretty consistant) when mprime is full throttle. (running 3x LL-Ds and 1x P-1)

With mprime turned off the same transfer increases to 20-25Mb/sec. In my testing process, I've made sure not to time cached transfers. (As a side issue - I've seen Linux cache cifs/smb transfers which is pretty cool. I haven't seen WinXP do the same.)

In my past experience, mprime (or is that an OS task?) is really good at throttling cpu cycles, but not memory bandwidth.

To me, this is a great example of memory saturation. So I'm leaning towards memory saturation at 6 cores with default clock speeds (memory @ 1066MHz-ish). But, hey if testing proves otherwise, then fair enough, I'll accept defeat :)

The new 8x real cores coming will defiantely hit memory saturation with memory @ 1066MHz-ish.

6 real cores at 1333HHz - subject to testing, I'm leaning away from memory saturation. It may just be aok.
8 real cores at 1333HHz - subject to testing, but I'm leaning towards memory saturation.

-- Craig
We need cache!! cache, cashe, and more cache!!! I want 8 Gbyte of
L2 cache!!!!
R.D. Silverman is offline   Reply With Quote
Old 2010-01-15, 18:49   #7
stars10250
 
stars10250's Avatar
 
Jul 2008
San Francisco, CA

3·67 Posts
Default

Quote:
Originally Posted by R.D. Silverman View Post
We need cache!! cache, cashe, and more cache!!! I want 8 Gbyte of
L2 cache!!!!
I don't think you're going to get that any time soon. For L2, Gulftown is listed as 6 x 256 KB...so 256 KB per core. I know you'd like to fit everything in the cache, but you still have to admit these cpus are really taking performance to new levels. Can you estimate how much faster a core could do an LL iteration if it didn't have to communicate with main memory? That would be interesting.
stars10250 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
48-core chip - Intel hj47 Hardware 6 2009-12-20 13:59
Intel Quad Core optimization joblack Hardware 1 2009-06-11 02:42
Optimising work for Intel Core 2 Duo or Quad Core S485122 Software 0 2007-05-13 09:15
Intel's 80 core beast... TehPenguin Hardware 10 2006-11-27 11:13
Intel Core Duo processor drew Hardware 5 2006-05-29 07:00

All times are UTC. The time now is 14:11.

Fri Jan 15 14:11:18 UTC 2021 up 43 days, 10:22, 0 users, load averages: 1.62, 1.79, 1.93

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.