mersenneforum.org  

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

Reply
 
Thread Tools
Old 2007-12-24, 16:27   #45
ADBjester
 
Aug 2002

2×3×5 Posts
Default

Quote:
Originally Posted by db597 View Post
Given the memory bottleneck, for the similar cost, going for a DDR3 @ 1600MHz might have a bigger difference than watercooling and super overclocking your rig. Not to mention less worries about stability, topping up the water periodically, electricity bills etc.
"For the similar cost?" Where do *you* buy your DDR3?

The cost in terms of moving to the X34 chipset is only $260... and it was time for a coolant replacement anyway.

To move to DDR3 I'd *still* need to buy a new motherboard (the current one does not support DDR3), and I'd have to replace the RAM as well. The best price I can find on 4 GB of DDR3 PC3-12800 (1GBx4) is $1038. (I wouldn't even consider moving to DDR3 and then hobbling myself with 1066 Mhz RAM).

I doubt I could save that much in electricity over the course of a year. I do eventually intend to go DDR3... after it comes down in price. I'll probably do so when I build my first Nehalem system in 2009.

Jeff

Last fiddled with by WraithX on 2016-02-15 at 05:10
ADBjester is offline   Reply With Quote
Old 2007-12-25, 03:19   #46
db597
 
db597's Avatar
 
Jan 2003

7·29 Posts
Default

Quote:
Originally Posted by ADBjester View Post
The cost in terms of moving to the X34 chipset is only $260... and it was time for a coolant replacement anyway.
Ah... didn't notice that you already have your water cooling kit - this in itself ain't cheap. Was counting this, and the fact that you can get a 35 series motherboard instead of a X38, plus lower electricity bills.
db597 is offline   Reply With Quote
Old 2007-12-25, 03:56   #47
IronBits
I ♥ BOINC!
 
IronBits's Avatar
 
Oct 2002
Glendale, AZ. (USA)

3·7·53 Posts
Default

Quote:
Originally Posted by ADBjester View Post
ordered an ASUS Maximus Formula Mobo based on the Intel X38 chipset, which is well known for its overclockability.
Jeff
It is absolutely the best board I've ever owned. Rock solid, impressive features. Designed by gamers for gamers, not a cheap component anywhere.
With a Q6600, easily goes to 3GHz on air.
IronBits is offline   Reply With Quote
Old 2007-12-27, 08:35   #48
zs6nw
 
Dec 2007

1C16 Posts
Default

Quote:
Originally Posted by sgrupp View Post
1) Memory contention when all 4 cores are running P95. Are iteration times still OK?
No. When I load up 4 cores, I find the per-iteration time roughly double compared to when only two cores are working and 2 idling.

Quote:
2) HEAT. Can an air cooling solution handle all 4 cores doing LL testing?
Easily. A 2x E5430 system (8 cores + disk + 4M RAM) runs around 200W of power.

Quote:
3) Anyone running a water cooled overclocked rig like Dell's latest?
Water cooling seems pretty unneeded in light of only 200W for the entire computer system. Granted, I have low-power on-board video.

Quote:
4) Power. Any notion of what power consumption is on a QX6700 running 4 LL tests?
Entire system (2x E5430 quadcore, 4M 667 Ram, 160G Disk, SuperMicro Motherboard) runs around 200W.

My only peeve is the memory clash - seems like two cores working are already close to the bus bandwidth limit. That means the $/P90_CPU_year is much higher than anticipated. Perhaps some clued-in guru can confirm this...
zs6nw is offline   Reply With Quote
Old 2007-12-28, 21:33   #49
ADBjester
 
Aug 2002

368 Posts
Default

Quote:
Originally Posted by IronBits View Post
It is absolutely the best board I've ever owned. Rock solid, impressive features. Designed by gamers for gamers, not a cheap component anywhere.
With a Q6600, easily goes to 3GHz on air.
Sadly, due to an incompetent vendor I will not have the parts to reconstruct the system this weekend. (I was sent the Maximus Formula, rather than the MF Special Edition -- meaning what I got has no watercooling hookups on the north/south bridges). It's going back and I'm waiting for the correct board.

Jeff
ADBjester is offline   Reply With Quote
Old 2008-01-03, 06:59   #50
ADBjester
 
Aug 2002

2×3×5 Posts
Default

Quote:
Originally Posted by ADBjester View Post
Sadly, due to an incompetent vendor I will not have the parts to reconstruct the system this weekend. (I was sent the Maximus Formula, rather than the MF Special Edition -- meaning what I got has no watercooling hookups on the north/south bridges). It's going back and I'm waiting for the correct board.

Jeff
Crosspost of what I just posted to the Mersenne mailing list follows. Thanks to all here who helped figure this out!

Summary of last week's posts:

My uber-juiced, liquid-cooled, overclocked (to 3.467 Ghz) quad core was turning in iteration times about double what the benchmarks page suggested was reasonable. For an exponent in the 41.5 million range (2560K FFT):

Core 0: 0.133 sec
Core 1: 0.110 sec
Core 2: 0.138 sec
Core 3: 0.109 sec

Note that "complementary" cores on the same die were not turning in identical times, i.e. Cores 0 and 2 ran slower, while Cores 1 and 3 ran faster.

After being steered to the MersenneForum and reading some other posts there, it became apparent that some thought that my use of the nVidia 680i chipset might be causing a memory bandwidth issue, and/or L1 cache/thrashing issues. I decided to test that theory, and ordered an Intel X38 motherboard from ASUS.

----

Results:

After a minor crisis in upgrading, I have the ASUS in operation. It is not yet overclocked, but it is the same system with the exception of the motherboard, now running stock operating speeds. The numbers are startling.

Core 0: 0.071 sec
Core 1: 0.071 sec
Core 2: 0.069 sec
Core 3: 0.072 sec

Now remember, this is without benefit of the 30% overclock I was enjoying... and I'm now seeing iteration times nearly twice as fast across the board.

I'm also experiencing 40-50% inprovement in framerates in bleeding edge video games like Crysis. (Wow, I thought I needed SLI to get this sort of improvement in framerate!)

There is absolutely no question about it. Intel knows how to manage the memory architecture on their own chips far better than nVidia does.

If you're planning on running P95 on a quad core (or if you are just a purist who wants the fastest system when pushed to its absolutle limit), stick with the Intel chipset. nVidia makes GREAT graphics cards... but apparently isn't so hot at motherboards.

Jeff Woods
Reading, PA
ADBjester is offline   Reply With Quote
Old 2008-01-03, 15:12   #51
ADBjester
 
Aug 2002

2·3·5 Posts
Default

Quote:
Originally Posted by ADBjester View Post
The numbers are startling.

Core 0: 0.071 sec
Core 1: 0.071 sec
Core 2: 0.069 sec
Core 3: 0.072 sec

Now remember, this is without benefit of the 30% overclock I was enjoying... and I'm now seeing iteration times nearly twice as fast across the board.
Here's another apples to apples comparison that I just posted about this elsewhere:

Here's how the 30%-overclocked 680i system responded as the number of cores stressed to the max was increased:

Single core stressed: 0.050 seconds
Two cores stressed: 0.063 seconds
Three cores stressed: 0.074 seconds
Four cores stressed: 0.124 seconds

The fourth core stressed results in a 148% increase in calculation time over a single core.

The X38 at standard clocking, by contrast, does this:

Single core stressed: 0.056 seconds
Two cores stressed: 0.063 seconds
Three cores stressed: 0.065 seconds
Four cores stressed: 0.071 seconds

That's an increase of only 30% from one core to four cores, which is much more within expectations. I expect once I get a stable overclock going, that those will drop a bit.

Again, these motherboards were:

EVGA 122-CK-NF68-AR 680i
ASUS Maximus Formula Special Edition X38

---

Other pertinent parts in the system include:

Vista Ultimate x64-bit
4 GB (1GBx4) Corsair Dominator PC2-8500-C5D RAM
QX6700 CPU

Jeff
ADBjester is offline   Reply With Quote
Old 2008-01-04, 16:32   #52
ADBjester
 
Aug 2002

1E16 Posts
Default

And finally, some numbers from the initial stable overclock on the ASUS X38.

A 13X multiplier turned out to be unstable on the X38. I'm now at 12X, 3.2 Ghz, and a standard FSB and RAM of 1066 Mhz. (Recall that I could not get the RAM to go the full linked 1066 Mhz on the nVidia.) That survived 12 hours of Torture testing last night.

All four cores crank merrily along at 0.0635. Benchmarking a single core gives a best time of 0.046, meaning running four cores on this machine at a stable 20% overclock loses 38% of its performance to cache thrashing.

I can live with that (especially when compared to 148% loss on the 680i!)... at least until Nehalem gives us dedicated, onboard memory controllers per core and QuickPath technology instead of the shared front side bus. THOSE systems should be pretty sweet. (Anyone plan to try a Skulltrail on a QX9650 in the meantime?)

The final conclusion: at least for the particular brands/models of motherboards I tried, the 680i chipset is completely inferior to the X38 when pushed to the outer limits of memory bandwidth requirements by Prime95.

I'll continue to try to squeeze out some overclock cycles going forward (I suspect/hope I can add more memory bandwidth by increasing the FSB and memory voltage here, something that was unstable on the 680i), but for now, book closed. Intel chipsets for me going forward.

Jeff Woods
ADBjester is offline   Reply With Quote
Old 2008-01-16, 02:19   #53
James Heinrich
 
James Heinrich's Avatar
 
"James Heinrich"
May 2004
ex-Northern Ontario

1101101100002 Posts
Default

For what it's worth, my Q6600 (stock 2.4GHz) overclocked to 3.49GHz on a P35 board (Gigabyte P35C-DS3R) show numbers like this:
1 thread: 0.045s
2 thread: 0.048s
3 thread: 0.054s
4 thread: 0.063s

RAM is clocked at 775MHz. System is 95% stable at 3.6GHz, but not Prime95-stable :(
James Heinrich is offline   Reply With Quote
Old 2008-01-25, 07:02   #54
nngs
 
nngs's Avatar
 
Jun 2004

22·3·5 Posts
Default

C2Q power consumption for LL tests:

Dell Inspiron530
CPU: C2Q 2.4GHz
1 GB DDR2 SDRAM @ 667MHz (seems dual channel, but not sure)
Integrated Video/Audio
SATA 250GB
16X DVD+/-RW
19in1 card reader

OS: VISTA Home Basic

Quote:
CPU0 CPU1 CPU2 CPU3 Power
X X X X 62 W
V X X X 104 W
V X V X 129 W
V V V X 139 W
V V V V 144 W
My another Dell machine, CPU P4 2.4G, consume about 65W at idle, and ~115W at 100% LL test
nngs is offline   Reply With Quote
Old 2008-01-25, 22:01   #55
ADBjester
 
Aug 2002

2×3×5 Posts
Default

Quote:
Originally Posted by dsouza123 View Post
Maybe the newer 680i motherboards will work better.
If you followed my earlier saga... it got worse. MUCH worse. And I found out the answer to the above the hard way. Newer 680i (and 780i) boards are no better than the older 68oi boards, and I infer that all Intel X38 chipsets do fine.

Short version:

Iterations under EVGA 680i suck, at 0.138 per core on a quad core QX6700. RAM Bandwidth saturation suspected.

I swap for an ASUS Maxiumus Formula with X38 chipset. It turns in iteration times of about 0.070 when all four cores crank 100%. This is very close to expectation, and nearly double the performance of the 680i.

That's where I left off last time. Now we rejoin our story, brought up to today.

Four days later, the Maximus died overnight. Poof. No boot. No POST. No more in the channel for replacement by the vendor. Only option: RMA for a 10 (business) day repair. Scrambling for parts (it's a work computer that I *have* to have in operation), I replace it with what I can get my hands on -- an ASUS P5N-T Deluxe. This is a 780i chipset, very new.

At this time, since I have the water loop broken down yet again, I replace the aluminum radiator with a copper one to stop an electrolysis problem, and replace the pump with one pushing 2.4 times as much water -- I was near redlining at 64C-68C before.

The P5N-T deluxe has a SEVERE problem that they have not managed to isolate in the lab. If you're running 4 GB RAM rated for 1066 Mhz (PC2-8500 or better, 4x1GB sticks), the board hard-locks several times a day. <sigh> Not suitable... but for the brief time I have this board operational, I find that the 780i chipset performs about the same as the 680i -- 0.138. Different manufacturers' nVidia chipsets did about as poorly.

So, yet again I buy another board. This time, an Abit IX38. This board thus far appears stable (knock wood) and is turning in iteration times of 0.071 or so at standard clocks. Thus, we have X38 boards from two different vendors turning in iteration times near 0.070 and nVidia boards (two different revs) from two different vendors turning in times of almost double that.

nVidia sucks for P95. Stick with Intel chipsets.

ASUS sucks for performance systems. Three dead boards in 2 weeks here. Stick with other vendors for Intel chipsets.

Jeff Woods
ADBjester is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Dual Core to Quad Core Upgrade Rodrigo Hardware 6 2010-11-29 18:48
exclude single core from quad core cpu for gimps jippie Information & Answers 7 2009-12-14 22:04
Quad Core Questions... TomYosho Information & Answers 2 2009-09-14 13:01
Quad Core R.D. Silverman Hardware 76 2007-11-19 21:57
Optimising work for Intel Core 2 Duo or Quad Core S485122 Software 0 2007-05-13 09:15

All times are UTC. The time now is 04:16.


Sat Oct 23 04:16:12 UTC 2021 up 91 days, 22:45, 0 users, load averages: 1.61, 1.68, 1.54

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.