mersenneforum.org  

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

Reply
 
Thread Tools
Old 2013-10-24, 11:05   #408
VictordeHolland
 
VictordeHolland's Avatar
 
"Victor de Hollander"
Aug 2011
the Netherlands

22308 Posts
Default

According to guys on the tomshardware forum, non K CPU can run XMP profiles, but possibly not all the way till DDR3-2400. But your BIOS/motherboard need to support it and you might have to set the vRAM yourself.
http://www.tomshardware.com/answers/...file-4770.html
VictordeHolland is offline   Reply With Quote
Old 2013-10-24, 16:33   #409
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

172710 Posts
Default

It also comes down to the fact that a K-series chip is a binned version of the non-k series. If I understand, the i7-3770k, i7-3770 and i7-3770s or whatever the last one is are all made from identical processes. The finest and highest performing chips get a boost to their basic frequency and are binned as a higher performing chip. Those chips that didn't make it through with flying colours but still function fine are underclocked, undervolted and sold cheaper as a budget / efficiency-minded chip.

If that is the case, then the non-K series doesn't meet the specs to be a K-series, in whatever way, which might be reflected in the memory controller's strength.
TheMawn is offline   Reply With Quote
Old 2013-10-24, 23:01   #410
ewmayer
2ω=0
 
ewmayer's Avatar
 
Sep 2002
Rep├║blica de California

22·3·7·139 Posts
Default

George, having re-perused this thread starting from early June, when you took delivery of your Haswell, I notice that your early start-CPU-at-stock-and-slowly-crank-up-the-clocking stability tests were all using your non-FMA v27 code. Do you recall having any stability issues with that above 4GHz, or did it run 24/7 stably for significantly longer than your current average-time-between-system-crashes?

Also, unless you did some with-FMA redos of the above starting-at-stock stability tests which I missed, it seems you have only ever run the v28 code with your CPU OCed - is that correct, or have you in fact run v28 @3.4GHz to test stability-at-stock-speed?
ewmayer is offline   Reply With Quote
Old 2013-10-25, 00:20   #411
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

23·312 Posts
Default

Ernst, you are basically correct. I had the system running at 4.1 GHz, DDR3-2400 on 27.9. I thought it was stable, but maybe I was just lucky in that the once-every-three-day-or-so reboot just never happened. For example, that first box has been up for the last seven days but I wouldn't be surprised if it rebooted tonight.

If it was truly stable under 27.9, two things have happened since then to introduce instability. 1) I started running 28.1 which places more stress on the CPU and memory. I downclocked 100 MHz hoping that would solve the instability. 2) I installed a GTX 570 card.

The second system has been unstable since the beginning running 28.1. It does not have a GPU. Even in its current unstable state, it has returned many successful double-checks.

I've been trying the increase voltage route (which may be why I haven't rebooted in 7 days -- or not). If the latest round of voltage increases doesn't work, I'll try a different strategy. Start at stock CPU speed and DDR3-2400 and increase CPU speed until unstable and then back off. The problem with a once-every-3-day reboot problem is it can take weeks to achieve the optimal settings.

I'll figure it out eventually. I've never had this much difficulty in OCing. In part because there are so many different voltage tweaks to try and in part because I've never pushed memory this hard. I'll be bummed if I can't get DDR3-2400 stable.
Prime95 is offline   Reply With Quote
Old 2013-10-25, 02:45   #412
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

32778 Posts
Default

George:

Could you list off your motherboard and power supply make and model? The GTX 570 is a pretty big power sink. It could be that your power supply can't take the load (if its rated power is low or it isn't a very good part) or it overheats from the sustained load.

Alternatively, your motherboard may not be able to fully handle the CPU. The PCI-E controller might also be having a hard time keeping up. It really depends on the motherboard. If it's some ASUS board with 3x PCI-E 3.0 x16 lanes and nice big VRM heatsinks, you might be fine. If it's some Biostar board that has only the one PCI-E slot and cheap generic CPU power delivery, a number of things could be happening.

Finally, your RAM could actually be giving you grief. I don't know that RAM errors would cause reboots. I could see the PSU overheating or missing a beat. I could see the VRM overheating. I could see the VRM briefly supplying dangerous voltage to the CPU. I could see the PCI-E lanes shitting the bed.


I would look into memtest86+ 4.20 just to initially give you some confidence in your RAM. I've never gotten a RAM setting past it to then proceed to failing the Prime95 stress test. My current settings are fine (all the DC's I've done have checked out, never failed a stress test, etc) and any tweak immediately makes me fail memtest86+ 4.20.


I was also wondering if the new instructions are giving your CPU grief. Maybe try taking out the GPU for a bit and seeing if that does anything.

Check that the VRMs aren't too hot. Maybe try lowering memory speed regardless of memtest results. See if you can tell what temperature the PSU is at. Check that your memory isn't too hot (chances are it's fine).
TheMawn is offline   Reply With Quote
Old 2013-10-25, 03:02   #413
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

23·312 Posts
Default

Mawn,

I think the PSes can handle the load. The RAM is not too hot to hold for 15 seconds or so. I haven't tried taking the temp of motherboard's VRMs. The mobos are definitely the cheapest parts in each system.

Haswell #1 with GTX570:

Mobo: http://www.newegg.com/Product/Produc...82E16813130695
PS: http://www.newegg.com/Product/Produc...82E16817182083
RAM: http://www.newegg.com/Product/Produc...82E16820231587

Haswell #2 no GPU:

Mobo: http://www.newegg.com/Product/Produc...82E16813128599
PS: http://www.newegg.com/Product/Produc...82E16817182264
RAM: http://www.newegg.com/Product/Produc...82E16820231585

Last fiddled with by Prime95 on 2013-10-25 at 03:05
Prime95 is offline   Reply With Quote
Old 2013-10-25, 03:52   #414
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

23·211 Posts
Default Haswell G2030 new to the family farm.

My son just installed it.
I've asked for Benchmarks.... stay tuned.
petrw1 is offline   Reply With Quote
Old 2013-10-25, 04:04   #415
db597
 
db597's Avatar
 
Jan 2003

7·29 Posts
Default

I suspect some memory modules are not fully stable at their rating. Under 99.9% of conditions they're fine. But P95 with AVX/FMA optimisations make us fall into the 0.1% of the users that exceed the standards the memory was validated for.

Some time ago I experienced this on a set of 1600MHz DDR3 sticks. It worked fine at 1333MHz, but would hang twice a day at 1600MHz. At first I thought it could be a CPU issue (added a lot of VCore, VCSA, VCCIO without success). However, when I replaced the memory with newer sticks at 2000MHz, and it continued to be stable - at the same CPU speed throughout.

To be clear, those 1600MHz sticks had been fine earlier running at 1600MHz on v26.x / early 27.x and never hanged on any other applications. If not for the new optimisations, I'd never have noticed. Also, those modules were already 2Y old, so perhaps running 24/7 had also "aged" them.
db597 is offline   Reply With Quote
Old 2013-10-25, 06:09   #416
TheMawn
 
TheMawn's Avatar
 
May 2013
East. Always East.

110101111112 Posts
Default

I can stand behind the memory, myself. G Skill has never disappointed me. They've my go-to people for RAM. I'm 0 for 1 with Corsair and 4 for 4 with G Skill. I'd say Trident is better than Ripjaws but only because it's their latest iteration. My modules are the 8 GB version of the ones George has and they're running happily at rated specs (but ---- me if they'll go one tiny tick higher).

My personal diagnosis on this one is the motherboard. I don't know if they're as capable of handling the constant load as they boast. You've got a certain level of stability for a certain amount of time, but you're probably stable for as long as they run their own tests.

I've attached a .png of your two motherboards as well as an ASUS one. ASUS does a huge amount of validation testing on their motherboards, like getting a certain frequency stable on the CPU and memory. Back with Z77, they got up to 4.6 GHz on an i7-3770k and up to 2800 MHz on memory on all of their boards. They probably do the same for Z87.

The VRM area is disturbingly empty in your two boards. ASUS have a whole other set of VRM goodies and another heatsink. I have this funny feeling your motherboard is having issues powering the CPU. I'm sure you've got your CPU supplemental power plugged in? See if you can't give the capacitors and inductors around the CPU area a quick tap to see if they're cooking themselves to death.
Attached Thumbnails
Click image for larger version

Name:	VRM.jpg
Views:	108
Size:	371.6 KB
ID:	10386  

Last fiddled with by TheMawn on 2013-10-25 at 06:10
TheMawn is offline   Reply With Quote
Old 2013-10-25, 09:49   #417
TheJudger
 
TheJudger's Avatar
 
"Oliver"
Mar 2005
Germany

100010101112 Posts
Default

Quote:
Originally Posted by TheMawn View Post
I can stand behind the memory, myself. G Skill has never disappointed me. They've my go-to people for RAM. I'm 0 for 1 with Corsair and 4 for 4 with G Skill. I'd say Trident is better than Ripjaws but only because it's their latest iteration.
Well, "Trident(X)" brand modules have a high chance to be based on Hynix "MFR" chips. They are very good for raw clock rate but they have crappy performance, especially the 4GiB modules. AFAIK one issue is that they are single rank modules. With one rank per memory channel you have always the additional read-after-write-on-same-rank penalty. That's why dualrank modules are prefered. Quadrank is possible two but the downside is that with increasing number of ranks the electrical charactaristics get worse, usually quadranked modules have lower clockrate than dualrank.
http://www.techpowerup.com/reviews/G...-8GTXDG/6.html

Oliver
TheJudger is offline   Reply With Quote
Old 2013-10-25, 10:39   #418
db597
 
db597's Avatar
 
Jan 2003

CB16 Posts
Default

Quote:
Originally Posted by TheJudger View Post
Not sure how that review concluded a 9.0 rating for the product. The benchmarks look to me like the 2933MHz sticks was consistently lagging the group - on average ~2nd slowest, although it had the highest speed rating.
db597 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Haswell-E Prelim. Benchmark sdbardwick Hardware 37 2015-02-10 18:49
Prime95 and Haswell Pleco Information & Answers 22 2014-07-13 16:03
Haswell Rig Mini-Geek Hardware 64 2014-05-27 13:22
Prime95 version 27.1 early preview, not-even-close-to-beta release Prime95 Software 126 2012-02-09 16:17
Missing mouse-over preview text retina Forum Feedback 1 2011-09-12 15:32

All times are UTC. The time now is 08:22.


Sun Dec 5 08:22:21 UTC 2021 up 135 days, 2:51, 0 users, load averages: 2.02, 1.69, 1.59

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.