I5 failing Prime95 at Stock

shadowsofthesun

Distinguished
Oct 17, 2009
8
0
18,510
For my first PC build, I decided to go with a core i5. I noticed early on that Everest's stress test would fail at stock voltages and clocks (usually 30 seconds - 10 minutes). Prime 95 also would give me errors at stock settings on occasion (usuallly 5 minutes - 30 minutes). I decided to fiddle around with overclocking a bit, and got a 3.2GHz clock that seemed stable. Throughout, I never exceeded the 1.4v Intel spec; temps also stayed cool.

I've since returned to standard settings across the board, but I'm still getting failures in Prime 95 with stock settings, which greatly concerns me.

- Corsair 620HX
- Gigabyte p55-UD4P motherboard
- Core i5
- Cooler Master Hyper212 heatsink
- Mushkin XP 1600 CAS7 1.9v 2x2GB (ran at Mushkin's recommended 1.9v for awhile until I realized this was bad. Returned to 1.5v since then. Problems happening throughout. Might RMA for 1600 CAS7 1.65v)
- Asus GeForce9600 GSO

I just ran Prime95 again with C states and EIST off. Some cores got errors (rounding was x, expecting y) and another failed because it reached 100 warnings (Illegal sumouts). Peak core temps only reaching between 39 and 44. I'll try running memtest overnight and report back with results.

Any suggestions on how to proceed? Not sure how much I can work on this this weekend, but I will try to stay responsive. Thanks!

*edit: added screen capture:
4018810652_f291b02dc6_b.jpg

And Prime95 just had to close on me due to an error. :(
 
Solution
If this held stable at 800mhz, then I'd say its the CPU and its on-die controller. If it couldn't handle 1333 at 1.5v Cas 9 (JDEC 3), it has degraded performance. If you want to keep the ram you have, realize that the limit is JDEC 3 or lower (based on CPU-Zs numbers.) If you are going to exchange the ram for something rated for the i5/i7, I'd do that first just to be sure the CPU memory controller is the problem.

It is possible this is a mobo issue. But the CPU/Ram issue should be cleared up first as its a known issue.

skora

Distinguished
Nov 2, 2008
1,498
0
19,460
If you were running 1.9vs through the memory controller, you may have toasted it already. Since its built into the chip, there's not much play at all for voltages like there would be with the memory controller on the old NB. 1.7 is the highest I'd go, but you blew right past that.

Running the ram at 1.5v, you'll have to underclock the ram alot to get the ram stable. Set it to 1.65v and see what you can do with its speed. Taking the hit for restocking on the ram is probably a good idea, because if you have to replace the CPU, you'll want the right ram anyway.

If that doesn't work, keep the OC, your warranty is void already or try to RMA it.
 

shadowsofthesun

Distinguished
Oct 17, 2009
8
0
18,510
Thanks for the advice, Skora.

I ran memtest overnight. 4 passes with 0 errors the RAM was running at 1333 at 1.5v. However, my system locked up again while I was writing this.

Is there any way to verify that the ram voltage is what killed the CPU? Are there any other tests I should do before trying to return on warranty? For example, could it be the motherboard causing trouble?

If I try to return the processor under warranty, I'm assuming I should go through Microcenter where I purchased it instead of Intel themselves, correct?
 

skora

Distinguished
Nov 2, 2008
1,498
0
19,460
For the sake of testing, dial back the ram to DDR3 800 and very loose timings. Set the voltage to 1.65, then stress it again. You might not have damaged anything and you'd be lucky, but 1333mhz may still be too much for the ram at 1.5v. Back it off all the way you can. If its stable with really slow ram, go buy new and then test with the correct stuff. If you still have problems, then look at the CPU.
 

skora

Distinguished
Nov 2, 2008
1,498
0
19,460
I was thinking, lower the speed/timings, and then prime it, not another round of memtest. While the ram is probably fine, its overcorrecting to isolate the CPU. Its like using a i7 975 system to test a 9800GT GPU. Its unrealistic, but assures the GPU will be the bottleneck in benchmarks, not the CPU. 1 variable is better than 2.

Either way, you should buy ram with the correct voltage limits. If MC will exchange the CPU, thats good too.
 

shadowsofthesun

Distinguished
Oct 17, 2009
8
0
18,510
Right now, I'm running prime95, and it seems to be doing fine.
4021791708_c64712a393_o.png


I have all advanced options turned off in on the CPU, such as EIST, sleep states, turbo mode, etc. The ram is set to 6x multiplyer, running at 800MHz CAS 6 1.5v. I feel like it's the things flagged as "advanced CPU features" in the BIOS that are causing the problem. Could this be a MB issue? I will attempt to isolate tomorrow before I try to return anything.
 

skora

Distinguished
Nov 2, 2008
1,498
0
19,460
If this held stable at 800mhz, then I'd say its the CPU and its on-die controller. If it couldn't handle 1333 at 1.5v Cas 9 (JDEC 3), it has degraded performance. If you want to keep the ram you have, realize that the limit is JDEC 3 or lower (based on CPU-Zs numbers.) If you are going to exchange the ram for something rated for the i5/i7, I'd do that first just to be sure the CPU memory controller is the problem.

It is possible this is a mobo issue. But the CPU/Ram issue should be cleared up first as its a known issue.
 
Solution

shadowsofthesun

Distinguished
Oct 17, 2009
8
0
18,510
Well, I tried to isolate the memory. I turned it back up to 1333MHz and it failed Prime and froze the machine. however, it works fine at 800 and 1066. I'm currently running a Prime95 test with the memory at 1333MHz, but at 1.64v. It seems to be doing ok. The plan is to get the right RAM this time around, and probably try to exchange the CPU as well.

Thanks for all the help! It's been a good learning experience and I definitely appreciate your assistance.
 

skora

Distinguished
Nov 2, 2008
1,498
0
19,460
Well, if this test ends up being stable at 1333, you're CPU may be fine and just the ram can't cut it 1333mhz speeds with only 1.5v. You're not going to get any noticeable performance increase with 1600mhz, so if its stable, you don't HAVE to do anything. You can try to OC the CPU and then make your decision. If you do keep this ram, just make sure you remember these settings anytime you wipe the bios and have to setup everything again.

Here's an article on the effects on ram speed for the i7s, the i5 will have similar results:
http://www.tomshardware.com/reviews/memory-scaling-i7,2325.html
 

shadowsofthesun

Distinguished
Oct 17, 2009
8
0
18,510
Well, I currently come to you from a i7 860 processor. The tests I was running still weren't stable at 1333 with 1.65v, so I exchanged the i5 for a replacement. I was even able to talk the service rep at Microcenter to let me upgrade to the the i7.

I've currently run one pass of Prime95, and it didn't have any errors with the Mushkin ram at 1333MHz 1.5v. I'll test more later, but I'm glad I took the CPU in. I'm also going to replace the memory with something correctly spec'd.

Thanks for the link. It's definitely surprising how little difference RAM speeds currently make on performance. Who knows, though, it might change in the future.
 

TRENDING THREADS