New prime95 for Haswell

P

prime95

n00b
Joined
Jun 17, 2005
Messages
5
I have a new version of prime95 for Haswell processors for you to try. It uses Haswell's new fused-multiply-add instructions for a little more CPU stress. On my system running the small FFT torture test, the new version is about 8C hotter (sorry about that!) than the last version. For me, I had to reduce my overclock by 100MHz to achieve stability with the new version.

To download, use this link (Windows 64-bit version): http://www.sendspace.com/file/mtfcor

This is a pre-release version, let me know if you find any bugs.
 
I tried this on my 4770s and I got an error that stopped a thread (not heat related it would seem) almost immediately.

I have since sent that processor back to Intel for RMA as it ran unreasonably hot for a 65 Watt rated chip. It would run Prime95 27.9 without errors or bluescreens but it would constantly throttle itself down to 2.5Ghz or lower when p95 got mean to it. That was with turboboost off. Perhaps running my Gskill Ares 2133Mhz Ram at full blast wasn't helping. It's 1.6v ram and I have 4X4GB modules installed on a AsRock Extreme4.

I'm afraid to try this version on my new 4770K. It runs Prime95 27.9 without errors through the first 14 tests (when it goes to the larger sets after first set of blend tests passes, the K goes on up into the 80s and I stopped it) at 4.2Ghz and 1.25Vcore. So many BIOS voltage options on these chips. I ended up with HIGH BCLK and Fixed voltage for both vcore and cache. Vcore and cache should be set the same, right?

Anyway, on topic, I'm not sure we needed more heat from Prime95 on Haswells... Is there any way to bring the program's heat generation down some without compromising speed if we want to actually run it fast instead of just torture testing for heat?
 
Bluesun311 said:
Vcore and cache should be set the same, right?

Anyway, on topic, I'm not sure we needed more heat from Prime95 on Haswells... Is there any way to bring the program's heat generation down some without compromising speed if we want to actually run it fast instead of just torture testing for heat?
Click to expand...

No, VCore and cache should NOT be the same. Cache should be set at 1.1-1.2v while VCore should be dependent on the clock you're trying to get.

And the heat generation comes from using the AVX instructions set if I'm not mistaken.
 
Nubbleet said:
No, VCore and cache should NOT be the same. Cache should be set at 1.1-1.2v while VCore should be dependent on the clock you're trying to get.

And the heat generation comes from using the AVX instructions set if I'm not mistaken.
Click to expand...
Ugh, I'm the real thing: "noob". Thanks much for response. I allowed my BIOS to load a preset for 4.2Ghz and it fixed my cache voltage at 1.15. It changed a couple other things I had not looked at but I wasn't way way off. I haven't tried their settings on Prime95 yet, but the overall effect atm is that the chip is running cooler at idle.

Well. Originally I had a bluescreen on Prime95 29.7 about 10 minutes in with 4.2Ghz at stock settings, so I really should have read a guide before I started messing with the cache voltage but I went to 1.25 only and for a short time with zero errors or bluescreens the half hour I ran p95 so hopefully no damage.

About the OP version of p95 that runs hotter than 29.7: is anyone else having instant errors with it like I did on my 4770S?
I bought it thinking it would run cooler than the K... and it was overheating in AIDA64 cpu only test and throttling to 2.4Ghz. Is it actually possible to get a bad chip like that? I almost felt guilty RMAing it. But I bought the 4770K first to test and so I would have a backup. I won't be pissed really if they just send it right back saying nothing wrong. It was only on the benchmarking side that I noticed excessive heat, I wouldn't have really had any reference for the temps I was seeing in applications and games. This thread is what brought me to this forum. If you are searching google for haswell and prime95 this thread pops up.

I got the K to test the problem, essentially, assuming the 2 chips would be almost identical with different factory settings. I don't remember where I got that idea and I don't know if it is supposed to be true or not. Certainly the chip names suggest that. The 4770K I received idles at about 15-17 degrees Celsius cooler than the S at 1.1Ghz higher clock. I wanted to know if I all of a sudden just now learned how to properly install a hyper 212 evo in a well ventilated case or if there's something wrong with the S chip.
 
I've ran this on my own 4770k and the new 28.1 does indeed run hotter than 27.9. No issues for me, everything works as expected.
 
May run hotter is something of an under statement. At stock clocks my 4770k hit 85 degrees.
 
How long does it take to get through the first test with this new 28.1? My setup can handle the temp but it seemed to be taking a really really long time to finish first test with 8 instances going... 27.9 It's like 1 test every 1-2 minutes.
 
Bluesun311 said:
How long does it take to get through the first test with this new 28.1? My setup can handle the temp but it seemed to be taking a really really long time to finish first test with 8 instances going... 27.9 It's like 1 test every 1-2 minutes.
Click to expand...

You need to pay careful attention to the details of the tests its running. For me with 27.9 FFT runs 9000 iterations using AVX while 28.1 runs 36,000 iterations (4x as many) using FMA3.
 
jasonorland88 said:
Does anyone have a rumor or source of what the TJMax is on the 4770K?
Click to expand...

Intel set TJ Max to 100C on the 4th Gen CPUs that I have seen. This value is written to every core of every Core i CPU so you can run Core Temp and see the correct value there. You can also run RealTemp and have a look in the Settings or Options window.

Core Temp
http://www.alcpu.com/CoreTemp/

Click on More Downloads... if you want to get the stand alone version of Core Temp without the installer.

RealTemp T|I Edition
http://www.mediafire.com/download/7h6578x6dh7hc8x/RealTemp_TI.zip
 
You must log in or register to reply here.
Back
Top