So I've downloaded a new prime 95

M76

M76

[H]F Junkie
Joined
Jun 12, 2012
Messages
14,031
Until now I used an old version from 1.5 years ago.

This ruined my oc completely. The new version stresses the cpu much more, and generates 10C° more heat, which is ridiculous.

Is this really comparable to any real-world application? I mean it jumps to 100C° instantly as soon as the test starts. It would settle at 90-91 with the old version.

Do we need to daily LN2 now to keep things cool? This is not even a hi-end CPU but a 3700x.
 
run the program torture test for a little while not all night.69C was as high as it got.That"s with a 6 core proc. though.
 
yes. its redicilous that a program that does a massive amount of realword application is optimized to use as many resources as possible to speed up the process.
OR
and this is just a suggesting. maybe you OC was not stable and you are going through a denial phase.
Different work takes different workload

This is why we do stress testing with hard to do stuff so we know its stable in lesser load.
Clearly your oc was not tested with this mentality for stability and now you see it breaking because the OC testing was done to weak.


Try removing the OC and se how it runs. it hard to bliame CPU/coolers when you are not running factory promised speeds anyway.
This is kinda on you are nobody has certified to run this speed. so putting out "do we need daily ln2" is clearly "no we don't" as its not needed on std clock. but it is and always have been an opportunity.
and this "issue" is from your choice. not the manufactors



Sidenote:
We see this kind of issue everyday on this and other forums
- nah use old prime26.6 becuase it has no AVX
- naah aida64 is fine prime is just to hard

All these excuses for bad overclock.
If the software runs fine on standard clock and not overclock. Your overclock is not stable
 
SvenBent said:
yes. its redicilous that a program that does a massive amount of realword application is optimized to use as many resources as possible to speed up the process.
OR
and this is just a suggesting. maybe you OC was not stable and you are going through a denial phase.
Different work takes different workload

This is why we do stress testing with hard to do stuff so we know its stable in lesser load.
Clearly your oc was not tested with this mentality for stability and now you see it breaking because the OC testing was done to weak.


Try removing the OC and se how it runs. it hard to bliame CPU/coolers when you are not running factory promised speeds anyway.
This is kinda on you are nobody has certified to run this speed. so putting out "do we need daily ln2" is clearly "no we don't" as its not needed on std clock. but it is and always have been an opportunity.
and this "issue" is from your choice. not the manufactors



Sidenote:
We see this kind of issue everyday on this and other forums
- nah use old prime26.6 becuase it has no AVX
- naah aida64 is fine prime is just to hard

All these excuses for bad overclock.
If the software runs fine on standard clock and not overclock. Your overclock is not stable
Click to expand...
What? I clearly stepped on a nerve. The torture test in prime is the exact opposite of real world application.
Did you even read my post? I said I was running an older prime version, no not 26.6 that, 28.something.
And compared to that temps jumped 10C°
I didn't think that was normal and started the topic to see anyone else noticed that.

After all I've seen furmark throwing artifacts on stock clocks on many a video card, so the same isn't as far fetched for a cpu test as you make it sound.

I didn't try to deny anything you don't have to be defensive about your precious 48 hour stable small fft oc, or whatever you consider a good OC. I don't need braggin rights, I need faster speeds for my renders, nothing more.
 
New version will use AVX instructions and you will see a heat climb due to that. Intel guys have the same problem when they run it as well. You have to have exceptional cooling if you want to run that and keep the clocks up high.
 
Gideon said:
New version will use AVX instructions and you will see a heat climb due to that. Intel guys have the same problem when they run it as well. You have to have exceptional cooling if you want to run that and keep the clocks up high.
Click to expand...
Thanks.

It doesn't mean I'm calling a retreat. Just going back to the drawing board and working my way up the chain again. I was up to 4300, now I'm back to 4150, not the end of the world. Still much faster than stock.

Seems to be temperature limited at this point.
 
SvenBent said:
yes. its redicilous that a program that does a massive amount of realword application is optimized to use as many resources as possible to speed up the process.
OR
and this is just a suggesting. maybe you OC was not stable and you are going through a denial phase.
Different work takes different workload

This is why we do stress testing with hard to do stuff so we know its stable in lesser load.
Clearly your oc was not tested with this mentality for stability and now you see it breaking because the OC testing was done to weak.


Try removing the OC and se how it runs. it hard to bliame CPU/coolers when you are not running factory promised speeds anyway.
This is kinda on you are nobody has certified to run this speed. so putting out "do we need daily ln2" is clearly "no we don't" as its not needed on std clock. but it is and always have been an opportunity.
and this "issue" is from your choice. not the manufactors



Sidenote:
We see this kind of issue everyday on this and other forums
- nah use old prime26.6 becuase it has no AVX
- naah aida64 is fine prime is just to hard

All these excuses for bad overclock.
If the software runs fine on standard clock and not overclock. Your overclock is not stable
Click to expand...

A few more side notes. Prime 95 isn’t exactly real world. If it is, please share with us some actual real world applications that most of us are likely to use that would do what prime 95 does. It is however a great program to test stability and cooling so while it’s anything but real works, it’s silly to wish it was less stressful than it is.

What OP experienced doesn't necessarily mean a bad overclock, unless the test actually failed. It means he has less than exceptional cooling.
 
RamonGTP said:
It means he has less than exceptional cooling.
Click to expand...
Actually afaik this is the best air cooler money can buy, or close to it, beating entry level aios. I don't think you can get much better than an NH-D15 without building a custom loop. And at that point I'd rather upgrade my cpu for the money than invest $4-500 in a cooling that might allow 100mhz more OC.
 
new version of prime maybe makes better use of avx and other instructions ...so it loads the cpu more efficiently than older versions

That could easily explain 10 degrees

in the end, regardless of what prime is or is not representing. If software is able to push the hardware to reach temps it shouldn't, the problem isn't in the software. If we're doing things right, the hardware should be configured to handle the worst case scenario without sacrificing the general case performance. Yours has been shown to be deficient in that sense - but it's up to them to decide if it matters, since this is the absolute worst case and it could be something they'll never ever see in real life.
 
M76 said:
Actually afaik this is the best air cooler money can buy, or close to it, beating entry level aios. I don't think you can get much better than an NH-D15 without building a custom loop. And at that point I'd rather upgrade my cpu for the money than invest $4-500 in a cooling that might allow 100mhz more OC.
Click to expand...

Best air cooler doesn't equate to exceptional cooling. There are cooling solutions that can keep from thermal throttling as soon as you hit the start button, so.... yeah. It's good, it's not exceptional.

I can run Prime for about a minute before hitting 95C on a 12 core CPU with 280mm AIO, and I don't consider my cooling exceptional by any stretch.
 
M76 said:
Until now I used an old version from 1.5 years ago.

This ruined my oc completely. The new version stresses the cpu much more, and generates 10C° more heat, which is ridiculous.

Is this really comparable to any real-world application? I mean it jumps to 100C° instantly as soon as the test starts. It would settle at 90-91 with the old version.

Do we need to daily LN2 now to keep things cool? This is not even a hi-end CPU but a 3700x.
Click to expand...

3700x is pretty damn high end. Stop judging it by its price.

It defeats almost everything Intel has except the 9900k.
 
RamonGTP said:
Best air cooler doesn't equate to exceptional cooling. There are cooling solutions that can keep from thermal throttling as soon as you hit the start button, so.... yeah. It's good, it's not exceptional.

I can run Prime for about a minute before hitting 95C on a 12 core CPU with 280mm AIO, and I don't consider my cooling exceptional by any stretch.
Click to expand...

a 3900x is probably technically impossible to currently cool to levels people have been used to seeing for similar tdp cpu's. Heat density has moved the new "exceptional" up many degrees C. I'd imagine custom loops in the same ambient temp will not do much better than that 280mm aio unless your setup is lacking enough air CFM thru the radiator. _maybe_ top at the low 90's. I'm tempted to boot into windows after work and see what mine settles at. 75-77F ambient temp. benchmarks max at the high mid 70's to low 80's. I suspect prime95 will settle around 90 ..but that's with undervolting
 
tangoseal said:
3700x is pretty damn high end. Stop judging it by its price.

It defeats almost everything Intel has except the 9900k.
Click to expand...
The fact that intel has nothing to compete doesn't make the 3700X hi end, after all there are two more cpus above it in line now, with one more coming soon.

If we can ever consider a consumer platform hi-end. AMD not just ripped the rug under from intel, but also from their own HEDT lineup. Except maybe for the 2970 and 2990.
 
M76 said:
The fact that intel has nothing to compete doesn't make the 3700X hi end, after all there are two more cpus above it in line now, with one more coming soon.

If we can ever consider a consumer platform hi-end. AMD not just ripped the rug under from intel, but also from their own HEDT lineup. Except maybe for the 2970 and 2990.
Click to expand...

If your just trying to get an ALL Core overclock ,try AMD Overclocking in the BIOS.Set LCC to a normal level to advoid Vdrop if you want.Leaving the CPU do it's thing for the best results overall.
3700X good starting point
4300Mhz
1325-1350 to test stability even at these lower voltage the Prmine95 test will hit 85c-90c most likely.

Here is an example of the most stressful thing I do on my computer which can generate higher temperatures than old Prime95 if interested. Re-encode videos with memory test 100%,Intel Burn test with drop 4250Mhz under load drops 1.26v.
4250-1-3v-png.png
 
Prime 95 was never indicative of real world usage. Unless you're into running some weird shit. It's always been meant to be as stressful as possible to test for any possible instability.

If you only do gaming run the most stressful game you can play. You can run much higher clocks and get much better performance tailoring your OC to your actual usage. Just know if you decide to do something else you may not be stable.
 
M76 said:
Until now I used an old version from 1.5 years ago.

This ruined my oc completely. The new version stresses the cpu much more, and generates 10C° more heat, which is ridiculous.

Is this really comparable to any real-world application? I mean it jumps to 100C° instantly as soon as the test starts. It would settle at 90-91 with the old version.

Do we need to daily LN2 now to keep things cool? This is not even a hi-end CPU but a 3700x.
Click to expand...

The newer P95's support AVX instruction sets so that will hammer your cpu harder than anything else it will ever do. PPL either prep for AVX loads or they do not since 99% of apps do not use AVX. I validated my 3900x with with P95 AVX just to for the extra padding. Then again I do have a big loop to cool it, [email protected] maxed out in P95 AVX = 77c.
 
  • Like
Reactions: Zinn
like this
sharknice said:
Prime 95 was never indicative of real world usage. Unless you're into running some weird shit. It's always been meant to be as stressful as possible to test for any possible instability.

If you only do gaming run the most stressful game you can play. You can run much higher clocks and get much better performance tailoring your OC to your actual usage. Just know if you decide to do something else you may not be stable.
Click to expand...

I agree for instance you can hold a lower voltage temperature and higher clock rate if you disable smt. If all you're doing is gaming on a 8 core then smt can actually hamper performance for gaming. This has been verified.

I drop like .10volts on my 3900x just turning smt off and temps are lower. Something like that.
 
gerardfraser said:
If your just trying to get an ALL Core overclock ,try AMD Overclocking in the BIOS.Set LCC to a normal level to advoid Vdrop if you want.Leaving the CPU do it's thing for the best results overall.
3700X good starting point
4300Mhz
1325-1350 to test stability even at these lower voltage the Prmine95 test will hit 85c-90c most likely.

Here is an example of the most stressful thing I do on my computer which can generate higher temperatures than old Prime95 if interested. Re-encode videos with memory test 100%,Intel Burn test with drop 4250Mhz under load drops 1.26v.
View attachment 182519
Click to expand...
I didn't see any vdrop at all under load. Smallffts hit 87c at 4150 already.
 
M76 said:
I didn't see any vdrop at all under load. Smallffts hit 87c at 4150 already.
Click to expand...

The only true way to see voltage droop is with a multimeter although hwinfo 64 may update fast enough to catch it.

I load up p95 and then watch my multimeter
You'll see a remarkable drop in voktage if your llc is set to low.

But LLC will allow more current for sure if set high enough.
 
tangoseal said:
I agree for instance you can hold a lower voltage temperature and higher clock rate if you disable smt. If all you're doing is gaming on a 8 core then smt can actually hamper performance for gaming. This has been verified.

I drop like .10volts on my 3900x just turning smt off and temps are lower. Something like that.
Click to expand...

AMD's SMT is super, no reason to disable.
 
x99 5930k oc'd to 4.5GHz at 1.28v.

First curve is running furmark, second plateau is prime95.

6lpzKZV.png


All cores temps:
4rcNSpi.png


No issues here =)
 
thesmokingman said:
It defeats the point in getting a high core cpu.
Click to expand...

I know but I'm just tossing the option out there.

But then again...with smt disabled your still getting 12 real cores. Smt is just a way to use under utilized cores to piggyback the big busy cores. It does show a gain in gaming to have it off. I assume it reduces latency and logic complexity penalties (maybe?).
 
tangoseal said:
I know but I'm just tossing the option out there.

But then again...with smt disabled your still getting 12 real cores. Smt is just a way to use under utilized cores to piggyback the big busy cores. It does show a gain in gaming to have it off. I assume it reduces latency and logic complexity penalties (maybe?).
Click to expand...

Ugh, it varies from game to game. There are some small gains vs small regression. In fact in some games there can be a huge regression with disabling SMT.

 
I have all but stopped using Prime95 (newer than 26.6) since it places completely unrealistic loads/heat on the CPU. I have been sticking with AIDA64, OCCT, and Reelbench.
 
It will be very hard to find a program that runs the torture prime 95 does with new avx instructions. It is basically a power virus and I think it is designed to test more your cooling system then actual overclock. Heck it even says that somewhere lol.
 
Same here but realbench for me doesn't work for some reason. I can run stock, it runs in to memory swap issue or something from what I researched. Its random and even on everything auto I get instability detected. It might have to do with me running two 16gb sticks. But all other stability tests including memtest passes just fine. Realbench is randon it can pass through for hour or 2 and give me error or after 10 mins. Not sure whats up.
 
Last edited:
NKD said:
Same here but realbench for me doesn't work for some reason. I can run stock, it runs in to memory swap issue or something from what I researched. Its random and even on everything auto I get instability detected. It might have to do with me running two 16gb sticks. But all other stability tests including memtest passes just fine. Realbench is randon it can pass through for hour or 2 and give me error or after 10 mins. Not sure whats up.
Click to expand...
Uh... You just replied to yourself?
 
GoldenTiger said:
Uh... You just replied to yourself?
Click to expand...

haha. didn't even realize. haha. happens studying late night on some real estate shit. lol

No wonder I took some time to find my glasses today. My eyes were tired as hell haha.

fixed.
 
Last edited:
My standard stability test has been the 'Intel Burn Test' with default settings. Any reason not to use that nowadays?
 
Load line calibration actually helps quite a bit. After enabling it I could lower voltage by about 0.025 while increasing frequency by 25Mhz, and it still doesn't fail the smallfft test in the new prime95 compared to where it previously produced errors in a few minutes.
 
Prime95 uses AVX which stresses your CPU like nothing else. Your OC was not stable.

Xeon chips have different turbo offsets for when AVX instruction sets are used (they run slower to reduce power usage). As far as I know, the consumer chips don't offer this.

Best bet would be to back off your OC or just not use anything that runs AVX code.
 
Zinn said:
Xeon chips have different turbo offsets for when AVX instruction sets are used (they run slower to reduce power usage). As far as I know, the consumer chips don't offer this..
Click to expand...

On Intel side, users can choose lower multipliers for AVX loads.
 
  • Like
Reactions: Zinn
like this
Tried Prime 95 on my new 3700X.

16 thread stress test - all cores @ 4GHz, 65C
2 thread stress test - two cores @ 4.2 GHz, 75C+, fans spinning like mad.

Is it supposed to work like that?
 
ashmelev75 said:
Tried Prime 95 on my new 3700X.

16 thread stress test - all cores @ 4GHz, 65C
2 thread stress test - two cores @ 4.2 GHz, 75C+, fans spinning like mad.

Is it supposed to work like that?
Click to expand...
Did you try with 8 threads? The amount of queuing going on with SMT might have actually bottlenecked the chip down to the point where it wasn't able to hit as high of a power load.
 
You must log in or register to reply here.
Back
Top