NEW MMO-Champion CPU Overclocking Leaderboards

[nocturnus]

snip[/IMG]

Very nice Tommi!

I'm sure quite sure the leaderboard requires Prime or Aida for a mention. Regardless, this is a pretty nice overclock with amazing temps. What are you cooling it with and have you done a de-lid?

I had the same thing you had by the way. Realbench runs fine for 8 hours at 5.1ghz while Aida reports errors relatively quickly. I'm still fidling with some settings before I attempt a new Aida run for the leaderboards.

[Tommi]

Very nice Tommi!

I'm sure quite sure the leaderboard requires Prime or Aida for a mention. Regardless, this is a pretty nice overclock with amazing temps. What are you cooling it with and have you done a de-lid?

I had the same thing you had by the way. Realbench runs fine for 8 hours at 5.1ghz while Aida reports errors relatively quickly. I'm still fidling with some settings before I attempt a new Aida run for the leaderboards.

Cooling with http://www.fractal-design.com/home/p...ng/celsius-s36

yeah, it's been delidded and had the TIM replaced with liquid metal based TIM.

[DeltrusDisc]

I'm inclined to add Realbench unless anyone has any issue with it. @Evildeffy you're pretty knowledgeable on these things, what do you think?

[Evildeffy]

I'm inclined to add Realbench unless anyone has any issue with it. @Evildeffy you're pretty knowledgeable on these things, what do you think?

RealBench, as far as I'm aware, is a little bit more relaxed to check for common stability for the most part.
It works and isn't an issue but I wouldn't count on that test for critical items like production.

Because things do slip through the cracks and the reason why Prime95 is such a powerful stability checker is the fact of that unrealistic load.
If it remains 100% stable with that, it'll remain stable with anything.

You can separate AVX stability tested or not in that sense meaning Prime95 26.6 tested for non-AVX loads (good deal of people don't care for AVX) and AVX tested with Prime95 29.3 (I think it was) for 100% stability with a precaution that AVX loads on Intel chips is VERY hot and shouldn't be attempted without a beast of a cooler.

So for general all-purpose non-critical things: Sure, RealBench is fine for that.
And for critical tasks: Fuck no, Prime95 fully loaded or bust!

Reason: Production critical tasks can put quite a bit of strain on the CPU, way more than gaming does right now (exceptions not counted), so what may be stable for gaming does not even remotely have to be the case for heavy CPU loads that aren't benchmarks.

That's my opinion on the matter in any case, do with it as you please.

[nocturnus]

RealBench, as far as I'm aware, is a little bit more relaxed to check for common stability for the most part.
It works and isn't an issue but I wouldn't count on that test for critical items like production.

Because things do slip through the cracks and the reason why Prime95 is such a powerful stability checker is the fact of that unrealistic load.
If it remains 100% stable with that, it'll remain stable with anything.

You can separate AVX stability tested or not in that sense meaning Prime95 26.6 tested for non-AVX loads (good deal of people don't care for AVX) and AVX tested with Prime95 29.3 (I think it was) for 100% stability with a precaution that AVX loads on Intel chips is VERY hot and shouldn't be attempted without a beast of a cooler.

So for general all-purpose non-critical things: Sure, RealBench is fine for that.
And for critical tasks: Fuck no, Prime95 fully loaded or bust!

Reason: Production critical tasks can put quite a bit of strain on the CPU, way more than gaming does right now (exceptions not counted), so what may be stable for gaming does not even remotely have to be the case for heavy CPU loads that aren't benchmarks.

That's my opinion on the matter in any case, do with it as you please.

What do you think about Aida? My 5.1 clock is stable in Realbench as well, but crashes in Aida. I never posted Realbench shots, because it isn't accepted for the leaderboards.

[Evildeffy]

What do you think about Aida? My 5.1 clock is stable in Realbench as well, but crashes in Aida. I never posted Realbench shots, because it isn't accepted for the leaderboards.

Honestly my AIDA64 opinion may be weird...

It's a tool used by a lot of people and I can understand why because of it's ease of use and monitoring tools but it's load stability test function I rated rather poorly since it's loads were less than that of Prime95 and OCCT (I think? Bright red coloured tester) when I messed about with them.

It may have changed over time like f.ex. the fact that AIDA64 can now do AVX testing as well which it couldn't do before (located in Test FPU!).
So right now my opinion on AIDA64 is "Unsure" in reality as I simply do not know properly what load it generates in comparison to Prime95.

The overall overclocking feeling is simple though, AIDA64 and RealBench test overall stability in non-critical areas which may be perfectly fine for gaming and other workloads, which if it does crash isn't a big impact on things, annoying yes but nothing critical.
Whereas if you have critical production such as YouTube videos or Twitch (or other platform streaming) at a more professional level my recommendation for testing remains Prime95 with AVX torture loads ... which people have to be careful with to not overload their cooling system with a big overclock.

If AIDA64 gets close to Prime95's torture ability (say 95% of the way) then you could use AIDA64 as a good standard for testing ... but from what I've seen so far Prime95 is still king by far for utmost stability testing, especially when AVX gets involved with Intel.
Like I said ... the "unrealistic" load is exactly why it's so powerful a tool and the load isn't as unrealistic as people think when professional multi-threaded software comes into play, granted small FFTs will still be insanely hot and power consuming but the load on the CPU itself can be equal.

So... AIDA64's current position is LIKELY the same as RealBench, general all-purpose non-critical things ... Fine for that but I'd make sure it doesn't bug out at least with FPU testing disabled since that involves AVX.

[nocturnus]

Honestly my AIDA64 opinion may be weird...

It's a tool used by a lot of people and I can understand why because of it's ease of use and monitoring tools but it's load stability test function I rated rather poorly since it's loads were less than that of Prime95 and OCCT (I think? Bright red coloured tester) when I messed about with them.

It may have changed over time like f.ex. the fact that AIDA64 can now do AVX testing as well which it couldn't do before (located in Test FPU!).
So right now my opinion on AIDA64 is "Unsure" in reality as I simply do not know properly what load it generates in comparison to Prime95.

The overall overclocking feeling is simple though, AIDA64 and RealBench test overall stability in non-critical areas which may be perfectly fine for gaming and other workloads, which if it does crash isn't a big impact on things, annoying yes but nothing critical.
Whereas if you have critical production such as YouTube videos or Twitch (or other platform streaming) at a more professional level my recommendation for testing remains Prime95 with AVX torture loads ... which people have to be careful with to not overload their cooling system with a big overclock.

If AIDA64 gets close to Prime95's torture ability (say 95% of the way) then you could use AIDA64 as a good standard for testing ... but from what I've seen so far Prime95 is still king by far for utmost stability testing, especially when AVX gets involved with Intel.
Like I said ... the "unrealistic" load is exactly why it's so powerful a tool and the load isn't as unrealistic as people think when professional multi-threaded software comes into play, granted small FFTs will still be insanely hot and power consuming but the load on the CPU itself can be equal.

So... AIDA64's current position is LIKELY the same as RealBench, general all-purpose non-critical things ... Fine for that but I'd make sure it doesn't bug out at least with FPU testing disabled since that involves AVX.

Aye, that's pretty much what I think of Prime95, which is the same reason I avoid it myself. This doesn't explain why both mine as Tommi's 5.1ghz stress test crash in Aida but survive Realbench.

[Evildeffy]

Aye, that's pretty much what I think of Prime95, which is the same reason I avoid it myself. This doesn't explain why both mine as Tommi's 5.1ghz stress test crash in Aida but survive Realbench.

It doesn't no, like I said I have no real answer as to how far AIDA64 has come in terms of load...

Either which way if the general stability is fine for you and you don't intend to do production... I don't see an issue.
Though I'd check if you could potentially solve the AIDA64 one.

However like I said... have you de-clicked the "FPU Test" bit?

[nocturnus]

It doesn't no, like I said I have no real answer as to how far AIDA64 has come in terms of load...

Either which way if the general stability is fine for you and you don't intend to do production... I don't see an issue.
Though I'd check if you could potentially solve the AIDA64 one.

However like I said... have you de-clicked the "FPU Test" bit?

No, but without the FPU test, the test isn't valid. Anyway, I changed some settings in the BIOS and Aida did a succesful 2 hour run (had it run while I was at training). So it might be ready for the full 8 hour torture test!

[Tommi]

No, but without the FPU test, the test isn't valid. Anyway, I changed some settings in the BIOS and Aida did a succesful 2 hour run (had it run while I was at training). So it might be ready for the full 8 hour torture test!

Mines failed at 2 hours, 4 hours and 7 hours in AIDA. It seems to be completely random. Everything else is fine though. It might just be a thing with the Coffee Lake processors above 5Ghz with AIDA.

[nocturnus]

Mines failed at 2 hours, 4 hours and 7 hours in AIDA. It seems to be completely random. Everything else is fine though. It might just be a thing with the Coffee Lake processors above 5Ghz with AIDA.

Nah, it's a pure stability issue. New Aida versions also test AVX intructions and while it's less hefty than Prime, it's more demanding than Realbench.

[Tommi]

Nah, it's a pure stability issue. New Aida versions also test AVX intructions and while it's less hefty than Prime, it's more demanding than Realbench.

Yet Prime passes 8 hours fine O.o

I'll run another test tonight and see where it fails.

[Evildeffy]

Yet Prime passes 8 hours fine O.o

I'll run another test tonight and see where it fails.

If AIDA64 fails when Prime95 v29.x does not than there's something wrong with AIDA64.

By no accounts is AIDA64 heavier in load than Prime95.

Are you sure you didn't just use Prime95 26.6? As this version doesn't stress AVX instructions.
Where the lastest versions of AIDA64, like I said above, stress AVX instructions in the FPU tests.

[Tommi]

If AIDA64 fails when Prime95 v29.x does not than there's something wrong with AIDA64.

By no accounts is AIDA64 heavier in load than Prime95.

Are you sure you didn't just use Prime95 26.6? As this version doesn't stress AVX instructions.
Where the lastest versions of AIDA64, like I said above, stress AVX instructions in the FPU tests.

29.3 is the prime version I have.

[nocturnus]

Yet Prime passes 8 hours fine O.o

I'll run another test tonight and see where it fails.

:/ I doubt you used a recent version of prime95 or you tweaked the settings; nothing fails if you pass an 8 our Prime avx blend test.

[Tommi]

:/ I doubt you used a recent version of prime95 or you tweaked the settings; nothing fails if you pass an 8 our Prime avx blend test.

Ah, I dunno whats happening then. Not too bothered about being on some leaderboard tbh :P

[DeltrusDisc]

Ah, I dunno whats happening then. Not too bothered about being on some leaderboard tbh :P

It's not just a leaderboard, it's a show of stability that you put the effort in to attain.

[Buildapanda]

this is worthless now with the new intel vulnerability update that slows performance by up to 50% i woulnd't even not patch even for a leaderboard

[nocturnus]

this is worthless now with the new intel vulnerability update that slows performance by up to 50% i woulnd't even not patch even for a leaderboard

It can be between 5% and 30%. But it's not perceptible for regular users in practice. Specific enterprise usage might paint a different picture.

This article is in Dutch, but the charts are clear enough regardless:

https://nl.hardware.info/reviews/784...aties-van-cpus

If you want to check your vulnerability you can download an Intel tool from this website.

[DeltrusDisc]

this is worthless now with the new intel vulnerability update that slows performance by up to 50% i woulnd't even not patch even for a leaderboard

Please do more reading and research before making broad, completely incorrect claims such as this. Where did you hear 50%? That's just not even remotely the truth of the matter. I suggest you delete this comment.