Intel Skylake / Kaby Lake

[lolfail9001]

Or maybe it is you that should think rationally? Wouldn't you rather run in 4.8 GHz in some apps and 4.3 GHz in the rest, instead of running at 4.3 GHz on all apps? Does it hurt your pride if your CPU has thermal throttling?

It hurts my knowledge of stability of overclock, if it thermally throttles during any stability testing, how am i supposed to learn whether it is stable at all? Get it or is it too hard to grasp? I won't even mention that CPUs do not thermal throttle like you think they do.

Of course, if the chips just can't get past 4.3 GHz, none of this is possible! I wonder what could be happening?

But since we know for a hard fact that it can run at 5.7Ghz on LN2, you are clueless. That's why i thought 10 seconds would be enough. Alas, it is not for people with preset agenda.

This doesn't look consistent with claims of 4.3 GHz OC limit. The Guru3D article showed 4.3 GHz at 1.25 V. I wonder which one is accurate.

CPU-Z does not detect voltage on SKL-X correctly and outputs random BS all the time at the time. So we can't really know what voltage that 4.3 was running at for time being.

 

[blue11]

It hurts my knowledge of stability of overclock, if it thermally throttles during any stability testing, how am i supposed to learn whether it is stable at all? Get it or is it too hard to grasp? I won't even mention that CPUs do not thermal throttle like you think they do.

Of course CPUs thermally throttle. How do you think laptops work? Based on your logic, we should lock all our laptops to base frequency so they never reach 100 C. If frequency is TDP limited, the best turbo target is higher than the maximum sustainable frequency. However, if frequency is electrically limited...

But since we know for a hard fact that it can run at 5.7Ghz on LN2, you are clueless. That's why i thought 10 seconds would be enough. Alas, it is not for people with preset agenda.

Everyone knows that running at cryogenic temperatures changes the frequency-voltage curve. It has nothing to do with what frequencies are electrically stable at room temperature. It also has no direct relation to cooling capacity.

CPU-Z does not detect voltage on SKL-X correctly and outputs random BS all the time at the time. So we can't really know what voltage that 4.3 was running at for time being.

Because you said so?

 

[TheF34RChannel]

The IMC is a significant upgrade compared to Broadwell-E, 4000 MHz+ should be relatively trouble-free.

Without much tweaking I hope. Basically I just want to set everything up manually without any tweaking (or just click XMP lol). I'd be equally happy with 3200-3466MHz.

This doesn't look consistent with claims of 4.3 GHz OC limit. The Guru3D article showed 4.3 GHz at 1.25 V. I wonder which one is accurate.

So they locked Intel RST behind a dongle? Not only are the X299s going to be starved out of lanes ($1000 for only 44), now you have to pay an additional charge to actually use them?! Meanwhile, on AMD you get 64 lanes, and there's no extra "DLC" to use them...

Well, the video shows proof whereas the article is the word from board vendors who may or may not have tried hard enough. I'd say: let's wait for reliable reviews.

Dongle; presumably because it's also paid for in the server platform they can't get away with giving it for free on HEDT.

 

[blue11]

Well, the video shows proof whereas the article is the word from board vendors who may or may not have tried hard enough. I'd say: let's wait for reliable reviews.

Dongle; presumably because it's also paid for in the server platform they can't get away with giving it for free on HEDT.

The video looks promising, since it doesn't look like much effort was put into the cooling (single tower with fan aligned wrong). However, board vendors usually quote optimistic figures for OC targets to create hype, so who knows what the deal is. As for the dongle, isn't RST free on client boards? It just plain looks bad if Intel is adding an upsell for something that's bundled on lower-end boards.

 

[TheF34RChannel]

The video looks promising, since it doesn't look like much effort was put into the cooling (single tower with fan aligned wrong). However, board vendors usually quote optimistic figures for OC targets to create hype, so who knows what the deal is. As for the dongle, isn't RST free on client boards? It just plain looks bad if Intel is adding an upsell for something that's bundled on lower-end boards.

Good point. Still, I'll take my own advice and wait. And even if it turns out that 4.3 is the average (seriously doubt it) it's overall still better than what I've got.

Hmm I thought it wasn't free, but don't quote me on that.

 

[mikk]

Because you said so?

Because it is business as usual over the years on new CPUs without CPUz support. So many times CPUz detects some Vcore nonsense on new CPUs and many people including you don't learn from their mistakes.

 

[lolfail9001]

How do you think laptops work?

They power throttle if you did not know, not thermal throttle. Or rather, when they do, they drop frequency below base.

If frequency is TDP limited, the best turbo target is higher than the maximum sustainable frequency.

Guess what, it is on 7900X. TBM3.0 is 4.5Ghz, remember?

However, if frequency is electrically limited...

Bring evidence and not speculation to it. So far you failed to even bring proper speculation to it speculating that having unstable overclock is better than having a stable overclock because unstable overclock will thermally throttle to a stable overclock. ROFL.

Everyone knows that running at cryogenic temperatures changes the frequency-voltage curve. It has nothing to do with what frequencies are electrically stable at room temperature. It also has no direct relation to cooling capacity.

Not quite right. Wrong. Wrong.

Because you said so?

No, because CPU-Z devs said so, look up your news.

 

[blue11]

They power throttle if you did not know, not thermal throttle. Or rather, when they do, they drop frequency below base.

Guess what, it is on 7900X. TBM3.0 is 4.5Ghz, remember?

Bring evidence and not speculation to it. So far you failed to even bring proper speculation to it speculating that having unstable overclock is better than having a stable overclock because unstable overclock will thermally throttle to a stable overclock. ROFL.

An "unstable" overclock is one that causes circuits to produce incorrect results. If the limiting factor is heat removal, a 4.8 GHz or whatever overclock is perfectly stable. Thermal throttling does in fact limit frequency climb on laptops, not only TDP throttling (which is also in play). You can take an old Haswell-era copper stock cooler and put it on a 7700K. It won't throttle to 800 MHz, but rather somewhere near base, and probably only when running AVX. It's not the TDP limit you'll be seeing, since Z-series boards disable TDP limits. If you are so scared of the thermal throttle boogeyman, you can even set a TDP limit to avoid it! Even though that makes no sense.

Not quite right. Wrong. Wrong.

Feel free to lock yourself in your padded room to keep those evil, evil wise men and their factual knowledge out of your fantasy land.

No, because CPU-Z devs said so, look up your news.

 

[SAAA]

http://browser.primatelabs.com/geekbench3/8370254
http://browser.primatelabs.com/geekbench3/8370249

Geekbench 3.4.1 results from i7-7800X and i7-7900X.

Wow that looks extremely interesting...If those scores are reliable a Skylake X 6/12 is quite a lot faster then an R7 1800X running at 4.1ghz which score around 30.000

Seriously, I know it's geekbench but this time is x86 vs x86, scores should reflect something meaningful.
As fair as it can get, a Skylake-X 8 core vs fastest 8 core Ryzen I could find:

http://browser.primatelabs.com/geekbench3/compare/8371149?baseline=8321340

Ryzen beats it only on SHA (it's specific hardware), for the rest there's quite a gap. Mind it might be L2 at work but certain subscores are 2x.

 

[blue11]

Seriously, I know it's geekbench but this time is x86 vs x86, scores should reflect something meaningful.
As fair as it can get, a Skylake-X 8 core vs fastest 8 core Ryzen I could find:

http://browser.primatelabs.com/geekbench3/compare/8371149?baseline=8321340

Ryzen beats it only on SHA (it's specific hardware), for the rest there's quite a gap. Mind it might be L2 at work but certain subscores are 2x.

Is GeekBench optimized for AVX (forget AVX-512)? The SGEMM/DGEMM scores look way too low on the 7820X vs the 1800X. There's only a 20% difference.

EDIT: Only 88 GFLOPS on "DGEMM". What a joke.

 

[lolfail9001]

An "unstable" overclock is one that causes circuits to produce incorrect results.

Yes, and testing that requires putting larger load on it than thermals can tolerate with your ingenious idea of overclocking.

If the limiting factor is heat removal, a 4.8 GHz or whatever overclock is perfectly stable.

No, it may be stable, or it may not be stable. You will never learn that until it's too late.

You can take an old Haswell-era copper stock cooler and put it on a 7700K. It won't throttle to 800 MHz, but rather somewhere near base

Evidence, please. AIDA64 and temp graph is all you need for that, don't you.

Feel free to lock yourself in your padded room to keep those evil, evil wise men and their factual knowledge out of your fantasy land.

Factual knowledge is on my side here, however. And i am an evil wise kid, too.

Seriously, I know it's geekbench but this time is x86 vs x86, scores should reflect something meaningful.

GB3 is notorious for being too cache-friendly.

 

[blue11]

Yes, and testing that requires putting larger load on it than thermals can tolerate with your ingenious idea of overclocking.

No, it may be stable, or it may not be stable. You will never learn that until it's too late.

So use your head? Instead of running prime95 for multiple days, you can cycle it on-off in 30 second intervals. That might actually be a harder stress test than just leaving it running, since voltage will be changing (overshoot, undershoot, etc.).

Evidence, please. AIDA64 and temp graph is all you need for that, don't you.

Factual knowledge is on my side here, however. And i am an evil wise kid, too.


GB3 is notorious for being too cache-friendly.

If you don't want to learn, nobody can teach you anything. You can just set a fan speed limit in your BIOS and run prime95 for a minute.

 

[SAAA]

Is GeekBench optimized for AVX (forget AVX-512)? The SGEMM/DGEMM scores look way too low on the 7820X vs the 1800X. There's only a 20% difference.

EDIT: Only 88 GFLOPS on "DGEMM". What a joke.

Yeah actual scores mine GB3 credibility as a whole, I did point out it's not a great tool.

GB3 is notorious for being too cache-friendly.

Must be that because there's no way this is happening for IPC gains:

http://browser.primatelabs.com/geekbench3/compare/8370952?baseline=8371149

I picked a casual but decent 7700K score and it's just absurd, getting consistent 30% over the same Skylake architecture...
On the other hand if some real programs also see this 30% gain, unlike cinebench meh scores, this year will get even more appealing for those who are willing to upgrade.

 

[wildhorse2k]

Seriously, I know it's geekbench but this time is x86 vs x86, scores should reflect something meaningful.
As fair as it can get, a Skylake-X 8 core vs fastest 8 core Ryzen I could find:

http://browser.primatelabs.com/geekbench3/compare/8371149?baseline=8321340

Ryzen beats it only on SHA (it's specific hardware), for the rest there's quite a gap. Mind it might be L2 at work but certain subscores are 2x.

Wow the difference is so huge. It's very surprising to see the latest tech from AMD being demolished so badly.

So they locked Intel RST behind a dongle? Not only are the X299s going to be starved out of lanes ($1000 for only 44), now you have to pay an additional charge to actually use them?! Meanwhile, on AMD you get 64 lanes, and there's no extra "DLC" to use them...

Threadripper is a paper CPU right now. Lots of PCIe lanes, cores, 2 dies but we don't know the real performance. It could be ok or very poor. Once we have benchmarks we can judge who will be more successful.

 

[blue11]

Yeah actual scores mine GB3 credibility as a whole, I did point out it's not a great tool.



Must be that because there's no way this is happening for IPC gains:

http://browser.primatelabs.com/geekbench3/compare/8370952?baseline=8371149

I picked a casual but decent 7700K score and it's just absurd, getting consistent 30% over the same Skylake architecture...
On the other hand if some real programs also see this 30% gain, unlike cinebench meh scores, this year will get even more appealing for those who are willing to upgrade.

They're either measuring memory/cache bandwidth or just plain wrong. A lot of the "integer" workloads look to be memory bound (encryption, compression, graph search). Who knows what kind of code is in their "floating-point" benchmarks if they only manage to get 88 GFLOPS in DGEMM (LINPACK).

 

[frozentundra123456]

Yea, way too much speculation from both camps right now. Plenty of FUD from Intel detractors and optimistic comments from supporters. We really wont know anything until we get side by side benchmarks in a variety of applications and get overclock results on Skylake X from a larger sample of chips.

 

[lolfail9001]

Yeah actual scores mine GB3 credibility as a whole, I did point out it's not a great tool.

There's a reason GB3 is not supported anymore, GB4 solved most of the dataset issues, but yeah, some still use GB3 for some reason.

On the other hand if some real programs also see this 30% gain, unlike cinebench meh scores, this year will get even more appealing for those who are willing to upgrade.

Well, i can imagine some with relatively small working set should see comparable gains, amount of cache and all.

That might actually be a harder stress test than just leaving it running, since voltage will be changing (overshoot, undershoot, etc.).

Changing, haha, changing.

You can just set a fan speed limit in your BIOS and run prime95 for a minute.

Why, i know how thermal throttling on my PC looks like: like a 800Mhz downclock.

 

[blue11]

Changing, haha, changing.

Why, i know how thermal throttling on my PC looks like: like a 800Mhz downclock.

If you don't want to learn, nobody can teach you anything.

Hey, if you want to double down on your ignorance, be my guest. The rest of us grown-ups will discuss the real world, while you play in your bubble pit.

 

[IntelUser2000]

6950X@4.5+DDR4 3200

ST 195
MT 2390

That doesn't really tell us anything. 5-10% differences exist on the same CPU from various results. The 6950X result might be the top one while SKL-X might be the lowest one. That's why you'd want everything equalized, like in a review.

 

[Shivansps]

Again, i would have expected that moving from 4-way L2 to 16-way L2 to already show some perf increase over mainstream Skylake, but no way it is like 30%.

 

[csbin]

That doesn't really tell us anything. 5-10% differences exist on the same CPU from various results. The 6950X result might be the top one while SKL-X might be the lowest one. That's why you'd want everything equalized, like in a review.

i9 7900X@5745MHz + DDR4 3426mhz CL12 (3209 cb ）World Records
http://hwbot.org/submission/3566634_the_overclocking_knights_cinebench___r15_core_i9_7900x_3209_cb/


i7 6950X@5323MHz + DDR4 3550mhz CL11（2878 cb ）World Records
http://hwbot.org/submission/3468936_xtreme_addict_cinebench___r15_core_i7_6950x_2878_cb/

 

[IntelUser2000]

I picked a casual but decent 7700K score and it's just absurd, getting consistent 30% over the same Skylake architecture...
On the other hand if some real programs also see this 30% gain, unlike cinebench meh scores, this year will get even more appealing for those who are willing to upgrade.

This is going to disappoint some.

7700K vs 6900K: http://browser.primatelabs.com/geekbench3/compare/8302732?baseline=8370952

The 6900K is faster in most of the single threaded portion of the subtests. So they are very sensitive to memory bandwidth.

Here's a comparison of 7820X vs 6900K: http://browser.primatelabs.com/geekbench3/compare/8371149?baseline=8302732

 

[mikk]

Why do you compare Windows vx Linux results? Geekbench runs slower on Windows.

 

[arandomguy]

Just speculation but could the future being more multi die designs be the engineering challenge that Intel is addressing with respect to soldier versus polymer TIM for use with IHSs?

 

[IntelUser2000]

Why do you compare Windows vx Linux results? Geekbench runs slower on Windows.

I just compared with the highest result. It just happened to be Linux.

Unless people have two identical systems you'll never get true perf/clock comparisons. Just wait for release. Everyone is so impatient. The fervor in enthusiast forums are just like in politics.

I made a mistake on the 7700K. I did not choose the top 3 score.

http://browser.primatelabs.com/geekbench3/compare/8305784?baseline=8204243

It still seems like a memory subsystem related benchmark. Geekbench is baffling to understand. Which is why we wait for the real thing. This changes things though.

The memory performance on the 7700K is noticeably higher than the 6900K. I've compared it based on Windows, Linux, highest single vs highest multi. Maybe 5-10% difference in the memory performance but 7700K is 40%+ higher than any 6900K results.