Intel Skylake / Kaby Lake

[Timmah!]

OC3D together with Der8auer:

So does this mean, i actually can overclock 7900x to say 4,5GHz and use it for regular gaming, 3D rendering, Cinebench, etc... simply usual stuff, and it will be fine, even possibly cooled by 240/280mm AIO? And all the throttling, 90+C temps on CPU, 100+C temps on VRMs and ridiculous power-draw would concern me only if i tried to stress test with the Prime?

 

[mikk]

No offense but that is all crap. The difference between Ryzen, a 7700, and let's say a 7900x on AVX is that AMD A.) Doesn't use 512 like the 7900 does. B.) Uses 2 128bit piplines for handling 1 AVX 2 instructions.

I know that. That's what I meant, nothing new here.

 

[tamz_msc]

I'm sure TahoeDust can just fire up some benchmarks, stress tests, and HWinfo, and show us the CPU Performance Limit Reasons read out. If there's throttling invisible to CPU-Z, it should still show up there. He actually owns the platform and has run a lot of benchmarks, unlike the people that just show up in this thread to troll.

Edit: I'm now wondering how in the world der8auer got a 250W VRM output to throttle at 105C.

FIVR input is 1.8V nominal, so that's 140A.

If these are using IR3550: Divide by 8 phases and you get 2W power loss per phase for a total of 16W waste heat. 6 phase VRM is only nominally worse at 18W.

This is why he observes throttling.

90% efficiency at 250W output is 28W waste heat, which is 3.5W/phase for 8-phase VRM. With 5*C/W delta between PCB and MOSFET that's 17.5*C. Now he showed HWinfo reading 105*C on the VRM, thus the actual MOSFET is running at 122.5*C, which is very close to the continuous 125*C rating for most MOSFET's junction temperature. Of course in the real world things will deviate from this simple calculation, so yes, when you are consuming 250W maximum there is a possibility that you'll have VRM throttling.

I mean, when your heatsink is like this(Asus X299-A), with a white finish on the other side,

you know you're heading for trouble.

 

[tamz_msc]

They don't need that because Ryzen doesn't run AVX1/AVX2 full speed. So they don't have these heat issues there but they also suffer from a poor performance if AVX2 is properly used, recent x265 binaries for example.

I've looked at numbers for the 1800X and 6900K here - slower AVX2 doesn't make that much of a difference in SPECint2006, for example, even with the Intel compiler heavily skewing the results due to libquantum. And in SPECfp2006 the Ryzen CPU is 12% faster than Broadwell-E in single-threaded performance.

Plus we have this

Summing up, the Ryzen CPU cannot keep up with current Intel designs in terms of in-core performance with AVX(2) code. On the other hand, its overlapping cache hierarchy and good memory bandwidth (for a desktop chip) can regain a lot of lost ground despite the non-scalable L3 cache.

So really, AVX2 isn't much of a problem, as long as you're not limited to small data sets.

 

[mikk]

I've looked at numbers for the 1800X and 6900K here - slower AVX2 doesn't make that much of a difference in SPECint2006, for example, even with the Intel compiler heavily skewing the results due to libquantum. And in SPECfp2006 the Ryzen CPU is 12% faster than Broadwell-E in single-threaded performance.

Not much AVX2 code in there. AVX2 support alone won't help. Kabylake has a 40% better IPC than Ryzen for x265 because of AVX2.

 

[coercitiv]

So does this mean, i actually can overclock 7900x to say 4,5GHz and use it for regular gaming, 3D rendering, Cinebench, etc... simply usual stuff, and it will be fine, even possibly cooled by 240/280mm AIO? And all the throttling, 90+C temps on CPU, 100+C temps on VRMs and ridiculous power-draw would concern me only if i tried to stress test with the Prime?

What it means is you can overclock the 7900X to 4.5Ghz+, but heavy loads like Cinebench may end up being throttled depending on a number of factors (BIOS settings, VRM cooling, CPU quality & cooling etc). CPU thermal throttling will be the most likely though.

So enthusiasts looking to overclock the 7900X should choose their boards and cooling carefully. Anybody else on 8C and 6C will likely never see this behavior on 4.5-4.7Ghz overclocks.

 

[tamz_msc]

Not much AVX2 code in there. AVX2 support alone won't help. Kabylake has a 40% better IPC than Ryzen for x265 because of AVX2.

Just check for yourself on their website. It's laughable how much libquantum can skew those results, yet Zen isn't far behind.

Prime95 is far more useful as a stress-testing tool - how many people in the world care about finding the next Mersenne prime? Though it can check how fast you can compute FFTs, which is of academic interest only unlike SPEC which constitutes real-world code.

Where do you get 40%? On hardware.fr Skylake-X is only 20% faster clock-for-clock in x265.

 

[Atari2600]

I mean, when your heatsink is like this(Asus X299-A), with a white finish on the other side,

you know you're heading for trouble.

Is that... foam?!?!

 

[mikk]

Where do you get 40%? On hardware.fr Skylake-X is only 20% faster clock-for-clock in x265.

From user benchmarks with a current binary. Many reviews might use older x265 binaries, also the result differ from setting to setting.

https://docs.google.com/spreadsheets/d/1z9OMFSLGyCKeFG0Q6jzIbp-RRL5OpjJ_WLUzFUpuo7g/edit#gid=0

 

[tamz_msc]

Is that... foam?!?!

Some sort of alloy, with a thermal pad. What happened to the era of copper heatsinks, I wonder?

 

[Atari2600]

What happened to the era of copper heatsinks, I wonder?

I don't understand why they even stick on these big metal lumps. Anyone who knows anything about heat transfer will know they are idiotic! They might act as a sink for very brief transient power peaks, but any sustained operation will quickly saturate the heat capacity of the lump.

A few copper fins on the VRMs to increase surface area, then have a small fan running air across them. If the VRMs are running at >100 degC, then that small bit of airflow will keep them cool - regardless of whether the internal temperature of the box climbs up above 50/60.

 

[tamz_msc]

From user benchmarks with a current binary. Many reviews might use older x265 binaries, also the result differ from setting to setting.

https://docs.google.com/spreadsheets/d/1z9OMFSLGyCKeFG0Q6jzIbp-RRL5OpjJ_WLUzFUpuo7g/edit#gid=0

Again, can't tell anything from that without seeing what options were used at runtime. Even if it's 40% faster, that's just one scenario. Every architecture has it's strengths and weaknesses. I'm sure that AMD offering Broadwell-E levels of performance on average doesn't put AMD in a bad position, even if you compare it with Skylake-X, or the upcoming Coffee Lake.

 

[mikk]

Again, can't tell anything from that without seeing what options were used at runtime. Even if it's 40% faster, that's just one scenario. Every architecture has it's strengths and weaknesses. I'm sure that AMD offering Broadwell-E levels of performance on average doesn't put AMD in a bad position, even if you compare it with Skylake-X, or the upcoming Coffee Lake.

https://onedrive.live.com/?authkey=!AJ1HAbfOqf75wTI&id=8D4396DA2D9EBA37!242309&cid=8D4396DA2D9EBA37


Download and have a look. Feel free to post your benchmarks.

 

[tamz_msc]

https://onedrive.live.com/?authkey=!AJ1HAbfOqf75wTI&id=8D4396DA2D9EBA37!242309&cid=8D4396DA2D9EBA37


Download and have a look. Feel free to post your benchmarks.

Thanks for the link, will test on the only Windows systems I have access to - a Core 2 Duo E6300 and a laptop with a Core i3 7100U.

 

[MarkPost]

You know P95 wasn't updated to support Ryzen until very recently. P95 testing with Ryzen was done at launch, with it showing abnormally low power consumption for a stress test.



Someone with Ryzen should test 29.2 and see how much it drives power consumption up. Same for those with SKL-X as they get a AVX512 workout... especially the 7900x without execution penalty vs the 6c and 8c parts.

der8auer's P95 power numbers were done with 27.1... this version implements up to Sandy/Ivy AVX... the AVX2/FMA/AVX512 hardware in SKL-X goes completely unutilized. 4.5GHz 1.25v 7900x under those conditions and running small FFTs measured 502w at the wall . I don't even want to know how much those power numbers would shoot up with these parts of the CPU running full tilt and having the CPU configured to not throttle.

I've done a quick run (~15min, small-FFT) with my 1700X, and it seems with 29.2 consumes less power than with 28.10 (specially CPU package). Next week I'll receive a 7800X & MSI Carbon, so I will check with them.

28.10:

29.2:

 

[Markfw]

I can show you one that will outrun anything AMD makes, even when overclocked, while drawing 165w.

That 170 watts is the entire systems draw, motherboard, memory, a 1080TI, EVERYTHING. So what CPU are you talking about ?

 

[TahoeDust]

That 170 watts is the entire systems draw, motherboard, memory, EVERYTHING. So what CPu are you talking about ?

If he was talking about at the wall, why did he specify Motherboard Specs?

Honestly, system draw is the absolutely last thing I am concerned with. I certainly did not build a x299 machine to be economical.

 

[Markfw]

If he was talking about at the wall, why did he specify Motherboard Specs?

Who, What ??? I was replying to your reply to me about your CPU that can beat anything AMD with 165 watts overclocked, but I think that not including the rest of the system,

 

[TahoeDust]

Who, What ??? I was replying to your reply to me about your CPU that can beat anything AMD with 165 watts overclocked, but I think that not including the rest of the system,

My reply was not to you. It was to this:

You know what? If you can find a 7820x that runs Prime95 at stable 4.3Ghz without drawing way over the rated power for Skylake-X motherboards (165W) I'd be happy to to concede that Skylake might just be a "huge mistake" and not an absolute unmitigated disaster. Too bad there isn't much chance of that

Hence, my quoting it in my reply.

 

[Nothingness]

Don't know if this was already posted: http://www.mersenneforum.org/showpost.php?p=462751&postcount=64

It looks like SKL-X can reduce IPC without lowering clock. AVX code of course is the most affected.

 

[TheF34RChannel]

Well, I don't have my kill-a-watt on it right now, but its running and 52c is the highest its gotten.

The Ryzen 1800x

Edit, its now hit 55c after 16 minutes

OK, I shut it down and put the kill-a-watt in. Idle, 65 watts, full load 170 watts. Temps went from 40c to 55c

Edit: This is the complete system, from the wall, and yes thats all 8 cores and 16 threads.

Oh, this is running blend. Do you want something else ? Task manager says its running 100% CPU

We have had mods continuously coming in warning people this is not an AMD thread. Now we have a mod coming in turning it once again into an AMD thread. How should I take this? Is there an AMD thread where we can talk Intel? Because this one sure isn't it anymore.

So does this mean, i actually can overclock 7900x to say 4,5GHz and use it for regular gaming, 3D rendering, Cinebench, etc... simply usual stuff, and it will be fine, even possibly cooled by 240/280mm AIO? And all the throttling, 90+C temps on CPU, 100+C temps on VRMs and ridiculous power-draw would concern me only if i tried to stress test with the Prime?

Yes.



You have an issue with moderation?
Make an MD thread. Do not post mod grievances in an active thread.
For the record, Markfw is posting as a member, not a mod, so you make
not make reference to his mod title.
Secondly, he responded to your query below.

This is a zero-point infraction.


esquared
Anandtech Forum Director

 

[DrRamtop]

It looks like SKL-X can reduce IPC without lowering clock. AVX code of course is the most affected.

That's by no means a new capability. Some PowerPC processors implemented a system that could cut the processor's throughput without lowering clockspeed, essentially injecting a CPU-wide idle cycle. I remember playing with that system on a PPC750CXe, because there were instructions to manually control the throughput loss; essentially you could choose how many idle cycles were present per 8 clocks, from 0 for full speed down to 7 for 12.5% performance.

 

[Nothingness]

That's by no means a new capability. Some PowerPC processors implemented a system that could cut the processor's throughput without lowering clockspeed, essentially injecting a CPU-wide idle cycle.

Very interesting, thanks!

But I guess it's the first x86 chip to do that.

 

[DrMrLordX]

Yes.

So how much of a power draw delta is there between a 7820x running Prime95 29.2 and CBR15 when OCed to 4.5 GHz with a standard -500 MHz AVX offset?

 

[Markfw]

We have had mods continuously coming in warning people this is not an AMD thread. Now we have a mod coming in turning it once again into an AMD thread. How should I take this? Is there an AMD thread where we can talk Intel? Because this one sure isn't it anymore.



Yes.

Someone here asked for anyone to run this software on a Ryzen, so I thought it was OK. If no one wants to know, I will ignore future requests. I was not arguing the case like others were before.