Intel X299, Kaby Lake-X & Skylake-X MegaThread! FAQ and Resources

[juanrga]

7980XE pulls over 500w by itself when overclocked:

http://www.pcgamer.com/intel-core-i9-7980xe-review-an-incredibly-fast-cpu-but-gamers-dont-need-it/

Intel is rumored to be working on a nuclear fission adapter to power their new chips, rumored price tag to be in the $1,000,000.00 USD range. Overclocking will void the warranty on both parts however...caveat emptor.

(i) He is not measuring power at socket level, but power at platform level.
(ii) At stock settings the 7980XE pulls less power than TR 1950X, despite the ThreadRipper has two less cores and is much slower.
(iii) The 1950X hits a wall at around 3-9--4.0GHz. Using LN2 and pushing the chip at same clocks than OC XE, bring power consumption well above 500W. In fact already with a weak 3.9GHz overclock the 1950X platform pulls 550W. ThreadRipper is terribly inefficient and power hungry!

 

[goldstone77]

7980XE pulls over 500w by itself when overclocked:

http://www.pcgamer.com/intel-core-i9-7980xe-review-an-incredibly-fast-cpu-but-gamers-dont-need-it/

Intel is rumored to be working on a nuclear fission adapter to power their new chips, rumored price tag to be in the $1,000,000.00 USD range. Overclocking will void the warranty on both parts however...caveat emptor.

(i) He is not measuring power at socket level, but power at platform level.
(ii) At stock settings the 7980XE pulls less power than TR 1950X, despite the ThreadRipper has two less cores and is much slower.
(iii) The 1950X hits a wall at around 3-9--4.0GHz. Using LN2 and pushing the chip at same clocks than OC XE, bring power consumption well above 500W. In fact already with a weak 3.9GHz overclock the 1950X platform pulls 550W. ThreadRipper is terribly inefficient and power hungry!

The problem with current 14nm Zen was built on Samsung 14nmLPP for 3GHz operation, and is being pushed to it's architectural limitations 3.8,3.9,4.0, and 4.1GHz depending the the CPU this varies greatly. Once a particular CPU hit above normal operation frequency threshold the voltage required to increase just 100-300MHz spikes! If kept within normal operation it's very efficient, and runs cool.

 

[juanrga]

7980XE pulls over 500w by itself when overclocked:

http://www.pcgamer.com/intel-core-i9-7980xe-review-an-incredibly-fast-cpu-but-gamers-dont-need-it/

Intel is rumored to be working on a nuclear fission adapter to power their new chips, rumored price tag to be in the $1,000,000.00 USD range. Overclocking will void the warranty on both parts however...caveat emptor.

(i) He is not measuring power at socket level, but power at platform level.
(ii) At stock settings the 7980XE pulls less power than TR 1950X, despite the ThreadRipper has two less cores and is much slower.
(iii) The 1950X hits a wall at around 3-9--4.0GHz. Using LN2 and pushing the chip at same clocks than OC XE, bring power consumption well above 500W. In fact already with a weak 3.9GHz overclock the 1950X platform pulls 550W. ThreadRipper is terribly inefficient and power hungry!

The problem with current 14nm Zen was built on Samsung 14nmLPP for 3GHz operation, and is being pushed to it's architectural limitations 3.8,3.9,4.0, and 4.1GHz depending the the CPU this varies greatly. Once a particular CPU hit above normal operation frequency threshold the voltage required to increase just 100-300MHz spikes! If kept within normal operation it's very efficient, and runs cool.

I know. I predicted ~3GHz as normal clocks and a ~4GHz wall for overclocking.

It is not Intel fault that AMD does a wrong mixture of muarch and node. Intel chips are more efficient thanks to correct combination of muarch and node.

 

[goldstone77]

Intel is not done with Kabylake-X yet. A prototype of i3 dual-core processor has been spotted in China.

This is a really bizarre processor. A platform that is designed for high-performance computing might soon get a new CPU… with two cores.

The i3-7360X is a HEDT alternative to i3-7350K, which, by the way, will soon be succeeded by 8350K.

The specs of this new processor are not overwhelming. It’s only 100 MHz faster than 7350K. The turbo clock is 4.3 GHz. The TDP though, skyrockets to 112W.

According to the leaker, the i3-7360X is 1.25% faster than 7350K. The price of 7360X is expected around 1699 Yuans (220 USD), so it’s not cheap.
A side by side comparison between i3-7360X and i3-7350K:

 

[TechyInAZ]

Intel is not done with Kabylake-X yet. A prototype of i3 dual-core processor has been spotted in China.

This is a really bizarre processor. A platform that is designed for high-performance computing might soon get a new CPU… with two cores.

The i3-7360X is a HEDT alternative to i3-7350K, which, by the way, will soon be succeeded by 8350K.

The specs of this new processor are not overwhelming. It’s only 100 MHz faster than 7350K. The turbo clock is 4.3 GHz. The TDP though, skyrockets to 112W.

According to the leaker, the i3-7360X is 1.25% faster than 7350K. The price of 7360X is expected around 1699 Yuans (220 USD), so it’s not cheap.
A side by side comparison between i3-7360X and i3-7350K:

Yep saw that on Awesome Hardware. For the first time I can finally say we have a more terrible CPU in the market than the FX 9590.

 

[juanrga]

As shown in other forums, that i3 KBL-X is fake. The CPU identifier corresponds to one of current XBL-X models.

 

[juanrga]

As expected the new SKL-X models have excellent power consumption, often consuming much less power than ThreadRipper despite having more cores and higher clocks

The combination of controlled power and superb performance produces excellent efficiency ratios

7980XE is 34% more efficient than ThreadRipper 1950X even in a benchmark favorable to Zen muarch. Wow!

 

[-Fran-]

As expected the new SKL-X models have excellent power consumption, often consuming much less power than ThreadRipper despite having more cores and higher clocks

Spoiler

The combination of controlled power and superb performance produces excellent efficiency ratios

7980XE is 34% more efficient than ThreadRipper 1950X even in a benchmark favorable to Zen muarch. Wow!

Actually, I find it impressive that TR is actually able to consume less in most of them with an inferior process AND still be "up there" with them.

I'd be willing to say that, under the same process, AMD would be putting big hurt into Intel's bottom line with Zen. And I think Intel agrees with me there, hence the speed up in their releases. We all win at the end, so no complains on my part!

Cheers!

 

[juanrga]

As expected the new SKL-X models have excellent power consumption, often consuming much less power than ThreadRipper despite having more cores and higher clocks

Spoiler

The combination of controlled power and superb performance produces excellent efficiency ratios

7980XE is 34% more efficient than ThreadRipper 1950X even in a benchmark favorable to Zen muarch. Wow!

Actually, I find it impressive that TR is actually able to consume less in most of them with an inferior process AND still be "up there" with them.

I'd be willing to say that, under the same process, AMD would be putting big hurt into Intel's bottom line with Zen. And I think Intel agrees with me there, hence the speed up in their releases. We all win at the end, so no complains on my part!

Interesting remark considering that AMD precisely chose 14LPP for its efficiency focus, and considering that other companies using 14LPP are able beat Intel on efficiency metrics.

It would be also interesting academic exercise to port those Intel chips from the performance node to the efficiency process node and see how much the efficiency gap increases.

 

[-Fran-]

Interesting remark considering that AMD precisely chose 14LPP for its efficiency focus, and considering that other companies using 14LPP are able beat Intel on efficiency metrics.

It would be also interesting academic exercise to port those Intel chips from the performance node to the efficiency process node and see how much the efficiency gap increases.

Yep, that is something that would be interesting indeed.

I do believe Intel's process is better and well rounded and most of the time the problems their CPUs face are due to *other* weird decisions that are not process related, like we see with the "ceiling" of ~4Ghz for RyZen.

All in all, my remark is not far-fetched when you do look at the power consumption, even if the performance of the CPUs is not the same (I do agree Intel still has an edge over AMD), Zen is not left completely in the dust like it happened with PD and Ivy.

Cheers!

 

[8350rocks]

7980XE pulls over 500w by itself when overclocked:

http://www.pcgamer.com/intel-core-i9-7980xe-review-an-incredibly-fast-cpu-but-gamers-dont-need-it/

Intel is rumored to be working on a nuclear fission adapter to power their new chips, rumored price tag to be in the $1,000,000.00 USD range. Overclocking will void the warranty on both parts however...caveat emptor.

(i) He is not measuring power at socket level, but power at platform level.
(ii) At stock settings the 7980XE pulls less power than TR 1950X, despite the ThreadRipper has two less cores and is much slower.
(iii) The 1950X hits a wall at around 3-9--4.0GHz. Using LN2 and pushing the chip at same clocks than OC XE, bring power consumption well above 500W. In fact already with a weak 3.9GHz overclock the 1950X platform pulls 550W. ThreadRipper is terribly inefficient and power hungry!

Sure it does...with a 250W GPU...it absolutely does.

The 7980XE pulls over 500W on the CPU though, this is validated in multiple locations.

To further add to this...

The only reason SKL-X has lower idle consumption (notice you did not provide full load consumption, interestingly), is because Intel has far more aggressive core parking on their CPUs.

 

[juanrga]

7980XE pulls over 500w by itself when overclocked:

http://www.pcgamer.com/intel-core-i9-7980xe-review-an-incredibly-fast-cpu-but-gamers-dont-need-it/

Intel is rumored to be working on a nuclear fission adapter to power their new chips, rumored price tag to be in the $1,000,000.00 USD range. Overclocking will void the warranty on both parts however...caveat emptor.

(i) He is not measuring power at socket level, but power at platform level.
(ii) At stock settings the 7980XE pulls less power than TR 1950X, despite the ThreadRipper has two less cores and is much slower.
(iii) The 1950X hits a wall at around 3-9--4.0GHz. Using LN2 and pushing the chip at same clocks than OC XE, bring power consumption well above 500W. In fact already with a weak 3.9GHz overclock the 1950X platform pulls 550W. ThreadRipper is terribly inefficient and power hungry!

Sure it does...with a 250W GPU...it absolutely does.

The 7980XE pulls over 500W on the CPU though, this is validated in multiple locations.

To further add to this...

The only reason SKL-X has lower idle consumption (notice you did not provide full load consumption, interestingly), is because Intel has far more aggressive core parking on their CPUs.

I provided full load consumption figures in this post

http://www.tomshardware.co.uk/forum/id-3455797/intel-x299-kaby-lake-skylake-megathread-faq-resources/page-3.html#20223499

When paired with the same GPUs, the Threadripper platform often consumes more power than the X299 platform because AMD CPUs are more power hungry. The same happens when measuring power at the socket level.

Your pcgamer link measured 576W with the 18C overclocked at 4.1GHz. The review claims that threadripper 1950X overclocked to 3.8GHz used 422W in the same test. Let us do some computations. Pushing Threadripper up to 4.1GHz would pull ~25% more. So 530W. And adding two extra cores would increase to ~600W.

 

[goldstone77]

Intel Core i9-7960X review: It beats Threadripper, but for a price
80% more cash than a 1950X for less than 20% more performance? Tough sell.
MARK WALTON (UK) - 9/25/2017, 9:38 AM

What this means, when paired with a huge die and a TDP of 165W like in the i9-7960X, is users are saddled with CPU thermals that severely limit overclocking and require expensive cooling solutions at stock. Given the ability to overclock is much of the reason to purchase an X-series processors over a comparable Xeon, that's quite a problem. Intel recommends liquid cooling for Skylake-X, even for operation at stock speeds, but I'd go as far as saying you need 280mm liquid cooler at a minimum, with a 360mm liquid cooler being ideal (custom loops not withstanding).

If all you want to do is run the i9-7960X leaving everything on "Auto," CPU temperatures hover around 67°C under full load when paired with a 360mm liquid cooler like the BeQuiet Silent Loop. I measured a total system power draw of 520W running Rog RealBench with a GTX 1080 Ti. 67°C isn't a bad temperature, but it's higher than a stock Threadripper 1950X, which hits 64°C with a full system power draw of 451W.
Part of the problem is the X299 platform is a little over-zealous with voltage, even at stock speeds (VID is at 1.030v at stock). By how much varies from motherboard to motherboard, but you can claw back some power draw and temperature by reducing the amount of voltage going to the CPU, instead of increasing it like when overclocking.

It is possible to run the i9-7960X using "Auto" settings at 4.1GHz by simply increasing the multiplier. However, the motherboard opts to pump 1.206v into the CPU, resulting in temperatures peaking at 97°C and total system power draw hitting a whopping 687W. That's far from a useable, everyday overclock.

I found the lowest stable voltage at 4.1GHz to be 1.116v. Again, this will vary from motherboard to motherboard and chip to chip, but the result is a more reasonable 83°C peak and a total system power draw of 600W.

Ultimately, 4.1GHz at 1.116v (with a -500MHz offset for AVX) is what I'd consider a reasonable overclock for most people, without getting into custom loops or delidding, which is the complex process of removing the heat spreader and replacing the thermal material. That's a shame, because I've no doubt that had Intel used solder, higher overclocks would be possible (I got as high as 4.4GHz without crashes, but quickly encountered thermal throttling).

The good

The fastest slice of silicon going
Good gaming performance
Improved memory support
Modern complement of I/O

The bad

Not that much faster than Threadripper 1950X
Fewer PCIe lanes than the competition
Power hungry
Continual use of TIM instead of solder
High clock speeds and quiet systems out of the question without custom liquid cooling or delidding
The ugly

There's little justification for the price tag. Buy a Threadripper 1950X instead.

It is possible to run the i9-7960X using "Auto" settings at 4.1GHz by simply increasing the multiplier. However, the motherboard opts to pump 1.206v into the CPU, resulting in temperatures peaking at 97°C and total system power draw hitting a whopping 687W. That's far from a useable, everyday overclock.

And that's the 16 core vs. 16 core! Multiple reviews show the numerous issues with these higher core count Skylake-X CPU's. Even the one where they delided, applied liquid metal, and custom cooled it! The multiple reviews I post, and there are more than I posted, show similar power consumption issues with Skylake-X!


Edit:

The combination of controlled power and superb performance produces excellent efficiency ratios

Juanrga note that I am supplying links to all the benchmarks I'm postings. Can you please post links, so your benchmarks can be fact checked. Thank you.

 

[goldstone77]

Intel-Core-i9-7900X-10Core-3-3GHz-LGA-2066-140W-BX80673I97900X-Desktop-Processor
Selling on Ebay at a discount if anyone is looking to buy!
List price: $1,149.99 What does this price mean?
You save: $200.00 (17% off)
Now:US $949.99
From $45 for 24 months *

US $900.00

 

[juanrga]

Intel-Core-i9-7900X-10Core-3-3GHz-LGA-2066-140W-BX80673I97900X-Desktop-Processor
Selling on Ebay at a discount if anyone is looking to buy!
List price: $1,149.99 What does this price mean?
You save: $200.00 (17% off)
Now:US $949.99
From $45 for 24 months *

US $900.00

The i7-9000X had a launch price of $999. This is reflected in the OP of this thread, and also in reviews

http://hexus.net/tech/reviews/cpu/107017-intel-core-i9-7900x-14nm-skylake-x/

So finding an Ebay store that overprices to give the sensation of discount means nothing.

 

[goldstone77]

Intel-Core-i9-7900X-10Core-3-3GHz-LGA-2066-140W-BX80673I97900X-Desktop-Processor
Selling on Ebay at a discount if anyone is looking to buy!
List price: $1,149.99 What does this price mean?
You save: $200.00 (17% off)
Now:US $949.99
From $45 for 24 months *

US $900.00

The i7-9000X had a launch price of $999. This is reflected in the OP of this thread, and also in reviews

http://hexus.net/tech/reviews/cpu/107017-intel-core-i9-7900x-14nm-skylake-x/

So finding an Ebay store that overprices to give the sensation of discount means nothing.

Regardless the Ebay offerings at $949.99 and $900.00 are available for 5% and 10% discount if anyone wants one.

 

[8350rocks]

7980XE pulls over 500w by itself when overclocked:

http://www.pcgamer.com/intel-core-i9-7980xe-review-an-incredibly-fast-cpu-but-gamers-dont-need-it/

Intel is rumored to be working on a nuclear fission adapter to power their new chips, rumored price tag to be in the $1,000,000.00 USD range. Overclocking will void the warranty on both parts however...caveat emptor.

(i) He is not measuring power at socket level, but power at platform level.
(ii) At stock settings the 7980XE pulls less power than TR 1950X, despite the ThreadRipper has two less cores and is much slower.
(iii) The 1950X hits a wall at around 3-9--4.0GHz. Using LN2 and pushing the chip at same clocks than OC XE, bring power consumption well above 500W. In fact already with a weak 3.9GHz overclock the 1950X platform pulls 550W. ThreadRipper is terribly inefficient and power hungry!

Sure it does...with a 250W GPU...it absolutely does.

The 7980XE pulls over 500W on the CPU though, this is validated in multiple locations.

To further add to this...

The only reason SKL-X has lower idle consumption (notice you did not provide full load consumption, interestingly), is because Intel has far more aggressive core parking on their CPUs.

I provided full load consumption figures in this post

http://www.tomshardware.co.uk/forum/id-3455797/intel-x299-kaby-lake-skylake-megathread-faq-resources/page-3.html#20223499

When paired with the same GPUs, the Threadripper platform often consumes more power than the X299 platform because AMD CPUs are more power hungry. The same happens when measuring power at the socket level.

Your pcgamer link measured 576W with the 18C overclocked at 4.1GHz. The review claims that threadripper 1950X overclocked to 3.8GHz used 422W in the same test. Let us do some computations. Pushing Threadripper up to 4.1GHz would pull ~25% more. So 530W. And adding two extra cores would increase to ~600W.

So, is that why Jayz2cents did a test review and had to swap an 800W PSU out to a 1500W PSU because the 7980XE test bench drew more power than the 800W PSU could produce?

Just trying to get clarification.

 

[juanrga]

So, is that why Jayz2cents did a test review and had to swap an 800W PSU out to a 1500W PSU because the 7980XE test bench drew more power than the 800W PSU could produce?

Just trying to get clarification.

So if tomorrow Jay or any similar youtuber joker decides to publish the nonsense that the CPU alone consumes 3000W, we suddenly take his words as gospel and ignore all real reviews. Right? Specially all those reviews that demonstrate that 7980XE consumes significantly less power than TR 1950X.

 

[-Fran-]

So, is that why Jayz2cents did a test review and had to swap an 800W PSU out to a 1500W PSU because the 7980XE test bench drew more power than the 800W PSU could produce?

Just trying to get clarification.

So if tomorrow Jay or any similar youtuber joker decides to publish the nonsense that the CPU alone consumes 3000W, we suddenly take his words as gospel and ignore all real reviews. Right? Specially all those reviews that demonstrate that 7980XE consumes significantly less power than TR 1950X.

That logic swings both ways, Juan.

I already said that Toms and Anands had issues with their i9-7980XE and even more, Anands reported the CPU sucked way more than standard.

Don't put a blind fold to reality yourself either. MoBo+CPU combos have issues, but they are a platform with issues none the less. When big site reviews have issues, you can really expect the end user to have them as well in one way or another.

 

[Jsimenhoff]

Join us in the Laptop General Discussion Forumhttp://www.tomshardware.com/forum/id-3533785/ama-spectre-laptop-giveaway.htmlhttp://www.tomshardware.com/forum/id-3533785/ama-spectre-laptop-giveaway.html for a very special AMA and Spectre x360 giveaway with our friends from HP.

 

[juanrga]

So, is that why Jayz2cents did a test review and had to swap an 800W PSU out to a 1500W PSU because the 7980XE test bench drew more power than the 800W PSU could produce?

Just trying to get clarification.

So if tomorrow Jay or any similar youtuber joker decides to publish the nonsense that the CPU alone consumes 3000W, we suddenly take his words as gospel and ignore all real reviews. Right? Specially all those reviews that demonstrate that 7980XE consumes significantly less power than TR 1950X.

That logic swings both ways, Juan.

I already said that Toms and Anands had issues with their i9-7980XE and even more, Anands reported the CPU sucked way more than standard.

Don't put a blind fold to reality yourself either. MoBo+CPU combos have issues, but they are a platform with issues none the less. When big site reviews have issues, you can really expect the end user to have them as well in one way or another.

Anand had problems with the i9-7980XE, but not with the i9-7960X. Anand got that the i9-7960X is more efficient than both ThreadRipper models. Even when had issues with the 7980XE, the power measured was only 27W above the expected. Anand measured less than 200W for the full package.

Even considering outliers, the whole set of reviews show that on average the new i9 models consume less power and are more efficient than TR models. Comparing the power consumed by OC 7980XE with the power consumed by stock TR is nonsense. Either we compare stock vs stock or we compare OC vs OC.

 

[juanrga]

AVX512 HPC benchmarks

https://www.pugetsystems.com/labs/hpc/Intel-Core-i9-7900X-and-7980XE-Skylake-X-Linux-Linpack-Performance-1059/

 

[connerweatherly17]

What is the best motherboard for a 7800x? I have been trying to look around and I don't want to spend a hole bunch of money (not over 350) but all of them I look at have a lot of bad reviews, is there an atx board that will actually work well without a bunch of issues?

 

[uguv]

Unless you need the pci lanes, avx512 or quad channel ram provided by the skylake-x platform, I think you're better off with an i7-8700K.

 

[8350rocks]

Spoiler

So, is that why Jayz2cents did a test review and had to swap an 800W PSU out to a 1500W PSU because the 7980XE test bench drew more power than the 800W PSU could produce?

Just trying to get clarification.

So if tomorrow Jay or any similar youtuber joker decides to publish the nonsense that the CPU alone consumes 3000W, we suddenly take his words as gospel and ignore all real reviews. Right? Specially all those reviews that demonstrate that 7980XE consumes significantly less power than TR 1950X.

That logic swings both ways, Juan.

I already said that Toms and Anands had issues with their i9-7980XE and even more, Anands reported the CPU sucked way more than standard.

Don't put a blind fold to reality yourself either. MoBo+CPU combos have issues, but they are a platform with issues none the less. When big site reviews have issues, you can really expect the end user to have them as well in one way or another.

Anand had problems with the i9-7980XE, but not with the i9-7960X. Anand got that the i9-7960X is more efficient than both ThreadRipper models. Even when had issues with the 7980XE, the power measured was only 27W above the expected. Anand measured less than 200W for the full package.

Even considering outliers, the whole set of reviews show that on average the new i9 models consume less power and are more efficient than TR models. Comparing the power consumed by OC 7980XE with the power consumed by stock TR is nonsense. Either we compare stock vs stock or we compare OC vs OC.

The rest of the reviews out there show otherwise...curious to see such a difference in results.