Article New Intel 28 core unlocked Fire breathing monster review !

[Markfw]

Here: https://www.anandtech.com/show/13748/the-intel-xeon-w-3175x-review-28-unlocked-cores-2999-usd

A 32 phase power with 6 CPU power plugs, and 2 24 pin power plugs just for the motherboard. At 4.4 ghz with a custom water setup, it consumes 510 watts just for the CPU !

And yes, the CPU alone is $2999. The motherboard is estimated at $1500. And a quote from the conclusion:

"On the power side of the equation, again the W-3175X comes in like a wrecking ball, and this baby is on fire. While this chip has a 255W TDP, the turbo max power value is 510W – we don’t hit that at ‘stock’ frequency, which is more around the 300W mark, but we can really crank out the power when we start overclocking.

This processor has a regular all-core frequency of 3.8 GHz, with AVX2 at 3.2 GHz and AVX-512 at 2.8 GHz. In our testing, just by adjusting multipliers, we achieved an all-core turbo of 4.4 GHz and an AVX2 turbo of 4.0 GHz, with the systems drawing 520W and 450W respectively. At these frequencies, our CPU was reporting temperatures in excess of 110ºC! This processor is actually rated with a thermal shutoff at 120ºC, well above the 105ºC we see with regular desktop processors, which shows that perhaps Intel had to bin these chips enough that the high temperature profile was required."

 

[TheGiant]

You can always high clocked epyc, it can turbo to 3,8 GHz.

3,8GHz with AVX performance of ryzen is simply too low
in fact it it slower than Broadwell xeon and faster only in memory bottlenecked situations
if Intel released a 16C unlocked xeon W with 6CH memory=win atm

the wider core of ryzen 3 will be better, but I believe the performance when I see it

 

[Zucker2k]

The most I have ever seen it draw is 350 watts. I don't remember what speed that was, but I think it was only 4.0 and it was 75c, not 110c. For 4 more cores this is way better than thats POS @ 510 watts.

As far as why I created it, it was the Anandtech front page article yesterday, so I linked it. The the "fire" came from their own conclusion, not mine.

Nice callout.

This summarizes everything nicely. A TR2 2990WX consuming 350W @ 4Ghz? Did you see this?: https://www.anandtech.com/show/13124/the-amd-threadripper-2990wx-and-2950x-review/13

"POS," right?

 

[TheELF]

3,8GHz with AVX performance of ryzen is simply too low
in fact it it slower than Broadwell xeon and faster only in memory bottlenecked situations
if Intel released a 16C unlocked xeon W with 6CH memory=win atm

the wider core of ryzen 3 will be better, but I believe the performance when I see it

Are you aware of any software that can use a wider core?
Ryzen 1 already is wider then kaby but it's still slower in pretty much anything,how will going even wider then already wider make any difference?

 

[maddie]

Are you aware of any software that can use a wider core?
Ryzen 1 already is wider then kaby but it's still slower in pretty much anything,how will going even wider then already wider make any difference?

Maybe cores are more than just # of execution units.

 

[DrMrLordX]

"POS," right?

Cute, but misleading. Yes, a 2990WX can consume 500W or more (Markfw was wrong in his assumption that the CPU would not consume more than 350W @ 4 GHz). However, you are comparing an overclocked 2990WX to a stock W-3175X. Overclock that thing and it will consume much more than 510W. For comparison:

https://www.gamersnexus.net/hwreviews/3066-intel-i9-7980xe-7960x-thermals-power-review

Here an 18c/36t 7890XE chews up a whopping 492W @ 4.5 GHz. Reduce two cores and up the clockspeed to 4.6 GHZ for the 7960X and . . . still 492W. See what's happening here? Now imagine that effect on 28 cores of the same uarch.

 

[Zucker2k]

Cute, but misleading. Yes, a 2990WX can consume 500W or more (Markfw was wrong in his assumption that the CPU would not consume more than 350W @ 4 GHz). However, you are comparing an overclocked 2990WX to a stock W-3175X. Overclock that thing and it will consume much more than 510W. For comparison:

https://www.gamersnexus.net/hwreviews/3066-intel-i9-7980xe-7960x-thermals-power-review

Here an 18c/36t 7890XE chews up a whopping 492W @ 4.5 GHz. Reduce two cores and up the clockspeed to 4.6 GHZ for the 7960X and . . . still 492W. See what's happening here? Now imagine that effect on 28 cores of the same uarch.

I don't even know why you're posting links to other chips when the link in the OP has all the numbers? Check it out: https://www.anandtech.com/show/13748/the-intel-xeon-w-3175x-review-28-unlocked-cores-2999-usd/3

Oh, don't forget 110c, obviously, you're not going to be getting the best consumption numbers. This is an impressive chip. I don't understand why all the hate when the TRs can't do better.

Edit: And that's POVRay with AVX2!

 

[DrMrLordX]

I don't even know why you're posting links to other chips when the link in the OP has all the numbers?

The hell it does. AT was stuck at 4.4 GHz. Try running one at 4.5, 4.6, or higher. See what happens then.

 

[Zucker2k]

The hell it does. AT was stuck at 4.4 GHz. Try running one at 4.5, 4.6, or higher. See what happens then.

I'm not sure if serious. You want to see the chip pushed beyond all reasonable limits in order to prove your point? Sure, would love to see how much a 2990WX consumes at 4.5GHz, too.

Edit: Also, they're using a 500w cooler. The chip is already consuming 500w @ 4Ghz meaning they were nursing a runaway thermal/power consumption scenario. No wonder they were seeing temps of 110c.

 

[DrMrLordX]

You want to see the chip pushed beyond all reasonable limits in order to prove your point?

What do you think you've done when you run a 2990WX at 4 GHz?

See my point yet?

It's all about voltage efficiency curves. Zen and Zen+ run into ugly power vs. voltage scaling past a certain point (for Zen+, it's probably at around 3.8-3.9 GHz).

 

[Zucker2k]

What do you think you've done when you run a 2990WX at 4 GHz?

See my point yet?

It's all about voltage efficiency curves. Zen and Zen+ run into ugly power vs. voltage scaling past a certain point (for Zen+, it's probably at around 3.8-3.9 GHz).

And a 28core running an AVX2 load @ 4GHz isn't near that efficiency curve, right? Moreover, when the chip is registering 110c on the thermals!!!! Intel gets no credit on this forum.

 

[DrMrLordX]

And a 28core running an AVX2 load @ 4GHz is near that efficiency curve, right? Moreover, when the chip is registering 110c on the thermals!!!! Intel gets no credit on this forum.

For Skylake-X? If you look at the lower core-count products, you'll see that it starts to get punished at 4.6-5.0 GHz. 4.4 GHz is sitting closer to its optimal efficiency curve. I would expect power to be much more tame at 4 GHz than the 4.4 GHz speed cited.

Remember the 5 GHz 28c/56t demo Intel gave months ago? The one that wowed audiences? With the chiller under the table? This is that chip, albeit probably better-binned. I'll give them credit for binning the hell out of it, which is why there are so few.

 

[Zucker2k]

For Skylake-X? If you look at the lower core-count products, you'll see that it starts to get punished at 4.6-5.0 GHz. 4.4 GHz is sitting closer to its optimal efficiency curve. I would expect power to be much more tame at 4 GHz than the 4.4 GHz speed cited.

Remember the 5 GHz 28c/56t demo Intel gave months ago? The one that wowed audiences? With the chiller under the table? This is that chip, albeit probably better-binned. I'll give them credit for binning the hell out of it, which is why there are so few.

And we're talking about a cooling capacity of 1,700w vs 500w. Also, you keep citing lower core count products, which naturally enjoy higher frequencies than higher core count products. the W-3175x comes in with 28 cores sitting next to each other (monolithic). You expect it to overclock as well as a 16 core and an 18 core? Come on, man!

 

[DrMrLordX]

. Also, you keep citing lower core count products, which naturally enjoy higher frequencies than higher core count products. the W-3175x comes in with 28 cores sitting next to each other (monolithic). You expect it to overclock as well as a 16 core and an 18 core? Come on, man!

Those cores in the W-3175X are no different than the ones in the 7890XE. With proper cooling it will clock to the same levels as any Skylake-X. Same uarch, same process. In the case of the 7890XE it's the same die. Try thinking about what I say for a moment instead of being defensive. Why do you think Intel used that chiller for their 5.0 GHz demo? They wouldn't have had to do it if voltage scaling from 4.4 GHz to 5.0 GHz was the same as the voltage scaling from 4.0 GHz to 4.4 GHz on that uarch/process. The same thing starts to happen to the 2990WX once you push it past 3.6-3.8 GHz or so. How much power do you suppose the 2990WX consumes if it runs at a steady 3.8 GHz or 3.6 GHz? Maybe if you asked @Markfw nicely he might tell you, assuming he'll stop his from folding long enough to run accurate power benchmarks. Sadly, very few reviewers seem to have run benchmarks like that (we mostly got fixed 4.0 GHz tests or PBO tests).

To reiterate, you can't take a CPU that's in a bad part of its voltage curve (2990WX @ 4.0 GHz) and compare it to a CPU that's in a decent-to-good part of its voltage curve (W-3175X @ 4.0-4.4 GHz) and pretend like you're making a useful comparison in terms of performance/watt.

The real problem here is that there are folks jumping out of the woodwork to defend a goofy halo HEDT processor that exists in vanishingly small quantity at a price that only a few people would be willing to pay. Intel wouldn't be selling this chip if they had working 10nm HEDT dice shipping today. It's a testament to how well they've done with their 14nm process as well as their failure to move on to anything else.

 

[Zucker2k]

Those cores in the W-3175X are no different than the ones in the 7890XE. With proper cooling it will clock to the same levels as any Skylake-X. Same uarch, same process. In the case of the 7890XE it's the same die. Try thinking about what I say for a moment instead of being defensive. Why do you think Intel used that chiller for their 5.0 GHz demo? They wouldn't have had to do it if voltage scaling from 4.4 GHz to 5.0 GHz was the same as the voltage scaling from 4.0 GHz to 4.4 GHz on that uarch/process. The same thing starts to happen to the 2990WX once you push it past 3.6-3.8 GHz or so. How much power do you suppose the 2990WX consumes if it runs at a steady 3.8 GHz or 3.6 GHz? Maybe if you asked @Markfw nicely he might tell you, assuming he'll stop his from folding long enough to run accurate power benchmarks. Sadly, very few reviewers seem to have run benchmarks like that (we mostly got fixed 4.0 GHz tests or PBO tests).

To reiterate, you can't take a CPU that's in a bad part of its voltage curve (2990WX @ 4.0 GHz) and compare it to a CPU that's in a decent-to-good part of its voltage curve (W-3175X @ 4.0-4.4 GHz) and pretend like you're making a useful comparison in terms of performance/watt.

The real problem here is that there are folks jumping out of the woodwork to defend a goofy halo HEDT processor that exists in vanishingly small quantity at a price that only a few people would be willing to pay. Intel wouldn't be selling this chip if they had working 10nm HEDT dice shipping today. It's a testament to how well they've done with their 14nm process as well as their failure to move on to anything else.

I'm not being defensive. Here's what you're refusing to see:
28 Cores on a monolithic die running AVX2 code at 4Ghz, and already exceeding the cooling capacity of the test cooler. Is there a more horrifying scenario for a chip? I'm amazed you think a chip running under those conditions and registering 110c is far from its efficiency curve. The stock AVX2 clocks for the W-3175X is 3.2GHz. Here, it's running at 25% above stock clocks while being undercooled.

 

[LTC8K6]

Now you can overclock with confidence as your purchase of an Intel® Xeon® X-3175X processor includes the Performance Tuning Protection Plan; allowing for a one-time replacement from damage caused when operating outside of Intel's published specification.

 

[Markfw]

Those cores in the W-3175X are no different than the ones in the 7890XE. With proper cooling it will clock to the same levels as any Skylake-X. Same uarch, same process. In the case of the 7890XE it's the same die. Try thinking about what I say for a moment instead of being defensive. Why do you think Intel used that chiller for their 5.0 GHz demo? They wouldn't have had to do it if voltage scaling from 4.4 GHz to 5.0 GHz was the same as the voltage scaling from 4.0 GHz to 4.4 GHz on that uarch/process. The same thing starts to happen to the 2990WX once you push it past 3.6-3.8 GHz or so. How much power do you suppose the 2990WX consumes if it runs at a steady 3.8 GHz or 3.6 GHz? Maybe if you asked @Markfw nicely he might tell you, assuming he'll stop his from folding long enough to run accurate power benchmarks. Sadly, very few reviewers seem to have run benchmarks like that (we mostly got fixed 4.0 GHz tests or PBO tests).

To reiterate, you can't take a CPU that's in a bad part of its voltage curve (2990WX @ 4.0 GHz) and compare it to a CPU that's in a decent-to-good part of its voltage curve (W-3175X @ 4.0-4.4 GHz) and pretend like you're making a useful comparison in terms of performance/watt.

The real problem here is that there are folks jumping out of the woodwork to defend a goofy halo HEDT processor that exists in vanishingly small quantity at a price that only a few people would be willing to pay. Intel wouldn't be selling this chip if they had working 10nm HEDT dice shipping today. It's a testament to how well they've done with their 14nm process as well as their failure to move on to anything else.

Thanks for the thought ! And it might be interesting. Except the person it is meant for thinks Intel can do no wrong, and only make the best processors, and AMD is crap, and can't do anything right, and no changing their mind, so why bother trying to prove anything ? Like trying to get blood out of a rock. It isn't going to happen. When somebody who isn't so biased asks, I will work on it.

Oh, and another aside. Aside from the fact that Intel recommends water cooling as the cooling solution STOCK where as AMD can use air, the CPU runs at 110c, where the AMD runs at 75c, its costs almost twice as much and the motherboard is like 4 times the price of a 2990wx motherboard, you can't even buy one due to the limited supply. newegg is OOS on the CPU, and NOBODY has the motherboard even listed ! This is not even a real product, just a lame attempt to beat AMD. Oh, and 2 power supplies to power it ? 6 cpu plugs and 2 24 pin plugs ????? I could go on, but why ?

 

[Zucker2k]

Now you can overclock with confidence as your purchase of an Intel® Xeon® X-3175X processor includes the Performance Tuning Protection Plan; allowing for a one-time replacement from damage caused when operating outside of Intel's published specification.

This is actually great. I wonder if it's a lifetime or limited duration warranty.

 

[Markfw]

This is actually great. I wonder if it's a lifetime or limited duration warranty.

It says right in the quote "allowing for a one-time replacement from damage caused when operating outside of Intel's published specification. ",
So its ONE TINE for a $3000 processor (if you can find one to buy somewhere)

 

[LTC8K6]

This is actually great. I wonder if it's a lifetime or limited duration warranty.

It's one time, and it's in addition to the 3 year warranty.
The plan can be purchased within the first year, but is free with the W-3175X

https://click.intel.com/tuningplan/faq

https://click.intel.com/tuningplan/purchase-a-plan

 

[DrMrLordX]

I'm not being defensive. Here's what you're refusing to see:
28 Cores on a monolithic die running AVX2 code at 4Ghz, and already exceeding the cooling capacity of the test cooler. Is there a more horrifying scenario for a chip? I'm amazed you think a chip running under those conditions and registering 110c is far from its efficiency curve. The stock AVX2 clocks for the W-3175X is 3.2GHz. Here, it's running at 25% above stock clocks while being undercooled.

How am I refusing to see that? The chip is even worse above 4.4 GHz! It is not efficient. It is not a marvel of engineering. It's a Xeon on an old process that's unlocked, with a lot of binning to get it to run at clocks below 4.5 GHz with some semblance of efficiency. Believe it or not, 4.0-4.4 GHz is a part of Skylake-X's voltage curve where it's still somewhat manageable. Somewhat. There's a reason why AT could not get the thing to OC past 4.4 GHz. That's when it starts to get real.

Yeah it'll run with greater efficiency at 3.2 GHz, but if you're running it at that speed then why are you even buying such a chip? There are already Skylake-SP Xeons that can do the same thing. And now Cascade Lake-SP as well, or soon anyway.

There are only a tiny number of people that actually need such a chip for anything. Most tasks that would fit a 28c/56t chip can be carried out just as well on a 2P Xeon Gold system, if not better.

 

[TheGiant]

Are you aware of any software that can use a wider core?
Ryzen 1 already is wider then kaby but it's still slower in pretty much anything,how will going even wider then already wider make any difference?

well apple is using the wider core and more execution units while maintaining low power with low frequency
I think the road from 4 to 5GHz is too long and complicated to have the desired benefit

I think changing the architecture completely to lower freq/better flow per MHz means lower power while performance can be very high (remember Athlon 64 rating comparing to p4)
then again fine tuning can bring it back high and we reset the circle
this 28C with high frequency is powerful but 500W, really desktop 500W ?
really whatever 500W on the CPU? we went the wrong way...

 

[TheELF]

well apple is using the wider core and more execution units while maintaining low power with low frequency
I think the road from 4 to 5GHz is too long and complicated to have the desired benefit

I think changing the architecture completely to lower freq/better flow per MHz means lower power while performance can be very high (remember Athlon 64 rating comparing to p4)
then again fine tuning can bring it back high and we reset the circle
this 28C with high frequency is powerful but 500W, really desktop 500W ?
really whatever 500W on the CPU? we went the wrong way...

Performance on a wide low freq core can only be very high if there is software that can take advantage of the additional execution units,right now there is nothing that can do this,not even one single piece of software.

 

[maddie]

Performance on a wide low freq core can only be very high if there is software that can take advantage of the additional execution units,right now there is nothing that can do this,not even one single piece of software.

Maybe you should be clearer in this.

I assume that you mean, "no single threaded piece of software". It would be a good assumption to think that SMT would improve with a wider core if the proper supporting units such as decoders, cache, etc, would also increase. So yes, there is software that would benefit from a wider core, just not a single threaded one.

 

[TheELF]

Maybe you should be clearer in this.

I assume that you mean, "no single threaded piece of software". It would be a good assumption to think that SMT would improve with a wider core if the proper supporting units such as decoders, cache, etc, would also increase. So yes, there is software that would benefit from a wider core, just not a single threaded one.

Yes this happend a couple of times with intel in the last years and nobody noticed any improvement...

 

[TheGiant]

Yes this happend a couple of times with intel in the last years and nobody noticed any improvement...

so how is it Apple doing? what is that performance miracle based on ?