Intel Unveils 28-core/56-thread HEDT Processor

[iO]

Ha, they're in full panic mode, "Please don't buy Threadripper 2000s, we will launch something at some time this year..."

I wonder how much power that thing consumed if a 7980xe @ 5Ghz already pulls 600+Watts from the wall.

 

[Silent_Scone]

Cooled by tears from AMD, brought by pubg idiots.

It's bought, not brought. At £2,500 at the least I doubt the last part of your post will be right, either.

 

[T4C Fantasy]

did anyone notice intel uses an AMD gpu in cinebench? xD

 

[jabbadap]

Yep , 32 core 64 thread Threadripper 2 incoming

Uhm not likely, by the time frame it have been sampling TR2 will most probably be based on 12nm Zen+ cores, thus 16 core for x399 platform is max.

7nm Zen2 will be all about adding cores among the other things, but that is 3rd gen ryzens including TR3.

 

[champsilva]

Ha, they're in full panic mode, "Please don't buy Threadripper 2000s, we will launch something at some time this year..."

I wonder how much power that thing consumed if a 7980xe @ 5Ghz already pulls 600+Watts from the wall.

TR2 wont have 32c.

 

[Kaotik]

did anyone notice intel uses an AMD gpu in cinebench? xD

Of course they use AMD GPU - that thing obviously doesn't have integrated GPUs and Intel has good relations with AMD and bad with NVIDIA.
(similar to this, first Cannon Lakes which have their GPU disabled are all shipping with AMD GPUs bundled on the side, be it NUC or laptop)

 

[Vya Domus]

Something ain't right about all of those cores running at 5 Ghz.

TR2 will most probably be based on 12nm Zen+ cores, thus 16 core for x399 platform is max.

How did you conclude that ? There is nothing about the 12nm process that would prevent them from enabling all 4 dies.

 

[xkm1948]

Uhm not likely, by the time frame it have been sampling TR2 will most probably be based on 12nm Zen+ cores, thus 16 core for x399 platform is max.

7nm Zen2 will be all about adding cores among the other things, but that is 3rd gen ryzens including TR3.

Marking your post and I will come back to this later. To see who wins.

More pics:

 

[CrAsHnBuRnXp]

someone give me two in a server motherboard so I can use it for BOINC.

 

[iO]

Uhm not likely, by the time frame it have been sampling TR2 will most probably be based on 12nm Zen+ cores, thus 16 core for x399 platform is max.

7nm Zen2 will be all about adding cores among the other things, but that is 3rd gen ryzens including TR3.

There is a rumor that '12nm' Zen+ is pretty much the same B2 stepping which Epycs are built upon for a year now.
And nothing really stops them from binning a single socket Epyc into a special 32 core TR Titan XTX XE CPU just to piss off Intel a little bit more.

 

[jabbadap]

Something ain't right about all of those cores running at 5 Ghz.

How did you conclude that ? There is nothing about the 12nm process that would prevent them from enabling all 4 dies.

And in which motherboard would you put that all 4 dies enabled processor? And if your answer is x399 then it would be badly crippled by running half the memory channels trough infinity fabric.

 

[cadaveca]

5 ghz on 28 cores, is impressive, but the cooling needed for that is not practical for every day use.

So if its a skylake-x based cpu, a new i9 model then?

No matter what, knowing intel prices. This thing gonna be very exspensive.

Just to be clear all 28cores were @5GHz & not using liquid helium for cooling?

I agree that the TDP and power consumption will be high, it will need a new much larger cooling solution too.

Well guys, you know that Intel cores at the high-end of performance use about 8-12W on average? So we are talking about 300W, which people have been cooling on HEDT for like a decade without any issues. Heck, let's look at nearly any high-end GPU... and sure, the power density might seem to be higher on a GPU, but with the large size of these XCC dies, that's not really that true any more.

Now, I have both i9-7980XE and 1950X running on my desk now as I am working on platform analysis for various reasons. Guess which of these two CPUs pulls more power at stock? and on OC? Intel can EASILY do this, but AMD will have more issues matching Intel here, so likely they'll just have everything priced accurately, which they have had since Ryzen launched, so they'll do just fine. It's cool to have more cores if you need them, and clearly there is some need, or Intel would not bother. Both Intel and AMD seem really focused on releasing products that their customers are asking for, rather than releasing stuff and trying to shove it down everyone's throat, and that's extremely good for us as enthusiasts.

 

[T4C Fantasy]

Well guys, you know that Intel cores at the high-end of performance use about 8-12W on average? So we are talking about 300W, which people have been cooling on HEDT for like a decade without any issues. Heck, let's look at nearly any high-end GPU... and sure, the power density might seem to be higher on a GPU, but with the large size of these XCC dies, that's not really that true any more.

Now, I have both i9-7980XE and 1950X running on my desk now as I am working on platform analysis for various reasons. Guess which of these two CPUs pulls more power at stock? and on OC? Intel can EASILY do this, but AMD will have more issues matching Intel here, so likely they'll just have everything priced accurately, which they have had since Ryzen launched, so they'll do just fine. It's cool to have more cores if you need them, and clearly there is some need, or Intel would not bother. Both Intel and AMD seem really focused on releasing products that their customers are asking for, rather than releasing stuff and trying to shove it down everyone's throat, and that's extremely good for us as enthusiasts.

run some cinebench and post the score in my thread xD
https://www.techpowerup.com/forums/threads/post-your-cinebench-score.213237/

 

[LiveOrDie]

I hope you're joking, we've been hearing this for what 3 decades now?So PUBG is for idiots?

It's bought, not brought. At £2,500 at the least I doubt the last part of your post will be right, either.

If there was a poop emoji i would use it.

 

[efikkan]

Well guys, you know that Intel cores at the high-end of performance use about 8-12W on average? So we are talking about 300W, which people have been cooling on HEDT for like a decade without any issues. Heck, let's look at nearly any high-end GPU... and sure, the power density might seem to be higher on a GPU, but with the large size of these XCC dies, that's not really that true any more.

Now, I have both i9-7980XE and 1950X running on my desk now as I am working on platform analysis for various reasons. Guess which of these two CPUs pulls more power at stock? and on OC? Intel can EASILY do this, but AMD will have more issues matching Intel here, so likely they'll just have everything priced accurately, which they have had since Ryzen launched, so they'll do just fine. It's cool to have more cores if you need them, and clearly there is some need, or Intel would not bother. Both Intel and AMD seem really focused on releasing products that their customers are asking for, rather than releasing stuff and trying to shove it down everyone's throat, and that's extremely good for us as enthusiasts.

Intel do already have:
Xeon Platinum 8180 28 cores at 2.5 GHz base, 3.2-3.8 GHz boost at 205W
Xeon Platinum 8176 28 cores at 2.1 GHz base, 2.8-3.8 GHz boost at 165W
And there is a chance the refreshed Cascade Lake-X will be slightly better than this, so cooling should be no issue at stock.
(BTW don't look at the price of those Xeons…)

 

[TheoneandonlyMrK]

Intel do already have:
Xeon Platinum 8180 28 cores at 2.5 GHz base, 3.2-3.8 GHz boost at 205W
Xeon Platinum 8176 28 cores at 2.1 GHz base, 2.8-3.8 GHz boost at 165W
And there is a chance the refreshed Cascade Lake-X will be slightly better than this, so cooling should be no issue at stock.
(BTW don't look at the price of those Xeons…)

Price, and construction cost are also what makes this a diamond unicorn like consumer titan V's that some expect to happen.

If it did happen the cost will be next level.
They might make a whole new teir above Hedt ,maybe HedtX for it i suppose.

Make it so Intel , and take less time with it then 10nm.

I really can't see it happening as they are saying , it's going to need a new consumer level board to fit that 700mm die in IMHO , likely with two memory channels cut.
And if it ships at 5ghz all core boost id be speachless i assure you , truly.

And after a mild browse i found this statement from wccf.


"Intel’s 28 Core Monster Processor Was Running on an LGA 3647 Platform With Hexa Core Memory Support"


So ,a new platform entirely, possibly six channel memory too , I'd wager they just demoed a overclocked xeon and are in the process of making this happen on a pivot to what they just heard of tr2 or zen3 s performance ie vaporware that might appear within 6months to 2 years.
Still thats some saving time there.

 

[Vya Domus]

Well guys, you know that Intel cores at the high-end of performance use about 8-12W on average? So we are talking about 300W, which people have been cooling on HEDT for like a decade without any issues.

I'm pretty sure everyone knows very well how much this chip would draw at stock. The bit that got people's attention is the 5 Ghz overclock , the 7980XE draws nearly 500W at 4.5 Ghz I am sure you can easily infer from that how feasible would it be to cool and power a CPU with 50% more cores and higher clocked.

Now, I have both i9-7980XE and 1950X running on my desk now as I am working on platform analysis for various reasons. Guess which of these two CPUs pulls more power at stock? and on OC? Intel can EASILY do this, but AMD will have more issues matching Intel here.

Interestingly enough , in nearly all reviews that I have seen the 7980XE draws a fair bit more power , granted it's a bit faster as well so I don't know what you are trying to say with that. If anything it's Intel who will face way more difficulties trying to keep these monolithic giants within certain power envelops.

 

[cadaveca]

I'm pretty sure everyone knows very well how much this chip would draw at stock. The bit that got people's attention is the 5 Ghz overclock , the 7980XE draws nrealy 500W at 4.5 Ghz I am sure you can easly infer from that how feasible would it be to cool and power a CPU with 50% more cores and higher clocked.

Um, no. My 7890XE DOES NOT pull 500W. That's hilarious.

Here's my results @ 4.4 GHz:

https://www.techpowerup.com/reviews/ASRock/X299_Professional_Gaming_i9_XE/11.html

4.6 GHz:

https://www.techpowerup.com/reviews/ASRock/X299_Extreme4/11.html

nowhere near 500W. Sorry.

Interestingly enough , in nearly all review that I have seen the 7980XE draws a fair bit more power , granted it's a bit faster as well so I don't know what you are trying to say with that.

Max I got out of my 1950X:

https://www.techpowerup.com/reviews/ASRock/X399M_Taichi/11.html



Like I mean, this isn't new stuff; it's been posted here on TPU for a while now. And it's not exactly surprising that my results differ from everyone else because they always do, because I measure power over the 8-pin (which is true CPU power) and not from the wall like everyone else does. Been doing it this way for YEARS now.

I'm not TRYING to say anything. I'm stating facts that were already disclosed if you read OUR reviews. So yeah, I guess that everyone DOES NOT KNOW what these CPUs pull, because clearly you don't...

And boy, if my next board review gets posted, it's going to cause yet another shit storm. I specifically waited with these reviews until other sites were done because I knew I was going to show a different perspective. You know, I got boards specifically designed for the 7890XE.... and for the 1950X.

At stock, Intel's 165W chip does better. Overclocked is another matter, but this is partially deflated on AMD's side because they cannot scale up the speed. No big deal. In fact, I was kind of hinting at the same thing you're suggesting, but because of a lack of playing with this hardware directly, I think you've made some assumptions about this hardware that give you the wrong impression about how these chips actually work. Could I make my 7980XE pull 500W? You bet. Can you guess the speed required to do so? I'ts FAR above 4.5 GHz.

 

[R0H1T]

There is a rumor that '12nm' Zen+ is pretty much the same B2 stepping which Epycs are built upon for a year now.
And nothing really stops them from binning a single socket Epyc into a special 32 core TR Titan XTX XE CPU just to piss off Intel a little bit more.

Did everyone forget the TR delidding video by Der8auser? Now looks a good time as any to unleash the kraken 32 core beast

Um, no. My 7890XE DOES NOT pull 500W. That's hilarious.

Here's my results @ 4.4 GHz:

https://www.techpowerup.com/reviews/ASRock/X299_Professional_Gaming_i9_XE/11.html

4.6 GHz:

https://www.techpowerup.com/reviews/ASRock/X299_Extreme4/11.html

nowhere near 500W. Sorry.



Max I got out of my 1950X:

https://www.techpowerup.com/reviews/ASRock/X399M_Taichi/11.html



Like I mean, this isn't new stuff; it's been posted here on TPU for a while now. And it's not exactly surprising that my results differ from everyone else because they always do, because I measure power over the 8-pin (which is true CPU power) and not from the wall like everyone else does. Been doing it this way for YEARS now.

I'm not TRYING to say anything. I'm stating facts that were already disclosed if you read OUR reviews. So yeah, I guess that everyone DOES NOT KNOW what these CPUs pull, because clearly you don't...

And boy, if my next board review gets posted, it's going to cause yet another shit storm. I specifically waited with these reviews until other sites were done because I knew I was going to show a different perspective. You know, I got boards specifically designed for the 7890XE.... and for the 1950X.

At stock, Intel's 165W chip does better. Overclocked is another matter, but this is partially deflated on AMD's side because they cannot scale up the speed. No big deal. In fact, I was kind of hinting at the same thing you're suggesting, but because of a lack of playing with this hardware directly, I think you've made some assumptions about this hardware that give you the wrong impression about how these chips actually work. Could I make my 7980XE pull 500W? You bet. Can you guess the speed required to do so? I'ts FAR above 4.5 GHz.

I'm guesstimating the 28 cores @5Ghz should consume at least 500W, let me reiterate ~ guesstimate. It could top 600W even if they are cherry picked samples, remember it'll still need the absolute best WC setups at that speed. I'm predicting lots of boards/PS going up in smoke should buyers push their OC too hard.

 

[Vya Domus]

And it's not exactly surprising that my results differ from everyone else because they always do, because I measure power over the 8-pin (which is true CPU power) and not from the wall like everyone else does. Been doing it this way for YEARS now.

Uhm , that shouldn't change anything in terms of the distribution you get ranking these CPUs. If the power measured from the 8-pin ends up being higher so will be the power measured at the wall , it's not like there will be a void somewhere else randomly eating up all power screwing with the distribution.

You said it yourself , your results differ from everyone else , as someone looking at these things I am definitely more inclined to pay attention to results that turn up more often in a more consistent manner.

 

[cadaveca]

Uhm , that shouldn't change anything in terms of the distribution you get ranking these CPUs. If the power measured from the 8-pin ends up being higher so will be the power measured at the wall , it's not like there will be a void somewhere else eating up all power.

You said it yourself , your results differ from everyone else , as someone looking at these things I am definitely more inclined to pay attention to results that turn up more often in a more consistent manner.

However, the difference between TPU and all other sites is that we all do this for fun, to help enthusiasts, not for money (note the lack of ads). That's why we OFTEN have different results. It's not just me either... we simply DO NOT CARE who is better or not, because we aren't being paid to say ANYTHING, and aren't paying our bills doing reviews. We test, post results, and you guys decide what those results mean. I simply DO NOT CARE what's best. What I do care about is making sure the information out there is ACCURATE.

there's also those guys dropping reviews right after they get hardware, which commonly means they only played with the hardware for 2-3 days, but I take at least 2-3 weeks, which is why my reviews lag behind everyone else's... I'm exploring far more. I'm not rushing reviews out to get paid.. I'm taking my time to make sure my results are accurate and repeatable. I spend 100's of hours testing every single item I review. When this next board review gets posted (if it does, because it's that controversial), look at that review, and then look at all the other reviews of the same item, and tell me that those other places are always accurate...

As to your "power void".. it's drives, memory, board, GPUs, fans, PSU loses...15% of most other numbers are PSU efficiency loses, for instance. say just 10%... that 500W turns to 450. Add in another 100 for all that other hardware, and you have pretty much my result. So you were saying...? My point is that unless you isolate what you're testing, you're not giving accurate results. those reviews reporting power from the wall, are not accurately reporting CPU power use. Full system use, sure, but that has so many other factors that come into paly that I truly feel using system power use numbers instead of actual CPU power is doing the entire industry a disservice. I mean, after all, a clamp meter that will allow you to monitor this is way under $100 these days, as is an oscilloscope and any of the other electrical testing hardware a reviewer might need to do this right.

I will stand by my remark a while ago that if a reviewer doesn't see the value in that small expense in order to deliver accurate information to their readers, then I question anything they write. While consistency is nice, consistently being wrong isn't good at all, but hey, if that's what you like is to read bullshit and half-truths and marketing regurgitation, then be my guest.

 

[Patriot]

Marking your post and I will come back to this later. To see who wins.

More pics:

Do you have more pictures? Seem to be a lot of missing power plugs... wonder if this is a recorded demo under LN2

 

[Vya Domus]

So you were saying...?

Actually what I was saying initially didn't necessarily had to do with any of that. It all stems from Intel showing this 28 core chip running at 5Ghz , that thing will be insanely hot and power hungry and people were questioning how feasible would it be to cool this thing down or maybe even powering it. You said Intel cores are around 10W parts , well that thing surely wont follow those numbers in the way they portrayed it.

I think they are mostly putting on a show rather than showcasing a new product being used effectively and practically.

 

[cadaveca]

Actually what I was saying initially didn't necessarily had to do with any of that. It all stems from Intel showing this 28 core chip running at 5Ghz , that thing will be insanely hot and power hungry and people were questioning how feasible would it be to cool this thing down or maybe even powering it. You said Intel cores are around 10W parts , well that thing surely wont follow those numbers in the way they portrayed it.

I think they are mostly putting on a show rather than showcasing a new product being used effectively and practically.

I think it will for stock, but of course, nto for 5 GHz. You see, I mentioned that 10W number because when you get a bunch of these chips and test, it seems that this is some sort of internal limit that Intel is using to decide clockspeeds of all these parts. The fact they mention 5 GHz says that they aren't power limiting the chips tho, so those that want to pay a bit more for more exotic cooling to get more out of these chips can, but it is also side-ways blow to AMD because their CPUs are at their limits already (like the 2700X that performs worse with manual overclocking at times).

You've got to read between the lines here is what I am really saying, but if I just said that without giving any context, it wouldn't make any sense. Intel just said "we can match AMD's core count, and scale up higher, if you like." Do keep in mind, intel only makes CPUs these days, not boards, so it's up to the ODMs to decide whether we see this in retail or not. After all, that's what Computex is really about.. the ODMs showing their stuff to stores and partners for future launches.

 

[jabbadap]

Do you have more pictures? Seem to be a lot of missing power plugs... wonder if this is a recorded demo under LN2

That is actually Asus showcasing ROG Dominus, not this topics intel's PC. I'm not sure if it's even same CPU, and yeah that Asus has obviously lga3647 socket(Pure size+6-channel memory).

I think it will for stock, but of course, nto for 5 GHz. You see, I mentioned that 10W number because when you get a bunch of these chips and test, it seems that this is some sort of internal limit that Intel is using to decide clockspeeds of all these parts. The fact they mention 5 GHz says that they aren't power limiting the chips tho, so those that want to pay a bit more for more exotic cooling to get more out of these chips can, but it is also side-ways blow to AMD because their CPUs are at their limits already (like the 2700X that performs worse with manual overclocking at times).

You've got to read between the lines here is what I am really saying, but if I just said that without giving any context, it wouldn't make any sense. Intel just said "we can match AMD's core count, and scale up higher, if you like." Do keep in mind, intel only makes CPUs these days, not boards, so it's up to the ODMs to decide whether we see this in retail or not. After all, that's what Computex is really about.. the ODMs showing their stuff to stores and partners for future launches.

You are talking about Skylake X aren't you. This one still might have that ~10W power limit, but with different process(14nm+ vs 14nm++) clock roof for to reach that 10W has risen.