Intel Xeon Processor Supporting Up to 56 Cores & 12 Channels Memory

It's actually amazing that Intel finds a way to fit all those cores with 14nm shrink.

 

400watts TDP

Holy moly

 

Does anyone smell glue on those top chips?

 

Noooo, that's so lame. And so AMD...

I wonder how it will do against AMD's 64C/128T in terms of performance per watt.

 

There goes Intel glueing cores together after they criticized AMD for it.

 

It doesn't look like Intel's MCM is yet quite up to AMD's standards. But I guess that's natural seeing how they are hastily trying to follow in AMD's footsteps. It'll take them some more time.

 

"Innovation Day" ... OK...
On CPU side- Most of them are Tick&Tock refresh-right?, and you can tell by the ppl on stage how confident they felt and looked , sad.
Good part was the Optane DDR4 memory, hope they will come to HEDT/Desktop and we could have 64GB~256GB "RamDisk" .

 

just for info: Intel was the 1st to do it, and AMD were bashing Intel to do it, then AMD do it and Intel use the exact same argument used against the company to bash AMD.

In the end AMD or Intel, Intel or AMD... Same way of working...

 

The tic/toc logic is dead now (or the clock doesn't work well: tic tic toc toc toc tic coocoo coocoo ).
It's just a new gen and many update (can be refresh or just small change).
AMD do the same way.

 

Look-Can you guess the price of Xeon Platinum 8256?
https://ark.intel.com/content/www/u...inum-8256-processor-16-5m-cache-3-80-ghz.html

 

$7007.00 .....holy sheepshit Batman! .....Dam!

 

$7000 for a quad core, only Intel can do this, pre-order now!

I hope they didn't stacked 2 cores on top of another 2 cores just like the 56C/112T model.

 

Intel did original gluing. AMD did pretty advanced design with shared parts.

 

I love how AMD has Intel on its heels. The market needs AMD to become a proper CPU competitor in the datacenter. Intel's MCM'ing is such an ugly solution compared to AMD's Infinity fabric and chiplet approach. Intel is working on AgileX on a FPGA which is actually a bit more advanced approach than AMD's but they are way behind AMD in this front. Also, the Intel tax AMD is speaking about is very real like some posters noticed a $7,000 4-core Xeon part makes you roll your eyes even for a server part. AMD is about to mop up in the server space if what they stated about doubling the cores in the same power envelope stays true. AMD's 32 core EPYC 7601 is pulling 177 what's so their 65 core part should be about the same wattage versus Intels 400 watt 56 core part.

 

Let's not forget that AMD also supports dual-socket motherboards. 128c/256t for x86 in a single U1 server for what will likely be nearly half the price is going to be a major challenge for Intel to deal with.

Right now, Intel's saving grace is the fact that their 56-core servers perform better on average (than a dual Epyc 7601), despite having fewer cores. I think this is largely due to how AMD's memory channels are divided, with the whole UMA/NUMA situation. If AMD can get that sorted out with their 64 core Epycs, they've got an all-around winner.

 

AMD is only dreaming of reaching 10% of the server market, so I don't think the Intel execs with wooden eyes are too worried yet. Intel is probably spending as much money in manipulating the market as they spend in developing their MCM tech.

 

Right now server marketshare isn't/shouldn't be AMD's priority, especially considering how difficult it is to measure compared to desktop PCs. What they need is getting their foot in the door. CPUs of this caliber are bought in bulk and require a year (or more) of planning. Companies who buy them seek to optimize software for them, so they're committed to the platform. That being said, it's a major challenge for AMD to try taking existing Intel customers, though it's not impossible.
AMD might only see a handful of major customers per year but each one of them accounts for millions of dollars worth of product, which is a very big win for AMD and a relatively small loss for Intel.

It's worth pointing out that AMD's x86 compatibility is important, because depending on your workload, there are other theoretically better alternatives like ARM, RISC-V, or POWER. But those require you to rebuild just about all of your software, which might not be worth the other benefits.

So - having AMD "defeat" Intel is a pipe dream and it isn't what they should be focusing on. They just need to focus on themselves, and I think what will make Epyc succeed most is if they focus on attracting new customers, rather than trying to steal Intel's.

 

I honestly think AMD will do great getting datacenter customers. The big cloud providers like AWS are itching for a CPU vendor I mean Amazon was building its own CPU just to have another option. Couple that with the fact that they all have had AMD's EPYC CPU's for over a year we should see Zen2 have great datacenter adoption.

 

This is just 2 28 core dies stuck together, isn't it?