Discussion Intel current and future Lakes & Rapids thread

[Tigerick]

I don't think it's unrealistic to say that Intel 7 (current) aka 10nm+++++ ~ TMSC 7.

But TMSC 5 is a pretty large improvement over TMSC 7, sure power has gone up but clocks have increased dramatically.

I'll be impressed if Intel 4 is better than TMSC 5. Actually let me rephrase that. I'm be impressed if they can do it in the next year. I'm sure they'll get there but if it takes 4 years who cares?

Actually, ES1 of Meteor Lake can hit all cores of 3.6GHz which is slightly slower than 4GHz of Phoenix based on N4 process. Of course, it is not a direct comparison as one is monolithic design another is tile based and I suspect Meteor Lake tCPU is limited by 25W TDP. Anyhow, below silde from last year actually pretty helpful on my calculation of power per core numbers.



And Intel claimed 15% performance per watt on 20A process, we could have rough idea on how much speed improvements based on above slide.

 

[Geddagod]

Actually, ES1 of Meteor Lake can hit all cores of 3.6GHz which is slightly slower than 4GHz of Phoenix based on N4 process. Of course, it is not a direct comparison as one is monolithic design another is tile based and I suspect Meteor Lake tCPU is limited by 25W TDP. Anyhow, below silde from last year actually pretty helpful on my calculation of power per core numbers.

View attachment 77934

And Intel claimed 15% performance per watt on 20A process, we could have rough idea on how much speed improvements based on above slide.

This is based on an industry standard core, prob some random ARM core, not RWC itself.

 

[Exist50]

What a load of crap.
in a compute dense rack you can go 4 sockets an RU , people most certainly care about 88kw of CPU vs 52kw*. That is significate power and cooling works and that's just CPU budget .
Even if you go for something more sane like 2 P an RU 44kw vs 26kw is still stupid bad.

i buy lots of servers for NFV purpose i can about maximum number of cores (HT off) with maximum clock rate with scalar / narrow SIMD widths , biggest SIMD user will be TLS decrypt/ encrypt .

Most server workload will never care about AVX-512 or AMX. So unless your telling orgs not to buy intel unless they need AMX really because Zen4 does AVX-512 just fine.

* 48ru with 4ru of switching 44ru of scale out compute , 6 x2.5inch drives a server , 1/2 width 2P server.

It sounds like for your purposes, something like Sierra Forest would be ideal. Tons of cores and a focus on power efficiency and integer performance.

It might be the power of hindsight, but it seems that Intel would have been much better off launching a ~200 core Gracemont part instead of the SPR that we got. Sure it wouldn't have the fancy AI extensions, but it only makes sense for focus on those niches after you have something competitive for the rest of the market.

 

[Exist50]

Its $34K just on the CPUs alone.

View attachment 77918

Those prices are essentially made up. No one buying significant volume pays list price for CPUs. Though what they actually pay is not publicly known.

 

[nicalandia]

Those prices are essentially made up. No one buying significant volume pays list price for CPUs. Though what they actually pay is not publicly known.

The Discounts are for Coorporate not your Average Joe trying to buy a Workstation. Good Luck Finding A 2S 8490H for 25K

 

[Exist50]

High bandwidth/core... a reference to the increased L2 per core for RWC vs GLC? Maybe? Don't think so because GLC server had 2MB of L2, as does RWC Client, but I suppose it's possible RWC L2 increases in server as well? Could be referring to the potential increased amount of L1i per core as well, RWC supposedly bumps that up in comparison to GLC. Lastly, this could also be referring to an increase in L3 cache. EMR is rumored to increase L3 cache quite substantially per core, so I would be surprised if RWC didn't carry that over to GNR, or atleast higher amounts per core than SPR.

My first guess would be "bandwidth per core" refers to the core to fabric interface, more so than the caches. Anyone keep track of what that has been for the last few gens?

 

[Exist50]

The Discounts are for Coorporate not your Average Joe trying to buy a Workstation. Good Luck Finding A 2S 8490H for 25K

The majority of the workstation market is corporate clients. Your Average Joe doesn't have any reason to buy one.

 

[ashFTW]

Sure, dream on...

You too dude! My estimate is based on my personal experience and the the historical discounts our university has gotten with Lenovo and Dell. Do you have any experience with such purchases, other than knowledge of the list price??

 

[Exist50]

I am glad to see this post from you. i think you missed to note Sandra Rivera's pervious role in managing network and edge, its Intel's best business segment (at least by growth/decline measures).

I think credit for that better goes to Gene Scuteri. Apparently the networking silicon team was so successful that they were commandeered to work on GNR for a while, with Scuteri briefly put in charge of the project before their umpteenth reorg split them back up.

 

[ashFTW]

Unless you are expecting to use AMX, why bother and just roll with Genoa?

I don’t expect to be using AMX, directly at least. And I haven’t evaluated Genoa yet, as it’s not yet available in workstation form factor from Lenovo or Dell.

 

[nicalandia]

You too dude! My estimate is based on my personal experience and the the historical discounts our university has gotten with Lenovo and Dell. Do you have any experience with such purchases, other than knowledge of the list price??

Sure, I will wait for you to post a Screenshot of the quote you get soon. Let us know just how low Intel is given those away.

 

[nicalandia]

I think this is rather misleading. Those numbers look like quad channel for that speed RAM.

Memory Channel does not have much saying in L3 or Memory access latencies. For example Desktop Alder Lake dont even have Quad channel and the latencies are pretty good

I was pointing out the RAM access latencies, they are very high on The Xeon W9 3495X

 

[ashFTW]

Sure, I will wait for you to post a Screenshot of the quote you get soon. Let us know just how low Intel is given those away.

Quotes are confidential. You would know that if you ever dealt with such things. i am done talking with you. Waste of energy

 

[nicalandia]

Quotes are confidential. You would know that if you ever dealt with such things.

Keep dreaming Troll

 

[ashFTW]

Keep dreaming Troll

Reported and blocked

 

[IntelUser2000]

@nicalandia In this case, you are at fault. Please stop it.

Intel 3 is definitely not just a +. The 18% number is better than what they said for SF and ESF. 10nm to 10nm SF equals to 14nm to 14nm+++.

20A is 15%. 18A is only 10% but that's still a very good gain.

Intel struggles in low perf at low voltages though. Whether this is an issue with their nodes, or their architecture designs, remains to be seen.

Nah, that's more due to architecture.

Process does play too. Intel has been traditionally extremely strong on the transistor drive current. Easily 1-2 generations ahead of TSMC.

But they don't get that for free.

What they are behind on is density and/or leakage. It's just a tradeoff. Density is the big part they sacrifice. If you ignore that few of their low power processes offer exceptional drive current as well.

 

[nicalandia]

@nicalandia In this case, you are at fault. Please stop it.

I agree.

I am Sorry for my beheavior and I ask for forgiveness to asFTW

 

[nicalandia]

These are up for sale in Asia, lets hope we see benchmarks soon...

 

[Exist50]

Nah, that's more due to architecture.

Process does play too. Intel has been traditionally extremely strong on the transistor drive current. Easily 1-2 generations ahead of TSMC.

It's both. Right now, their architectures are weak, so they're forced to pump more voltage in to make them competitive in performance. Supporting higher voltage, both from a design and particularly process perspective, requires sacrifices at low voltage/power. And add on top a culture that historically did not value low-V performance, and you've got the mess Intel's in.

Intel 3 is definitely not just a +. The 18% number is better than what they said for SF and ESF. 10nm to 10nm SF equals to 14nm to 14nm+++.

20A is 15%. 18A is only 10% but that's still a very good gain.

I'll believe these numbers when I see them in silicon. By all indications, 20A/18A will be more like competitors for TSMC's N3 family. Which would still be a massive improvement for Intel, but still.

 

[Hulk]

I'm still not sure what Intel 4 has to do with ML if it's gonna be TMSC 5 for the compute tiles?

 

[Exist50]

I'm still not sure what Intel 4 has to do with ML if it's gonna be TMSC 5 for the compute tiles?

Where's that coming from? N5 is only rumored for the GPU.

 

[Hulk]

Where's that coming from? N5 is only rumored for the GPU.

And my psychotic episode is over. Thank you.

 

[A///]

Sorry have to agree with Mr. @ashFTW here. University research labs will get discounts from the likes of Intel, AMD, NVidia etc. As will large OEMs such as HP, Dell, Lenovo. No OEM is paying tray price either. Apple will practically give away their computers, if you can call them that now, to schools.

Anyway looking forward to Intel's new workstation parts putting a fire under AMD's arse. Not long after Intel announced these parts Threadripper discussion began to gain traction. The last threadripper was a big middle finger to AMD pro customers. It's good to see AMD realizing they can't rest on their laurels.

Not as long as Pat "Gunslinger" Gelsinger is in charge at Intel.

 

[Henry swagger]

Keep dreaming Troll

Lol why you so mad good sir

 

[Henry swagger]

Sorry have to agree with Mr. @ashFTW here. University research labs will get discounts from the likes of Intel, AMD, NVidia etc. As will large OEMs such as HP, Dell, Lenovo. No OEM is paying tray price either. Apple will practically give away their computers, if you can call them that now, to schools.

Anyway looking forward to Intel's new workstation parts putting a fire under AMD's arse. Not long after Intel announced these parts Threadripper discussion began to gain traction. The last threadripper was a big middle finger to AMD pro customers. It's good to see AMD realizing they can't rest on their laurels.

Not as long as Pat "Gunslinger" Gelsinger is in charge at Intel.

Amd is not paying you so stop with the cheerleading lol