Question Intel Q1 Results

[jpiniero]

https://d1io3yog0oux5.cloudfront.net/_9c3e3f96270e337e2027289673ff9794/intel/db/887/8842/earnings_release/Q1+22_EarningsRelease.pdf

Revenue down 7%, which is actually better than their forecast. Stock getting pummeled anyway.

Client was the big loser. "Datacenter and AI" posted higher revenue but lower net income. That's terrible.

 

[DrMrLordX]

I can only guess that Intel doesn't have Alder Lake in sufficient quantity to offset the losses in client. DCG while losing margin fast seems to be a more stable bubble overall so Intel likely moved older node capacity to there which would explain expanding revenue (with stagnating profit).

Well the other possibility is that maybe Intel is finally able to sell IceLake-SP at better margin which might be propping up DCG, but . . . I'm skeptical. I have no useful data on Alder Lake other than supplies available through the DiY market which seem pretty robust.

And yeah 14nm has mostly been moved to DCG at this point. Intel wouldn't have enough product volume without it.

 

[IntelUser2000]

I can only guess that Intel doesn't have Alder Lake in sufficient quantity to offset the losses in client. DCG while losing margin fast seems to be a more stable bubble overall so Intel likely moved older node capacity to there which would explain expanding revenue (with stagnating profit).

The forced stay at home order is over in most areas so you don't need to be on zoom 24/7 for everything anymore. That'll cut into computer demand. There's also the impact of Alderlake mobile launch in Q1.

Sapphire Rapids launch in Q2 will also decrease server revenues. Launch quarters always result in lower revenue due to transition. The successive quarters will bounce back up.

 

[Markfw]

7% is actually great numbers when your top of the line Server/Datacenter and AI have been getting trashed for the last few years and your Upcoming product will not fair any different

View attachment 60818

So after the update where we saw that it was 2 52 core CPUs or 104 cores to get the 79,150 vs 74,422 for 64 cores, I would say that something is wrong here. (approx numbers, it does not show right now) I could believe that sapphire rapids beats Zen 2 by 6%, but 104 cores to do so vs 64 ? I think we need to wait for a full review my valid people on sapphire rapids vs EPYC Rome and Milan, or even those and Genoa.

 

[nicalandia]

So after the update where we saw that it was 2 52 core CPUs or 104 cores to get the 79,150 vs 74,422 for 64 cores, I would say that something is wrong here. (approx numbers, it does not show right now) I could believe that sapphire rapids beats Zen 2 by 6%, but 104 cores to do so vs 64 ? I think we need to wait for a full review my valid people on sapphire rapids vs EPYC Rome and Milan, or even those and Genoa.

I believe those numbers, low clocks and 8 Numa Domains can do that to scalability on Cine Bench. Remember how that affected OG Zen1

 

[Markfw]

I believe those numbers, low clocks and 8 Numa Domains can do that to scalability on Cine Bench. Remember how that affected OG Zen1

Well, I will not be as critical until we have retail silicon I guess is what I am saying. I still think it will loose badly to Genoa, MAYBE even Milan, but can't tell now with this test. And unfortunately, the core count will not be equal. Anyway, I think it WILL suck a lot of power if its like Alder lake.

 

[nicalandia]

Well, I will not be as critical until we have retail silicon I guess is what I am saying.

Understood..
This is Sapphire Rapids on a 2S configuration now

 

[Markfw]

Wit, it says "AMD EPYC Infinity Fabric NUMA communication Package Mapping Legend".... Whats that about ?

 

[TheELF]

Well, I will not be as critical until we have retail silicon I guess is what I am saying. I still think it will loose badly to Genoa, MAYBE even Milan, but can't tell now with this test. And unfortunately, the core count will not be equal. Anyway, I think it WILL suck a lot of power if its like Alder lake.

This isn't going to be a desktop part that has to reach high clocks...
Nobody in their right mind would run a server at full turbo anyway.

If we take the desktop parts as a guideline, as you clearly do, the super hot an un-efficient 12900k running at base TDP of 125W uses 5 whole watts more than the 5950x at its default settings...at which point it is about 7% slower in cinebench even with 24 against 32 threads, but also the 5950x is almost 20% hotter than the 12900k, especially for servers better temps will often trump better performance since better temps allow more systems in the same space without changing anything about cooling.
Also if a server has to pay for software by the logical core the 12900k makes much more sense due to the lower core count and still almost equal performance.

Another thing, but I don't know if this is true for server CPUs as well, but intel CPUs are extremely hard to trigger a thermal or any other kind of shutdown on, they will throttle down insanely but you wont have any down time unless you do crazy stupid stuff on it, desktop ryzen just immediately shuts off if they hit the thermal limit which would be ultra bad for servers.
Watch a goof trying to blow up an intel CPU

All these systems ran with the be quiet dark rock pro 4 and the 12900k was 13 degrees COOLER, 67 to 80.

Die Testsysteme - Seite 2 - Hardwareluxx

Core i9-12900K und Core i5-12600K: Hybrid-Desktop-CPUs Alder Lake im Test.

www.hardwareluxx.de

 

[DrMrLordX]

This isn't going to be a desktop part that has to reach high clocks...

Are we still talking about Sapphire Rapids here? Top SKU is going to be, what, 370W? High clocks or no, it's going to be power hungry.

 

[jpiniero]

Wit, it says "AMD EPYC Infinity Fabric NUMA communication Package Mapping Legend".... Whats that about ?

Sapphire's topology is pretty similar to Zen 1.

Are we still talking about Sapphire Rapids here? Top SKU is going to be, what, 370W? High clocks or no, it's going to be power hungry.

Yep. The boost clock should be pretty high actually as Intel has sold even the server parts as workstation. Yields will lower the boost clock on the highest core count models but I wouldn't be surprised to see a model with 5 boost.

 

[scineram]

So which group contains Altera revenue now?

 

[jpiniero]

So which group contains Altera revenue now?

Seems they lumped it into Datacenter & AI. Except possibly some of the revenue might be in Network & Edge.

 

[nicalandia]

Wit, it says "AMD EPYC Infinity Fabric NUMA communication Package Mapping Legend".... Whats that about ?

It was a reference to illustrate what is going to be Sapphire Rappid's NUMA Communication Fabric. It's remarkably similar to Zen1(perhaps this is where the whole Patent issue comes to play), there are going to be four NUMA nodes per package and eight in a 2S System. Cinebench does not scale well on 2S systems

 

[Exist50]

It was a reference to illustrate what is going to be Sapphire Rappid's NUMA Communication Fabric. It's remarkably similar to Zen1(perhaps this is where the whole Patent issue comes to play), there are going to be four NUMA nodes per package and eight in a 2S System. Cinebench does not scale well on 2S systems

It'll probably only be four nodes with SNC. Otherwise, likely to be treated as monolithic.

 

[nicalandia]

It'll probably only be four nodes with SNC. Otherwise, likely to be treated as monolithic.

This is what I believe it's going to be on an 2S system

 

[TheELF]

Are we still talking about Sapphire Rapids here? Top SKU is going to be, what, 370W? High clocks or no, it's going to be power hungry.

As I said, who runs their servers at full turbo all the time?! If such people do exist they will do it for a reason which means that it will be justified.
My point is that you are not forced to use max power turbo all the time and in server workloads I'm not aware of anybody running their CPUs at full turbo power for no reason, if you know a use case please tell us.

 

[Markfw]

As I said, who runs their servers at full turbo all the time?! If such people do exist they will do it for a reason which means that it will be justified.
My point is that you are not forced to use max power turbo all the time and in server workloads I'm not aware of anybody running their CPUs at full turbo power for no reason, if you know a use case please tell us.

First, servers don't run in turbo mode. If one core is loaded they run at a higher speed, but most servers are loaded more than 50% all the time. The 370 will be stock at base , but full load. Like the EPYC 7742 is 225 watt TDP at 2.25 ghz, but loaded. They will be hot power hungry monsters. And that is for the 56 core (or 52 core) model.

And as for how well its doing ? see here, it will be crushed:

Page 622 - Discussion - Intel current and future Lakes & Rapids thread

Page 622 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[jpiniero]

What is your point in relation to my post?

Think you are underselling using that much 10 nm silicon kills the margins. AMD jacking up Epyc prices should help the situation but it's still going to be rough.

 

[lobz]

The forced stay at home order is over in most areas so you don't need to be on zoom 24/7 for everything anymore. That'll cut into computer demand. There's also the impact of Alderlake mobile launch in Q1.

Sapphire Rapids launch in Q2 will also decrease server revenues. Launch quarters always result in lower revenue due to transition. The successive quarters will bounce back up.

How could SPR suddenly launch in a month? 🙄🙄

 

[DrMrLordX]

As I said, who runs their servers at full turbo all the time?!

Everyone? Utilization isn't necessarily going to be 100% but you're blind if you think TDP isn't factored into TCO. Unless you're suggesting people will buy expensive server hardware only for it to sit idle all of the time? Once that hardware has to actually do something, you'd better believe that the owner will be paying to feed it and cool the server room! You need as much perf/watt as you can get, and Sapphire Rapids will not be top dog.

 

[itsmydamnation]

As I said, who runs their servers at full turbo all the time?! If such people do exist they will do it for a reason which means that it will be justified.
My point is that you are not forced to use max power turbo all the time and in server workloads I'm not aware of anybody running their CPUs at full turbo power for no reason, if you know a use case please tell us.

I/big things I design do and with ht off, every power saving feature turned off. Anyone that cares about latency and or jitter will do the same.

My workloads only do 128 bit vectors so it's not as bad as it could be.

 

[TheELF]

Everyone? Utilization isn't necessarily going to be 100% but you're blind if you think TDP isn't factored into TCO. Unless you're suggesting people will buy expensive server hardware only for it to sit idle all of the time? Once that hardware has to actually do something, you'd better believe that the owner will be paying to feed it and cool the server room! You need as much perf/watt as you can get, and Sapphire Rapids will not be top dog.

I'm sure you can understand numbers, did you even look at the pic I uploaded?
With CB23 just as an example, the 12900k at 125W locked, gets 22831 points divided by 24 threads that is 932 points per thread, the 5950x at the "same" TDP of 120W gets 24838 points divided by 32 threads that is 776 points per thread.
Intel 932p/t ,ryzen 776p/t, and for server it's going to be bigger since the efficiency cores won't be dragging down the average.
At 67° for intel against 80 for ryzen you can get way more intel with the same cooling and way more performance.

If you are letting your servers suck down more power than it has to to get you the performance you need than you are doing it for a reason.

 

[Markfw]

I'm sure you can understand numbers, did you even look at the pic I uploaded?
With CB23 just as an example, the 12900k at 125W locked, gets 22831 points divided by 24 threads that is 932 points per thread, the 5950x at the "same" TDP of 120W gets 24838 points divided by 32 threads that is 776 points per thread.
Intel 932p/t ,ryzen 776p/t, and for server it's going to be bigger since the efficiency cores won't be dragging down the average.
At 67° for intel against 80 for ryzen you can get way more intel with the same cooling and way more performance.

If you are letting your servers suck down more power than it has to to get you the performance you need than you are doing it for a reason.

How in the world can you possibly justify anything desktop to server CPUs not even released yet ? Troll somewhere else.

 

[TheELF]

How in the world can you possibly justify anything desktop to server CPUs not even released yet ? Troll somewhere else.

Wait what?!
You started doing that by claiming high power use/heat on a not yet released server part.
I was just humoring your argument.
You go troll somewhere else.

 

[Markfw]

Wait what?!
You started doing that by claiming high power use/heat on a not yet released server part.
I was just humoring your argument.
You go troll somewhere else.

It may not be released, but there is a review with FACTS about it. You try to extrapolate desktop retail to server ES ? No you are just a troll.

And your reply was to DrMrLordX, not me.