Discussion Intel current and future Lakes & Rapids thread

Page 193 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
"Several years ago, we began a transformation to reposition the company to take advantage of the data revolution that is reshaping computing," CEO Bob Swan said in the earnings call. "We are accelerating growth by expanding the capabilities of our workload-optimized platforms and playing a larger role in our customers' success. Demand for our Intel Xeon Scalable processors is very strong as customers continue to make Xeon the foundation for their AI-infused data center workloads."



"In 2019, Intel expects to generate over $3.5 billion in AI-driven revenue, up more than 20 percent year-over-year. "



"Intel’s collection of data-centric businesses achieved record revenue in the fourth quarter, led by record Data Center Group (DCG) revenue. DCG revenue grew 19 percent YoY in the fourth quarter, driven by robust demand from cloud service provider customers and a continued strong mix of high-performance 2nd-Generation Intel Xeon Scalable processors. Intel acquired Habana Labs in the fourth quarter, strengthening its artificial intelligence portfolio for the data center. "



Seems pretty clear to me.

Intel has started to do some interesting things in how they group and report numbers under a new "data-centric" model which (tbh) I'm still trying to figure out. How they got to the $3.5B estimate ($3.8B actual) is a bit of a mystery to me as they have dedicated AI products which obviously must count in there but those products I don't believe are anywhere near enough to reach that number. So do they count all Xeon scalable sales as an AI sale? How do they distinguish the Xeon scalable sales for non AI HPC workloads different than AI workloads? Even if they know the sale is for an AI server, how do they know the server is even using those extensions? It seems to me that Intel is wanting to make their presence in AI seem much bigger than it is by counting everything with DL extensions as an AI sale, whether the customer actually intends to use them or not.

In a more straight forward view, the data center group's growth was mainly driven by cloud customers (up 48% Y/Y), enterprise (& Gov) was down 7% Y/Y.
 

Hitman928

Diamond Member
Apr 15, 2012
5,177
7,628
136
It looks like intel is set to gain pc share in q1 per amds guidance. I think once icelake server is out in 2h things get ugly. gg thanks for playing!

It looks like Intel has some big contracts lined up for Q1 for cloud customers which will give a big boost to their Q1 results (+$2.3B over 1Q19) but then full year revenue of +$1.5B over 2019. So basically they're predicting a large increase in Q1 but then flat or even a decrease over the rest of the year. Comparatively, AMD is predicting +28-30% revenue and significantly higher than that % increase in servers for full year 2020. We'll see how it plays out.
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
They should, "Core" is the most ridiculous marketing term I have ever seen a semi-conductor company use.

And even more ridiculous, someone being paid $100,000+ , likely came up with it and it has been allowed to run for years now.

It is kind of an arrogant name, isn't it. Like, we think our core is so good that it's the reference for what a core is. At least Atom makes sense.
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
It looks like Intel has some big contracts lined up for Q1 for cloud customers which will give a big boost to their Q1 results (+$2.3B over 1Q19) but then full year revenue of +$1.5B over 2019. So basically they're predicting a large increase in Q1 but then flat or even a decrease over the rest of the year. Comparatively, AMD is predicting +28-30% revenue and significantly higher than that % increase in servers for full year 2020. We'll see how it plays out.

Yeah, and for the year Intel is guiding double digit decrease in revenue for client although a decent part of that could be due to the loss of the modem revenue when Apple goes 5G.
 

liahos1

Senior member
Aug 28, 2013
573
45
91
intel is what you call "sandbagging" the 2h of the year. Its pretty obvious thats why the stock ripped. AMD on the other hand - looks like they lost server share in q4, and are set to lose pc share q4'19 to q1'20. Thats why the stock tanked.
 

RetroZombie

Senior member
Nov 5, 2019
464
386
96
It looks like intel is set to gain pc share in q1 per amds guidance. I think once icelake server is out in 2h things get ugly. gg thanks for playing!
How soo?
The line looks like a complete disaster to me (assuming the information in the asus slide is real!)

Platform
Sky Lake
Cascade Lake
Cooper Lake
Ice Lake
My notes
Process Node
(Purley, 14nm)​
(Purley, 14nm)​
(Whitley 14nm)​
(Whitley, 10nm)​
Same process ever and the new bad one
Launch Date
July 11,2017​
April 2, 2019​
Q2, 2020​
Q3,2020​
Literally the same stuff 3 years latter
Socket
Socket P​
Socket P​
Socket P+​
Socket P+​
Same refreshed socket, ok
PCH
C62x series​
C62x series​
C62x series​
C62x series​
Same old stuff, ok
Cores
Up to 28C per socket​
Up to 28C per Socket​
Up to 48C per socket​
Up to 38C Per Socket​
More cores nice, but the newer will have less cores than the predecessor?
TDP
Up to 205W​
Up to 205W​
Up to 300W​
Up to 270w​
Increase in TDP, the trade of for adding more cores, but the newer with less 10 cores for just 30 Watts decrease?
Scalability
2S, 4S, 8S​
2S, 4S, 8S​
2S​
2S​
End of scalability? Only 2S cpus? Severe platform downgrade?
Memory
6CH DDR4, up to 2666Mhz
12 DIMM Per Socket​
6CH DDR4, up to 2666Mhz
12 DIMM Per Socket
1st Gen Intel Persistent memory​
8CH DDR4, up to 3200Mhz
16 DIMM per socket​
8CH DDR4 up to 3200Mhz
16 DIMM per Socket
2nd Gen Intel Persistent memory​
added two more memory channels to feed 10 cores (and 20 cores) it obviously needed
PCI-e lanes
48 Gen3​
48 Gen3​
64 Gen3​
64 Gen4​
Major breakthrough, oh wait…
UPI Link
UP to 3 UPI​
UP to 3 UPI​
UP to 4 UPI​
UP to 3 UPI​
Who cares…

If I would put amd epic to compare it would be a blood bath, but since many are very sensible in discussing others guys stuff in other guys threads I will not do that.
Note: Used OCR software to make the table so it might have typos.
 
  • Like
Reactions: lightmanek

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
intel is what you call "sandbagging" the 2h of the year.

Once this cycle ends, Intel's going to have to really slash server prices to maintain share. As it is, they are already selling XCC dies for slightly more than HCC.

Have no idea how they are going to sell Cooper versus Rome/Milan.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
intel is what you call "sandbagging" the 2h of the year. Its pretty obvious thats why the stock ripped. AMD on the other hand - looks like they lost server share in q4, and are set to lose pc share q4'19 to q1'20. Thats why the stock tanked.
I saw somewhere, I think Hardware Unboxed, that high end gaming rig sales (from OEMs) is almost entirely i7 9900k. No one was touching AMD. That sentiment may have trickled down to lesser systems (years of Intel Inside advertising, etc.). Corporate customers are likely still buying Intel. Intel has been building their brand for Decades and is considered a reliable partner. AMD, unfortunately, has a spotty record. Momentum is a real asset, especially in servers. Remember when the saying was 'nobody got fired for buying IBM'? Now, replace IBM with Intel.

Anyway, low double digit growth is probably all we can expect from AMD for now. They just don't have the advertising budget to build brand recognition the way they need to right now. IMO.
 

liahos1

Senior member
Aug 28, 2013
573
45
91
Once this cycle ends, Intel's going to have to really slash server prices to maintain share. As it is, they are already selling XCC dies for slightly more than HCC.

Have no idea how they are going to sell Cooper versus Rome/Milan.

you could say the same thing about cascade vs rome. the proof is in the pudding.

amd did like ~250mln of epyc revs in the quarter. In q3 they said 50% seq growth for epyc. In q4 they said "strong double digit". Why be vague?

Intel did 7.2bln of DCG revenue in Q4 and had asp up mid single digits, a full two quarters after rome's release.

They also guided their PC biz down seasonally (double digits) while intel guided its pc biz down superseasonal (down 2-3%).
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
How soo?
The line looks like a complete disaster to me (assuming the information in the asus slide is real!)

Platform
Sky Lake
Cascade Lake
Cooper Lake
Ice Lake
My notes
Process Node
(Purley, 14nm)​
(Purley, 14nm)​
(Whitley 14nm)​
(Whitley, 10nm)​
Same process ever and the new bad one
Launch Date
July 11,2017​
April 2, 2019​
Q2, 2020​
Q3,2020​
Literally the same stuff 3 years latter
Socket
Socket P​
Socket P​
Socket P+​
Socket P+​
Same refreshed socket, ok
PCH
C62x series​
C62x series​
C62x series​
C62x series​
Same old stuff, ok
Cores
Up to 28C per socket​
Up to 28C per Socket​
Up to 48C per socket​
Up to 38C Per Socket​
More cores nice, but the newer will have less cores than the predecessor?
TDP
Up to 205W​
Up to 205W​
Up to 300W​
Up to 270w​
Increase in TDP, the trade of for adding more cores, but the newer with less 10 cores for just 30 Watts decrease?
Scalability
2S, 4S, 8S​
2S, 4S, 8S​
2S​
2S​
End of scalability? Only 2S cpus? Severe platform downgrade?
Memory
6CH DDR4, up to 2666Mhz
12 DIMM Per Socket​
6CH DDR4, up to 2666Mhz
12 DIMM Per Socket
1st Gen Intel Persistent memory​
8CH DDR4, up to 3200Mhz
16 DIMM per socket​
8CH DDR4 up to 3200Mhz
16 DIMM per Socket
2nd Gen Intel Persistent memory​
added two more memory channels to feed 10 cores (and 20 cores) it obviously needed
PCI-e lanes
48 Gen3​
48 Gen3​
64 Gen3​
64 Gen4​
Major breakthrough, oh wait…
UPI Link
UP to 3 UPI​
UP to 3 UPI​
UP to 4 UPI​
UP to 3 UPI​
Who cares…

If I would put amd epic to compare it would be a blood bath, but since many are very sensible in discussing others guys stuff in other guys threads I will not do that.
Note: Used OCR software to make the table so it might have typos.
My 7742 EPYC only draws 225 watts, VERIFIED with 64 cores @100%, I call that a bloodbath. On a single fan NH-14U air
 

jpiniero

Lifer
Oct 1, 2010
14,509
5,159
136
you could say the same thing about cascade vs rome. the proof is in the pudding.

Look at the Cost of Sales. They managed to maintain share with Cascade, but they had to give XCC for basically HCC prices. And that is what caused the shortage to get worse.
 
  • Like
Reactions: uzzi38

dmens

Platinum Member
Mar 18, 2005
2,271
917
136
"In 2019, Intel expects to generate over $3.5 billion in AI-driven revenue, up more than 20 percent year-over-year. "

LOL pretty easy to fudge that number by including any chip with so-called "DL boost" instruction support as "AI-driven"

Speaking of AI, expect news from Intel of yet another setback on that front soon. :grinning:
 
  • Haha
Reactions: Nothingness

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I just don't understand why anyone would buy a general-purpose CPU (even with bfloat16 and VNNI) for "AI" when it's more intelligent to use a dedicated accelerator from nVidia. Claiming that AI is driving Intel's DCG earnings seems really weird.

As far as the rest of the financials go . . . Intel is still making money, but they're doing it in a way that is generally bad for the industry (and Intel's long-term prospects). Expanding marketshare while consistently losing benchmarks in not a good sign, for anyone. Not sure exactly what all that has to do with the topic either, other than the fact that 38c Ice Lake-SP will probably help Intel maintain their benchmark losing streak due to delays.
 

liahos1

Senior member
Aug 28, 2013
573
45
91
I just don't understand why anyone would buy a general-purpose CPU (even with bfloat16 and VNNI) for "AI" when it's more intelligent to use a dedicated accelerator from nVidia. Claiming that AI is driving Intel's DCG earnings seems really weird.

As far as the rest of the financials go . . . Intel is still making money, but they're doing it in a way that is generally bad for the industry (and Intel's long-term prospects). Expanding marketshare while consistently losing benchmarks in not a good sign, for anyone. Not sure exactly what all that has to do with the topic either, other than the fact that 38c Ice Lake-SP will probably help Intel maintain their benchmark losing streak due to delays.

how are they making money in a way that is bad for the industry? According to many in here, they should be getting crushed by AMD bc of cinebench results? Maybe there is a misunderstanding between what actually drives purchases in the market vs the tech analysis people use in here?
 

Richie Rich

Senior member
Jul 28, 2019
470
229
76
Also, imagine calling a core prehistoric purely because of it's ALU count.
So do you disagree that there is a race for first 6xALU core in x86 world?

Because clearly there are very low hanging fruits for 6xALU as we can see at Apple's A13 example (+80% PPC). IMHO something like 20-30% PPC jump is very achievable in first generation (compare 4xALU A10 -> 6xALU A11). If Intel's Golden Cove will be first 6xALU core then AMD will be in trouble. At least AMD can fight back with higher core counts thanks to TSMC's 5nm (Zen4).

Don't forget that in 2013 there was introduced first 4xALU uarch Intel's Haswell.......... and Apple's A7 (4xALU ARM core) was 2013 product too. However Apple in 2017 just after 4 years moved to 6xALU uarch with A11. Whole x86 world is technologically stuck in 2016 and that's IMO prehistorical.




You have been warned now three times to stop with the Apple and ALU counts in this Intel thread.


esquared
Anandtech Forum Director
 
Last edited by a moderator:

BigDaveX

Senior member
Jun 12, 2014
440
216
116
So do you disagree that there is a race for first 6xALU core in x86 world?

Because clearly there are very low hanging fruits for 6xALU as we can see at Apple's A13 example (+80% PPC). IMHO something like 20-30% PPC jump is very achievable in first generation (compare 4xALU A10 -> 6xALU A11). If Intel's Golden Cove will be first 6xALU core then AMD will be in trouble. At least AMD can fight back with higher core counts thanks to TSMC's 5nm (Zen4).

Don't forget that in 2013 there was introduced first 4xALU uarch Intel's Haswell.......... and Apple's A7 (4xALU ARM core) was 2013 product too. However Apple in 2017 just after 4 years moved to 6xALU uarch with A11. Whole x86 world is technologically stuck in 2016 and that's IMO prehistorical.
Just sticking on another two ALUs isn't going to be worth the effort unless the rest of the chip is up to the task of keeping them fed. It'd be a repeat of K10, where the architecture on paper had the raw execution resources to completely bury Conroe clock-for-clock, but its inefficiencies meant it was only barely equal to it in practice (and that's before you get to everything else that went wrong with the original Phenom).
 

Richie Rich

Senior member
Jul 28, 2019
470
229
76
Just sticking on another two ALUs isn't going to be worth the effort unless the rest of the chip is up to the task of keeping them fed. It'd be a repeat of K10, where the architecture on paper had the raw execution resources to completely bury Conroe clock-for-clock, but its inefficiencies meant it was only barely equal to it in practice (and that's before you get to everything else that went wrong with the original Phenom).
Both K10 and Conroe had 3xALU, the only difference on paper was 3xAGU in K10 vs. 2xAGU in Conroe. That's not a good example regarding ALU count. Better one is Dozer vs. IvyBridge, 2xALU vs. 3xALU. Same situation today just twice as wide.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
how are they making money in a way that is bad for the industry?

They're still selling a modified Skylake core in their server products. That's horrible. Intel has been standing still in the server world since 2017 and the desktop since 2016. No significant TCO reduction, very little increase in rack density, etc. They're getting massacred in benchmarks. It's embarrassing for Intel and its customers who persist in buying respins of the same core design.

The products that we discuss in this thread (Cannonlake, Ice Lake, Tiger Lake, Sapphire Rapids, and related) are still not represented in Intel's bread & butter server lineup.

@Richie Rich

With all due respect, we've had some pretty significant thread derails in here already, and you've pushed the 6xALU narrative already in multiple other threads. Care to discuss something else?
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
With all due respect, we've had some pretty significant thread derails in here already, and you've pushed the 6xALU narrative already in multiple other threads. Care to discuss something else?

It is indeed annoying and disruptive. I thought @Richie Rich would stick to that in the ARM/Apple thread but he keeps bringing it up.
 
  • Love
Reactions: lobz

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
So do you disagree that there is a race for first 6xALU core in x86 world?

Because clearly there are very low hanging fruits for 6xALU as we can see at Apple's A13 example (+80% PPC). IMHO something like 20-30% PPC jump is very achievable in first generation (compare 4xALU A10 -> 6xALU A11). If Intel's Golden Cove will be first 6xALU core then AMD will be in trouble. At least AMD can fight back with higher core counts thanks to TSMC's 5nm (Zen4).

Don't forget that in 2013 there was introduced first 4xALU uarch Intel's Haswell.......... and Apple's A7 (4xALU ARM core) was 2013 product too. However Apple in 2017 just after 4 years moved to 6xALU uarch with A11. Whole x86 world is technologically stuck in 2016 and that's IMO prehistorical.
I think I realized what's going on.

You're a robot!!!

Welcome to our civilization!
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
witeken's latest tweet, and I quote:

"Bob Swan one year (officially) Intel CEO. Scorecard:
- 10nm launch: already ramped as many new process nodes as BK in five years
- restore Tick-Tock cadence
- DG1 dGPU SDV
- Foveros, 3D XPoint DIMM, NNP-I/T, Ponte Vecchio: new tech
- Solve shortages
- Acquisitions, hiring, reorg"


And just over a week ago, you all wanted me to take him seriously because how much he likes facts.

:pensive: