Discussion Intel current and future Lakes & Rapids thread

Page 620 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Exist50

Senior member
Aug 18, 2016
745
673
136
They are going to do Bronze and Silver with 4 dies? That's a lot of 10 nm silicon for $500 list (and Intel probably only gets half of that)
Probably using MCC or even something else. Doubt they're putting in the full 4xXCC.
 

Mopetar

Diamond Member
Jan 31, 2011
6,743
3,813
136
According to whom?
Anything we've seen from Intel has shown them attaching the tiles to each other. Each tile contains all of the hardware necessary to function independently.

The only issue is that using fewer tiles means less overall IO because each tile has a portion of the total. A two tile CPU wouldn't have as many memory channels or PCIe lanes as the full chip.

I suppose there's nothing stopping them from releasing some CPUs that have all 4 tiles with a lot of cores disabled on each in order to offer full channels and lanes, but it comes down to what the market wants and what they'll pay. There's probably money in a 4 tile chip that only has two cores active per tile assuming they can hit good clock speeds.

I think there's a market for chips with more cores that don't need as many memory channels or PCIe lanes as a full chip as well.
 
  • Like
Reactions: Tlh97

A///

Golden Member
Feb 24, 2017
1,234
870
136
Well, Intel pretty much crushed AMD for a decade+ and spend enormous $$s pushing their branding on TV, print and the whole Intel Inside program with stickers on every computer that folks bought. Overcoming that kind of momentum is very difficult. Add to that AMD's limited supply of CPUs compared to Intel and, here we are. Hyperscalers are eating 'our' CCDs like they are candy ;).
Intel had a better product and AMD fumbled a few times. They were also cash strapped due to illogical business choices in the mid 2000s, a lawsuit, and a plethora of other issues. Discounting some of the Intel releases in various sections of that decade, the chips Intel were releasing weren't largely leaps and bounds ahead of the prior generation. Alderlake is the first chip generation in years that's been worth buying. The last being Skylake.
 

Borealis7

Platinum Member
Oct 19, 2006
2,867
170
106
Intel had a better product and AMD fumbled a few times. They were also cash strapped due to illogical business choices in the mid 2000s, a lawsuit, and a plethora of other issues. Discounting some of the Intel releases in various sections of that decade, the chips Intel were releasing weren't largely leaps and bounds ahead of the prior generation. Alderlake is the first chip generation in years that's been worth buying. The last being Skylake.
if only i knew to buy AMD stocks when they were 2 bucks a pop...OTOH, they've had a rough 12 month period.
and on a semi-related note, Intel's earnings call is tomorrow.
 

JoeRambo

Golden Member
Jun 13, 2013
1,603
1,712
136
Not sure if this was posted already, seems like Xeon 8470 QS tested in Aida64:

结城安穗-YuuKi_AnS🇨🇳 on Twitter: "8470 QS https://t.co/khBDwD6Xbt" / Twitter

1651053403887.png


Some interesting data here:

1) From L1 cache latency clocks during test were ~3.6Ghz, matches the SKU real well
2) L2 cache despite being 2MB sized and being 16-way has same if not better latency than desktop ADL for 1.25MB with 10 ways. I guess the predictions that Intel already paid the latency tax for large L2 were completely true.
3) L3 does not look good at all, 2.5Ghz uncore speed and 40ns does not make much sense.
4) While memory is amazing @ 93.8ns, wow. Somehow having disaster level L3 is not holding them back, could be some aggressive optimization that runs memory access in parallel of check L3 cache or even ignoring it. No idea here, but 93.8ns is not far from Anandtech ADL DDR5 Jedec testing.
 

nicalandia

Golden Member
Jan 10, 2019
1,381
1,595
106
6% in r23 is "demolishes" ??? And what about Milan or Milan-x ?
Sorry, I was clearly being Sarcastic. This thing it's going to get Mauled by Milan and rendered DOA by Genoa which is going to be released earlier than this.

I mean Intel just can't catch a break. This CPU was supposed to beat Rome(which it barely does) and Fight off with Milan way back on 2020 when 56 vs 64 cores might have been not so bad. But 10nm 56 GC cores vs 5nm 96 Zen4... I just hope for our sake that this is The worst beating intel it's going to get from AMD. They need to catch up real fast
 
Last edited:
  • Love
  • Like
Reactions: Tlh97 and Markfw

nicalandia

Golden Member
Jan 10, 2019
1,381
1,595
106
Doesn't Threadripper 3990x scores much higher than EPYC though. It will compete pretty well against Milan EPYC at least.
By the time Sapphire Rapids is released, they would have to fight off High Core, High Speed EPYC Milan(4.1 Ghz), Threadripper PRO 5995WX(4.5 Ghz), Milan-X and the new released Genoa
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
22,926
11,448
136
By the time Sapphire Rapids is released, they would have to fight off High Core, High Speed EPYC Milan(4.1 Ghz), Threadripper PRO 5995WX(4.5 Ghz), Milan-X and the new released Genoa
And we haven't even seen its power usage yet, which to me (expecting it to be high) for a server will kill it unless the performance is consummate.
 

nicalandia

Golden Member
Jan 10, 2019
1,381
1,595
106
This was Intel's Hopeful if not somewhat already delayed(but still competitive roadmap) back on 2019.
1651186593255.png


This is their Roadmap now which makes their Xeons anything but competitive
1651186616270.png
 

RTX

Member
Nov 5, 2020
59
18
41
Sorry, I was clearly being Sarcastic. This thing it's going to get Mauled by Milan and rendered DOA by Genoa which is going to be released earlier than this.

I mean Intel just can't catch a break. This CPU was supposed to beat Rome(which it barely does) and Fight off with Milan way back on 2020 when 56 vs 64 cores might have been not so bad. But 10nm 56 GC cores vs 5nm 96 Zen4... I just hope for our sake that this is The worst beating intel it's going to get from AMD. They need to catch up real fast
Since they used the 52 core 8470 cpu, it should tie with Milan with the 56 core 8480 cpu?
 

IntelUser2000

Elite Member
Oct 14, 2003
8,188
3,046
136
Yea it can potentially get worse next year with Emerald Rapids. It sounds like the original big dog Sapphire Rapids is now called Emerald Rapids. Maybe they'll finally get the 72 core config working.
 

JoeRambo

Golden Member
Jun 13, 2013
1,603
1,712
136
I'd not read into these results too much. The reality is that average windows software does not support so called processor groups and therefore the maximum threads they can use is 64 ( and what gets assigned to those first 64 is mystery ):


Anandtech wrote about it, but in reality it's way more complicated, esp when two sockets and NUMA is also involved.
 

DrMrLordX

Lifer
Apr 27, 2000
19,382
8,170
136
Doesn't Threadripper 3990x scores much higher than EPYC though. It will compete pretty well against Milan EPYC at least.
Remember 3990x is just Zen2. Yes Threadripper does tend to have higher clocks, some of the time, but Milan has a significant IPC advantage over it.

I'd not read into these results too much. The reality is that average windows software does not support so called processor groups and therefore the maximum threads they can use is 64 ( and what gets assigned to those first 64 is mystery ):
3990x can handle 128 threads under win10. Is there some reason why a Sapphire Rapids Xeon couldn't?

I mean yeah I think it would be more fair to use Linux benchmarks where possible, but I don't have the chip so I can't really influence those decisions.
 

JoeRambo

Golden Member
Jun 13, 2013
1,603
1,712
136
3990x can handle 128 threads under win10. Is there some reason why a Sapphire Rapids Xeon couldn't?
It has nothing to do with Zen2 or SR. If You actually took time to read link to Anandtech's article, You'd see that software either supports "Processor groups" or it doesn't. And if it doesn't it uses 1st processor group of 64 threads and said threads are assigned in ways that depend on Windows SKU and BIOS table.
Even argument of 64 threads vs 64 threads in case just one group is used is flawed. As we might be comparing 32C with HT versus 64C in worse case. Or 16C from each socket with HT?
 

nicalandia

Golden Member
Jan 10, 2019
1,381
1,595
106
And we haven't even seen its power usage yet.
Since they used the 52 core 8470 cpu, it should tie with Milan with the 56 core 8480 cpu?
As it turns out. The 79,150 CineBench R23 MT score was for a 2S(Two socket processor for a total of 104 Cores), And 2S EPYC Rome gets 92,000 points and 2S Milan get's 113,000 points So it's looking Really Really Bad right now for Sapphire Rapids.


Not looking good for Sapphire Rapids

EPYC Rome
1651235543671.png


EPYC Milan
1651234955111.png
 
Last edited:

lobz

Golden Member
Feb 10, 2017
1,998
2,688
136
As it turns out. The 79,150 CineBench R23 MT score was for a 2S(Two socket processor for a total of 108 Cores), And 2S EPYC Rome gets 92,000 points and 2S Milan get's 113,000 points So it's looking Really Really Bad right now for Sapphire Rapids.


Not looking good for Sapphire Rapids

EPYC Rome
View attachment 60841


EPYC Milan
View attachment 60840
There is probably something wrong there with the Intel system, because Golden Cove is EXCEPTIONALLY strong in Cinebench.
 

ASK THE COMMUNITY