If Gracemont and subsequent atom cores are really performant, do you think Intel could expand on that technology and use them as a foundation for a new big core microarchitecture in the future?
Not only that but they could have an absolutely monster multicore server CPU right now. Why not?
If Moore's law is dead is to be believed, Raptor Lake will supposedly have configurations that have 8 big cores and 16 little cores, so 32 threads with potentially greater performance per watt and performance per die area. I doubt Intel would invest so heavily in the big/Little design philosophy if they didn't think it would pay off. Now as to whether these little cores would be competitive enough with ARM designs to merit their own separate product without any big core support is another question entirely.
When Microsoft finally introduces Windows 11, I think we're going to get some valuable insight as to both Intel and AMD's future intentions. Not going to lie though, even though I am frustrated by Intel keeping a relatively tight lid on things, I am excited about where the CPU industry is heading.
Asterox
Golden Member
- May 15, 2012
- 1,026
- 1,775
- 136
Well, in today world BigLittle is not revolutionary or spectacular at all.In the past Intel threw a lot of dollars into the water.
How Intel Lost $10 Billion and the Mobile Market
How did Intel -- the most advanced foundry in the world -- manage to spend so many billions of dollars on mobile technology with so little to show for it?
I'm still waiting for Alder Lake preview. Come on Intel, show at least the Cinebench R23 test or numbers.
Page 416 - Discussion - Intel current and future Lakes & Rapids thread
Page 416 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
forums.anandtech.com
Did Intel kill your dog? Seriously, I'd like to know. Stop the thread-crapping already!Well, in today world BigLittle is not revolutionary or spectacular at all.In the past Intel threw a lot of dollars into the water.
How Intel Lost $10 Billion and the Mobile Market
How did Intel -- the most advanced foundry in the world -- manage to spend so many billions of dollars on mobile technology with so little to show for it?www.extremetech.com
I'm still waiting for Alder Lake preview. Come on Intel, show at least the Cinebench R23 test or numbers.
Page 416 - Discussion - Intel current and future Lakes & Rapids thread
Page 416 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
IntelUser2000
Elite Member
- Oct 14, 2003
- 8,686
- 3,785
- 136
First, we don't know how big Gracemont is, at all.
My guess is it'll probably end up under 1.2mm2 in line with previous Atom-based cores. If it wasn't small, there would be no point, as Alderlake is going to be easily over 200mm2 since Tigerlake-H is already at 190mm2. Add 96 EUs, we are at maybe 210mm2. Roughly 80mm2 for 8 Tigerlake cores, so if we increase that by 50%, that's +40mm2 making it 250mm2. If each Gracemont cluster is equal to one Golden Cove core then it's +20mm2 and we're at 270mm2. You see, there's really no room for the die to grow. If Raptor Lake really does add two extra clusters, then we're looking at a potential 300mm2 die, or something very close to it.
Second, we don't know the exact scaling for the Atom-based cores. We know the 14nm process resulted in a 2.78x density increase, or a 64% reduction. That makes Airmont similar to the size of Tremont at about 0.8-0.85mm2.
Tremont is 0.85mm2, and Goldmont is at 1.1mm2. But there's Goldmont Plus with substantial performance improvements. Skylake is an 8.7mm2 core by the way.
Ian Cutress says Tremont performs like a Haswell Celeron(Haswell without AVX) in his testing. In my view, that's probably on the optimistic side and I stick to ~Ivy Bridge. If it really performs like Haswell on Integer applications, then it won't take much to outperform Skylake, since it's only about 10-15% faster.
Last edited:
If these 56 XCC and 34LCC rumors are true, i think the XCC tile might be actually 14core one (4x4, with 2 tiles being IMC) and the MCC tile (5x4, again with 2 tiles IMC, same case as like Skylake).
I wonder why people think that 4x4 die can get away with single IMC tile now, if Skylake needed 2 tiles for 5x4 design? Then again, i dont know a thing
Anyway, having 2 tile designs (and then need for mirrored version of each) is kind of weird. Does not look really cost-saving.
I wonder why people think that 4x4 die can get away with single IMC tile now, if Skylake needed 2 tiles for 5x4 design? Then again, i dont know a thing
Anyway, having 2 tile designs (and then need for mirrored version of each) is kind of weird. Does not look really cost-saving.
Remember it is going to be ddr5 so the bandwidth all but doubled.If these 56 XCC and 34LCC rumors are true, i think the XCC tile might be actually 14core one (4x4, with 2 tiles being IMC) and the MCC tile (5x4, again with 2 tiles IMC, same case as like Skylake).
I wonder why people think that 4x4 die can get away with single IMC tile now, if Skylake needed 2 tiles for 5x4 design? Then again, i dont know a thing
Anyway, having 2 tile designs (and then need for mirrored version of each) is kind of weird. Does not look really cost-saving.
Also we've seen the die shots now and it's unlikely to change much from that, 4x4 cores with 1 tile mc and was there a leak for a smaller die?. It is uncommon to have mirrored dies but if anyone has the scale of manufacturing to do such a thing then it's intel, this is compared to epyc1 that had an extra ifop on die so the four chiplets could be arranged only rotated.
The only new thing I think we'll see is a 60core halo sku, won't be widely available but it's going to be the cpu you see for benchmarks, there was a similar show with icelake-x, everything pointed to 38 cores but a 40core was released, even the dell workstation roadmap only has a 38core version.
Last I checked on Dell's site it does say the 40 core is available to purchase for servers.
As expected the performance increase in everything had nothing to do with Windows 11. What a noob.
Don’t believe those leaked Windows 11 benchmarks going around
While Windows 11 will might pack performance improvements over Windows 10, don’t be see quick to buy into recently leaked benchmarks.
www.windowscentral.com
It was an old leak that was specifically 1s workstations, not servers. I can't find it anywhere now.
But I'm referring to how everything was saying ice lake was 38 cores and comes out in 40, I don't think there's anything stopping a 60core sapphire rapids from existing so I expect some to be released even if they're rarer than hen's teeth.
dullard
Elite Member
- May 21, 2001
- 25,066
- 3,413
- 126
Sorry, I was out of range of internet since you posted this. I was trying to be non-combative by leaving AMD out of the discussion. But, if you only accept official slides, here are some direct from AMD itself:
"Glueless architecture to scale from one to four processors" slide 6
"Allows for “glueless” multiCPU designs" slide 9
"Scales to an 8-way without additional glue logic" slide 9
Does that count? AMD itself publically called Intel's dual core chips glued together for years. When Intel joked back in a leaked internal slide, suddenly Intel is in the wrong for using the term "glue".
The presentation you linked to was about multi-socket scaling and using the term 'glue' was (and is) fairly standard industry language referring to not needing additional network logic to connect the dies together. Intel's slide about Zen was just trying to take a dig at AMD when in reality, AMD was just well ahead of the curve and either Intel was blind to that or realized and was just trying to downplay AMD's lead.
Edit: I'll just add that at the time, the criticism against Intel was also justified. They were also behind AMD technologically at that time and rushed a multi-core approach on a tech that wasn't really made for what Intel was using it for. This is far different than what AMD did with Zen where the approach was designed as such from the beginning. The first iteration had some drawbacks but AMD has done a solid job with iterating on the tech in conjunction with process improvements as well.
Last edited:
Specifically it was intel calling out epyc in comparison to the new xeons. Stating epyc was something like "four desktop cpus glued together"
Very much downplaying what the zepplin die was to enterprise customers when arguably it was backwards with ryzen being a quarter of a server cpu.
I'll bring it up again but sapphire rapids looks far too similar to epyc 1 with a chiplet approach and distributed memory controllers among them, I expected an active interposer innitially.
dullard
Elite Member
- May 21, 2001
- 25,066
- 3,413
- 126
Yes, it was a joke dig at AMD. But AMD was doing the same joke digs at Intel for a decade before that with their Opteron vs Pentium D marketing (search for "AMD from the ground up" in the 2005 to 2015 time frame to see lots more examples). Either both sides are wrong, or both sides can have fun with the topic. "Glue" isn't an Intel is evil AMD is an angel topic like so many people think it is. It has been a series of digs at each other over decades going back and forth.
Last edited:
It's all just marketing. It's just that Intel makes itself look silly because their 'glue' comments made it look like they didn't understand what was happening and how their tech lead was quickly slipping away. It's not the comment itself that deserves so much criticism, it's the context and tone that make them receive so much (deserved) flak.
What cracked me up at the time was that same marketing material quoting a WCCFtech article to support their arguments against Zen.
dullard
Elite Member
- May 21, 2001
- 25,066
- 3,413
- 126
Few sane people would argue that Intel wasn't losing their lead. But that isn't what the anger was about here on the forums. The anger was all about the word "glue". Yes, AMD gained the lead, yes Ryzen is a good chip. But, I still find it humorous that now these same forums are arguing that Intel needs to glue chiplets together.
Who was arguing Intel should glue chiplets together? It seems you are siding with Intel in that AMD's tech is just gluing things together. If you didn't notice, Intel's solution with Saphire Rapids is more akin to 'gluing' dies together than what AMD has done with chiplets.
dullard
Elite Member
- May 21, 2001
- 25,066
- 3,413
- 126
This branch of the topic started here:
Page 455 - Discussion - Intel current and future Lakes & Rapids thread
Page 455 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
forums.anandtech.com
Instead of a monolithic chip, Intel rushed out the Pentium D as 2 chiplets. AMD made fun of Intel for doing so, calling it glued together. Intel later retorted making fun of Ryzen as chiplets glued together. People on this and other forums decided that it was awful for Intel to say AMD glued chiplets together. Now some people want Intel to go back to chiplets. I just think it is a humorous about face on these forums. If you don't find that humorous, that is okay.
If you want my "side", then I guess it is perfectly fine for Intel or AMD to "glue" chiplets together. Whatever it takes to get more performance with the process they have is fine with me. Monolithic is better if it was possible with the production technology they have. But if they can't produce it profitably, I'm perfectly fine with "glued" chiplets if it makes higher performing CPUs.
Last edited:
This branch of the topic started here:
Page 455 - Discussion - Intel current and future Lakes & Rapids thread
Page 455 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Instead of a monolithic chip, Intel rushed out the Pentium D as 2 chiplets. AMD made fun of Intel for doing so, calling it glued together. Intel later retorted making fun of Ryzen as chiplets glued together. People on this and other forums decided that it was awful for Intel to say AMD glued chiplets together. Now some people want Intel to go back to chiplets. I just think it is a humorous about face on these forums. If you don't find that humorous, that is okay.
If you want my "side", then I guess it is perfectly fine for Intel or AMD to "glue" chiplets together. Whatever it takes to get more performance with the process they have is fine with me. Monolithic is better if it was possible with the production technology they have. But if they can't produce it profitably, I'm perfectly fine with "glued" chiplets if it makes higher performing CPUs.
Chiplets are so much more advanced than connecting 2 full dies together over FSB. It's quite absurd to try to make them at all equivalent.
Edit: the post you linked to is arguing against Intel "gluing" dies together.
dullard
Elite Member
- May 21, 2001
- 25,066
- 3,413
- 126
In this forum terminology going back the 20 years that I've actively posted here, Chiplet and similar = glue, monolithic = no glue. I'm going with how this forum has been speaking for years. I don't quite see how you think the poster that wanted completely different cores on different chiplets was arguing against "glue".
Sure there are better and worse ways to connect chiplets together.
Edit: changed "chiplet" to "chiplet and similar" for Hitman928.
Last edited:
Please show me where the term chiplet was used further back than a couple of years ago. Again, putting multiple die on 1 substrate does not equal chiplets. You are completely discounting the work/tech behind AMD's chiplet solution.
Thunder 57
Platinum Member
- Aug 19, 2007
- 2,675
- 3,801
- 136
I think you are going out of your way to find a way to be offended. In your mind you've created this scenario that when AMD said glue no one cared but when Intel did, *GASP* it hit the fan. How about the fact that Intel called Epyc glued together "Desktop" Die? It was much more smartass and just not factual.
TRENDING THREADS
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 10K
-
Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)- Started by Vattila
- Replies: 13K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 7K
-
Facebook
Twitter