Giving in - New possible chiplet Xeon found

[williamcll]

Intel has known to be against AMD's Chiplet design for years. However a possible leak might mean the company is taking a 180 in their decision.

Quote

A member of ServeTheHome forums has published what he claims to be the first photos of Intel's Xeon Scalable 'Sapphire Rapids' processor. If the images are legitimate, they may shed some light on the design of the CPU and may indicate that it does not use a large monolithic die, but actually carries two dies. The photos depict an LGA processor with a metallic heat spreader carrying an 'Intel Confidential' mark, which indicates that this is a pre-production chip meant for testing and evaluation. Another engraving indicates a rather moderate 2.0 GHz frequency of the CPU which is something to be expected from an early sample. Also, since the processor is a pre-production sample, it has a four-character stepping: QTQ2. Since the device does not look like an existing Intel processor, it could well be a sample of Intel's upcoming Sapphire Rapids.

The back side of the CPU looks typical for Intel's latest server processors with its land grid array split into two domains. Meanwhile, there are two identical sets of capacitors in the middle of the package, which supports the theory that Intel's Sapphire Rapids is indeed a multi-chip-module (MCM) carrying two dies interconnected using one of Intel's latest technologies (e.g., EMIB). By contrast, Intel's monolithic dies have one set of capacitors on the back of their packaging.

Source: https://forums.servethehome.com/index.php?threads/lga-4677.31075/

https://www.tomshardware.com/news/intel-sapphire-rapids-leaked-photo-chiplets
Thoughts: Seems like a two NUMA chip? I wonder if it can out-core the EPYC 77X2 processors that has been AMD's server flagship for the past two years. Would they also bring the design to Desktop in the next generation? The competition sure is making Intel make decisions it never thought of.

[Bombastinator]

That heat spreader doesn’t look right to me.  Like it’s made out of fiberboard or aluminum or something.  The surface is really rough for an IHS.  The lower left hand corner is also very odd.  Like there are two layers of metal or something. 

[Antistatic12]

Can't beat em, join em.

I would really like to see Intel go all in on chiplet designs, The enormous R/D budget and their talented employees could really bring some interesting results.

[Letgomyleghoe]

36 minutes ago, Bombastinator said:

That heat spreader doesn’t look right to me.  Like it’s made out of fiberboard or aluminum or something.  The surface is really rough for an IHS.  The lower left hand corner is also very odd.  Like there are two layers of metal or something. 

i was thinking that exact same thing, maybe its some "ultra high conductive alloy" that intel came up with to market their processors with 

[leadeater]

59 minutes ago, williamcll said:

Thoughts: Seems like a two NUMA chip? I wonder if it can out-core the EPYC 77X2 processors that has been AMD's server flagship for the past two years. Would they also bring the design to Desktop in the next generation? The competition sure is making Intel make decisions it never thought of.

Intel has already done MCM Xeon products, Cascade Lake-AP (Xeon 9200 Series). Those required water cooling and were application specific and not available retail (BGA package not LGA). Whether Intel carries on with that for general Xeons I don't know, the second chip in the MCM package on Cascade Lake-AP provided no extra PCIe lanes so any new product implementing the same MCM approach would require a huge number of pins in an LGA socket to provide access to those extra PCIe lanes.

 

We could just being seeing a revision update to Xeon 9200 series (9300?) and is not the next generation general Xeons. Pictures of current 9200 series for reference:

 

[Master Disaster]

1 hour ago, Bombastinator said:

That heat spreader doesn’t look right to me.  Like it’s made out of fiberboard or aluminum or something.  The surface is really rough for an IHS.  The lower left hand corner is also very odd.  Like there are two layers of metal or something. 

Yep, it goes beyond just rough metal imo, it almost looks like its deliberately textured.

 

23 minutes ago, Letgomyleghoe said:

i was thinking that exact same thing, maybe its some "ultra high conductive alloy" that intel came up with to market their processors with 

And the way to make an "ultra high conductive alloy" dissipate more heat is by reducing its contact surface area?

[RejZoR]

1 hour ago, Bombastinator said:

That heat spreader doesn’t look right to me.  Like it’s made out of fiberboard or aluminum or something.  The surface is really rough for an IHS.  The lower left hand corner is also very odd.  Like there are two layers of metal or something. 

I was just installing Ryzen 5800X in my new system recently and it had the same slightly rough surface like IHS was bead blasted. Nothing odd about above photo apart from being a crappy one.

[Arika]

3 hours ago, williamcll said:

Intel has known to be against AMD's Chiplet design for years. However a possible leak might mean the company is taking a 180 in their decision.

And they should be allowed to.

 

Companies should never for pigeon holed into doing one thing forever, just because they may have said they didn't like it in the past. If it proves more effective, then they should be free invest into it. Innovation is key.

[leadeater]

59 minutes ago, Arika S said:

And they should be allowed to.

 

Companies should never for pigeon holed into doing one thing forever, just because they may have said they didn't like it in the past. If it proves more effective, then they should be free invest into it. Innovation is key.

Well the slightly more correct way to say what was said is that Intel's marketing and PR was tasked with highlighting potential/theoretical issues with a competing product to try and make their product more appealing. It's not like Intel the company or their engineers would have ever though this at all, no they would be looking at it so see how good it actually is and how AMD have approached the engineering and design challenges around chiplets and interconnects.

[jagdtigger]

Oh no, intel has gone so low they start to glue cpu's together.....

[WhitetailAni]

What they should do is just make two Xeon Platinum 8180's on a single chip and make a new socket called LGA 7309.

Cause then if Linus drops it it won't make the other one useless.

[Bombastinator]

5 hours ago, RejZoR said:

I was just installing Ryzen 5800X in my new system recently and it had the same slightly rough surface like IHS was bead blasted. Nothing odd about above photo apart from being a crappy one.

The purpose of bead blasting is to put microscopic surface cracks in the metal.  It was designed originally as a paint preparation. It drastically increases surface area.  It’s a strange preparation for an IHS which relies on smoothness and flatness for maximum heat transfer.  Might be a method to the madness I suppose. 

[themaniac]

45 minutes ago, Bombastinator said:

The purpose of bead blasting is to put microscopic surface cracks in the metal.  It was designed originally as a paint preparation. It drastically increases surface area.  It’s a strange preparation for an IHS which relies on smoothness and flatness for maximum heat transfer.  Might be a method to the madness I suppose. 

Maybe there's some thermal paint they have, or they just haven't finished the ihs processing?

[cj09beira]

3 hours ago, ragnarok0273 said:

What they should do is just make two Xeon Platinum 8180's on a single chip and make a new socket called LGA 7309.

Cause then if Linus drops it it won't make the other one useless.

they already did that and flopped really hard, too hot, too power hungry, not enough customization

[Bombastinator]

2 minutes ago, themaniac said:

Maybe there's some thermal paint they have, or they just haven't finished the ihs processing?

Paint seems unlikely, it would require both the binder and the carrier to be highly heat conductive and I don't know of any that are particularly.   Doesn’t mean the don’t exist. The pigment could be anything granular so that could easily be highly heat conductive, but I don’t see what such a surface treatment would have over paste.  Thermal paste kind of is a paint where the silicon oil is both a binder and a carrier but does not dry. Thermal paste afaik gains nothing from adhesion. Lack of finish seems a lot more possible.  That lower left corner still bothers me. 

[DildorTheDecent]

Have a feeling Intel will do chiplets better. Something something R&D money.

[tim0901]

9 hours ago, williamcll said:

Intel has known to be against AMD's Chiplet design for years.

You mean the same Intel that's using a similar chiplet design in their discrete graphics department? The same Intel that has developed a chiplet stacking technology in Foveros?

 

Intel aren't against chiplets at all, any issues they may seem to have with them are purely from their marketing department as an attempt to shit on AMD. Ask the engineering teams there and you'll find they're just as interested in chiplets as AMD.

[tech.guru]

7 minutes ago, tim0901 said:

You mean the same Intel that's using a similar chiplet design in their discrete graphics department? The same Intel that has developed a chiplet stacking technology in Foveros?

 

Intel aren't against chiplets at all, any issues they may seem to have with them are purely from their marketing department as an attempt to shit on AMD. Ask the engineering teams there and you'll find they're just as interested in chiplets as AMD.

There is trade offs to chiplet design 

I think intel getting pressure to make there products more competitive in pricing and performance.

 

I know from talking to enterprise OEMs they are trying hard to push there customers to stay with intel (kickbacks hmmm) but enterprise customers are seeing more and more value from AMD and putting pressure on OEM to develop skus around AMD for their products.

 

Intel high core count 56 core processors are ridiculously expensive compared to epyc 64 core.

 

Theres also advantages to 1P design and smaller footprint (1U) servers.

 

You can today get same performance from single socket amd compared to intel 2 socket for significantly cheaper.

 

Intel step up your game..

[Kisai]

7 hours ago, leadeater said:

Well the slightly more correct way to say what was said is that Intel's marketing and PR was tasked with highlighting potential/theoretical issues with a competing product to try and make their product more appealing. It's not like Intel the company or their engineers would have ever though this at all, no they would be looking at it so see how good it actually is and how AMD have approached the engineering and design challenges around chiplets and interconnects.

 

Tech companies PR and Marking tend to be miles away. That's why Apple had Steve Jobs, not Steve Wozniak as the CEO/PR/Marketing all in one. Both Steves understood the tech, but had the roles been reversed, Apple would be selling ApplePi's in Altoids tins now, and they'd also be on MacOS version 32 that's incompatible with 31. "Cool, but impractical"

 

Like Intel sometimes comes out and says things that are factually untrue in order to try and keep brand loyalty with companies like Dell and HP, and businesses that rely on "Intel" CPU's. The gaming, art, and video prosumer/professionals aren't married to any specific CPU, and the marketing doesn't matter, as the only decision that goes into the CPU choice is if their existing software will work on it. 

 

 

It could be true, that Intel's more monolithic design is better for performance, but there are diminishing returns in that design unless Intel is going to make cpu's that use the entire wafer and nothing gets bin'd.

[tim0901]

2 hours ago, tech.guru said:

There is trade offs to chiplet design 

I think intel getting pressure to make there products more competitive in pricing and performance.

Absolutely there are, the infinity fabric does cause issues. But Intel has avoided using chiplets in their CPU division mostly because they've been holding out hope for the last 4 years that 10nm was 'just around the corner' and therefore haven't wanted to invest significant amounts of money in developing a new architecture for their aging 14nm node.

 

Like, to give Intel credit, their processors are still very competitive compared to AMD's when you consider that they are essentially the same architecture and manufacturing process as their processors from five years ago. Even now the 5000 series CPUs are only ~10% faster in single-core performance over the 10900K. For a five year old CPU design, that's bloody impressive.

 

IMO the best thing for intel right now is for them to get their head back in the game and fix their manufacturing issues, not to try and shake up their entire CPU design philosophy by shifting to a Ryzen-like architecture. (Or they could just cut prices...)

 

Quote

I know from talking to enterprise OEMs they are trying harder to push there customers to stay with intel (kickbacks hmmm) but customers are seeing more and more value from AMD and putting pressure on OEM to develop skus around AMD

Kickbacks aren't the only reasons why Intel is popular in the enterprise space though. From an OEM's point of view, kickbacks are ultimately useless if the customer simply says no and goes elsewhere to buy AMD's offerings. There's more to the story than "Intel buys out the OEMs lol".

 

For one, AMD's server products just aren't as reliable as Intel's - even Linus has encountered this in his server experiments before, where he encountered some bug on AMD that was instantly solved by moving to Intel - and reliability plays a huge part in selling a product to enterprise customers. This isn't to say that they are unreliable, merely that Intel is still better in this regard. From the OEM point of view, pushing Intel can thereby also be a safer choice for them as their product is less likely to run into big issues down the line (which costs the OEM lots in support).

 

The price/performance metrics also aren't also as cut-and-dry the whole way down the stack. At the top yes AMD absolutely stomps over Intel, but when comparing mid-range frequency-optimised Epycs like the 7F52 to Intel's competing product (Xeon Gold 6242), the price/performance is practically identical. Certainly not worth the risk of swapping CPU vendors.

 

And to throw a further spanner into the equation: you can put four of those Xeon 6242s in one motherboard, whereas quad socket Epyc motherboards don't exist. So you've then got to factor in the price of a second server, extra rack space etc. and so the price argument starts to swing firmly in Intel's favour. It also means if you're looking for thread density per node, Intel is still on top despite their threads-per-cpu disadvantage as you can have four 58-thread Xeon 8180M CPUs in the same node, vs two 96-thread CPUs using the top end Epyc.

 

Yes, AMD's solution is very competitive and it makes a lot of sense to many, but that's not to say that Intel doesn't makes sense to many as well. Isn't competition great?

[pas008]

didnt intel glue back in the core days?

[Bombastinator]

8 minutes ago, tim0901 said:

Absolutely there are, the infinity fabric does cause issues. But Intel has avoided using chiplets in their CPU division mostly because they've been holding out hope for the last 4 years that 10nm was 'just around the corner' and therefore haven't wanted to invest significant amounts of money in developing a new architecture for their aging 14nm node.

 

Like, to give Intel credit, their processors are still very competitive compared to AMD's when you consider that they are essentially the same architecture and manufacturing process as their processors from five years ago. Even now the 5000 series CPUs are only ~10% faster in single-core performance over the 10900K. For a five year old CPU design, that's bloody impressive.

 

IMO the best thing for intel right now is for them to get their head back in the game and fix their manufacturing issues, not to try and shake up their entire CPU design philosophy by shifting to a Ryzen-like architecture. (Or they could just cut prices...)

 

Kickbacks aren't the only reasons why Intel is popular in the enterprise space though. From an OEM's point of view, kickbacks are ultimately useless if the customer simply says no and goes elsewhere to buy AMD's offerings. There's more to the story than "Intel buys out the OEMs lol".

 

For one, AMD's server products just aren't as reliable as Intel's - even Linus has encountered this in his server experiments before, where he encountered some bug on AMD that was instantly solved by moving to Intel - and reliability plays a huge part in selling a product to enterprise customers. This isn't to say that they are unreliable, merely that Intel is still better in this regard. From the OEM point of view, pushing Intel can thereby also be a safer choice for them as their product is less likely to run into big issues down the line (which costs the OEM lots in support).

 

The price/performance metrics also aren't also as cut-and-dry the whole way down the stack. At the top yes AMD absolutely stomps over Intel, but when comparing mid-range frequency-optimised Epycs like the 7F52 to Intel's competing product (Xeon Gold 6242), the price/performance is practically identical. Certainly not worth the risk of swapping CPU vendors.

 

And to throw a further spanner into the equation: you can put four of those Xeon 6242s in one motherboard, whereas quad socket Epyc motherboards don't exist. So you've then got to factor in the price of a second server, extra rack space etc. and so the price argument starts to swing firmly in Intel's favour. It also means if you're looking for thread density per node, Intel is still on top despite their threads-per-cpu disadvantage as you can have four 58-thread Xeon 8180M CPUs in the same node, vs two 96-thread CPUs using the top end Epyc.

 

Yes, AMD's solution is very competitive and it makes a lot of sense to many, but that's not to say that Intel doesn't makes sense to many as well. Isn't competition great?

A competitive dualopoly is better than a noncompetitive dualopoly it’s true.