I thought AMD had already brought chiplets to data center CPUs a long time ago [1], [2] and that is why they are leading currently, now that they have multiple generations of chiplet data center CPUs out? [3]
Yes. The article (second paragraph) clearly states that this is a first for Intel, not for Data Centers.
Given that this is a trade-ish publication, I think writers/editors will more or less assume their audience understands the smackdown that AMD has been laying on Intel and AMD's chiplet strategy.
Though it's not even really a first for Intel. They made the Xeon 9200 series two years ago, and while those were slapdash and hard to buy, it's hard to argue that they don't deserve this crown. Two dies in a single package, each with 28 cores and memory controllers and I/O.
Even when Intel moves to heterogeneous dies, that won't actually be a first for them, even in the modern era. Several early i3/i5/i7 models had one die with two cores and another die that handled memory and I/O.
You are right AMD has done that. I don't know if it is the reason they are leading -- they have better process technology and in many ways better core design as well.
But this story is about Intel. Though it depends on how you define "chiplet". In this article it looks like 4 identical chips wired on a package each with their own IO and memory controllers. That's very different from from AMD's core chips wired to an IO chip, and is nothing new for Intel. https://en.wikipedia.org/wiki/Multi-chip_module
It's the expected response from Intel when they are unable to compete on throughput per socket with a single chip (they went to DCMs during the period of Opteron dominance too). I don't know what the story is really -- they said the latency was too high to do this before but they've already done something effectively the same (and multi socket SMP systems have even higher chip to chip latency again, so clearly it can be done). I guess they just improved the on package interconnect performance a bit, but almost certainly they would have done a multi chip to compete with AMD even if they were not able to improve that latency.
Sapphire Rapids is more similar to AMD's "Naples" Epyc from 2017 which also used four identical dies. Intel's mesh and EMIB should give Sapphire Rapids dramatically more internal bandwidth than AMD designs. It counts as chiplets IMO because the dies aren't intended to be used standalone.
More similar to that than? The current AMD chip+io die approach? Sure.
As I said, "chiplet" just depends on a definition, there is really no good single one. Almost certainly the dies could be used standalone, and it wouldn't be surprising if some were for low end SKUs (modulo the statement in the article that they won't for this generation) -- there are some mesh IOs on some length about 1/2 the die on two edges, that's it. Not much area.
also done to allow more SKUs from the same components, allowing AMD to make up any mix of 8 to 64 core chips without manufacturing fixed quantities of each or disabling most of the die (expensive)
There's some truth to this in the sense that a single die would almost always have better performance than chiplets, but chiplets also reduce cost and they make larger core counts possible. A single die literally can't fit more than ~48 big cores with the associated cache and I/O, so 56, 64, or 96-core processors are only possible using chiplets.
That was indeed my point - you said it perfectly. I meant, that there is nothing for the customer in chiplets except cost.
Intel went the EMIB route to route through embedded silicon but people need to be reminded on HN that chiplet/fabric architecture is not something to brag about as a feature - which is what a lot of comments are doing (AMD did it first!). Intel went through this mess because they had to due to yield. It ain't a feature for the customer.
And at the time of the Phenom X4 launch, AMD was mocking Intel for not putting all cores on one die. Seems the technically superior solution wobbles depending on who's doing it.
Chip development cycles are long enough that sometimes the only response companies have is to downplay their competitor's innovations until they have their own version. The hilarious part is that sometimes it works.
Though there's something to be said about die size. If you can't reach 250mm2 on a single die, that's much more mockable than wanting multiple dies to build your 900mm2 server CPU.
> Later, it was reduced to a more narrow role as a server and high-end desktop processor and was used in supercomputers like ASCI Red, the first computer to reach the trillion floating point operations per second (teraFLOPS) performance mark.
Amazing. RTX 3090 has peak fp32 performance of 35 TFLOPS.
I believe that almost all floating point instructions on the Pentium Pro had the same performance whether you were using the x87 FPU in 32-bit, 64-bit or 80-bit mode, so it's probably more fair to compare ASCI Red against the fp64 performance of today's GPUs. NVIDIA usually doesn't let the consumer GeForce parts get anywhere near 1 TFLOPS for double-precision. You have to look at some of their Titan or datacenter GPUs to get that kind of double-precision performance.
I find it interesting that the dies aren’t completely identical like Zen chiplets are. If I am reading the diagram correctly, it looks like the tiles diagonal from each other are the same. Kind of neat that they’ll have left-hand and right-hand wafers.
If you are just as confused as I am at "fourth generation" when Xeon has been around for decades, they mean "fourth-generation Xeon Scalable Processor (SP)". Skylake-SP, Cascade Lake-SP, Ice Lake-SP were the previous generations. Where Xeon processors previously used E3/E5/E7 designations for 1/2/4+ socket capabilities, SP CPUs use metallic prefixes: Bronze representing basic processors, Silver low power, Gold adding different options for advanced interconnects and integrated accelerators, and Platinum offering the widest range of capabilities.
Intel is in the cusp of getting taking out. Architectures and process technologies are both multiple generations behind. AMD has 10x its server share. If the trend sustains for one more generation the majority of the market will go to AMD.
Not that Intel is in a good spot, but Intel for about the past 4 years has been in the position that AMD (and TSMC) had been in for much of the past 30 years (with some brief switches).
Actually it's not even that in the case of Intel vs AMD because Intel continues to have large revenues and profits (gross margins slipped from usual low-60s to low-50s in the past few years). Whereas AMD was really struggling as a company for long periods.
Could Intel be in terminal decline as a CPU design and/or silicon manufacturing leader? It's possible. Is it a done deal or are they "finished" any time soon? No way. It will be many years while this plays out.
Intel is nowhere close to the doldrums of yesteryear's AMD. It continues to command the lion's share of the x86 market, and hasn't needed to sell off assets in order to remain afloat. In fact, Intel's business has never been better. The AMD-favoring narrative that the firms have flip-flopped is not backed up by market share research, nor corroborated by quarterly filings.
That's what I was getting at. On technical execution they actually are doing pretty bad particularly in manufacturing but also in CPU design. But even technically it has been a few years in comparison to current leaders that previously lagged by similar amounts for many years. Of course financially they are doing far better.
It's just amazing to me that people who really follow the AMD / Intel battle quite closely can actually believe that Intel is finished because AMD emerged from their decade-long post-Opteron slump a few years ago with Zen.
I said they are on the cusp. Please tell me, what happens if we end up with 10nm+++? You really think we would end up with anything other than with >50% of datacenter purchases going to amd?
Currently, AMD is fab capacity limited and couldn't get enough cpus made to be 50% of datacenter purchases.
Add to that some reality about purchaser reluctance to switch, in part because of at least perceived quality of platform support, and Intel still has a lot of breathing room.
We're starting to see some positive signs from Intel fabs, so they could likely recover, as they have before. For AMD to become dominant, they will need to continue to do excellent work for a few more years while Intel continues to wallow as they have since maybe Skylake.
You said they're being on the cusp of being "taken out".
> You really think we would end up with anything other than with >50% of datacenter purchases going to amd?
Even if this did happen, how do you believe AMD survived for decades with probably under 1% of data center purchases? And small share of PC and other server revenue? How as AMD able to weather through all that and eventually come out the other side with a good product that would not be possible for Intel?
Intel's first proper competitor to Zen is coming out soon, and from what I've seen so far it looks like a monster. Couple this with them launching GPU's into the most demand-heavy market we might ever see, they don't have to do much to have a good few years.
Also, every measure I've ever seen has AMD still at least a doubling away from parity with Intel, so I find your figure very hard to believe.
AMD and Intel are fighting different battles right now.
AMD is competing with Intel today. They're offerings are better so they're winning market share and growing.
On the other hand, Intel really needs to be compared to the Intel of a few years back. They had ~100% of the server, laptop, and workstation markets. Whatever they released, people would buy it, and they could charge whatever they wanted. That level of dominance isn't going to come back. There is too much competition from AMD and ARM.
I'm not saying that Intel won't do well. But while they have to deliver competitive CPUs, they also have to succeed with this GPU push (which btw is not their first attempt) and probably also succeed with fabbing other people's chips (which has also flopped in the past). So they need to not flop this time, and they have to continue to execute well since these areas all have competition.
As a consumer, this is all excellent. It's seriously amazing what healthy competition brings.
Intel still has 90%+ of the server market. It only just started moving over to AMD. They did gain a lot of ground over Intel in the desktop market though.
That chart shows the installed base at the end of each quarter, not sales in each quarter - see note below the chart. So current Intel sales are likely significantly less than 90% of the server market.
AMD can’t make enough chips to take 50% of the current sales market yet alone 90%… you do understand that both Intel and AMD publish reports each quarter so we know how much revenue they generate in each segment right?
How about reading more carefully before being so patronizing. I just said Intel likely has less than 90% market share now. I never said AMD has 90%. Sheesh.
AMD’s Q2 revenue for enterprise, semi-custom and embedded is $1.6B, Intel’s datacenter revenue for Q2 was $6.5B. Intel still outsells AMD by at least 3 to 1 in datacenter and perhaps even a bit more in terms of volume since AMD has almost no entry level systems unlike Intel.
[1] https://www.wired.com/story/keep-pace-moores-law-chipmakers-...
[2] https://www.tomshardware.com/news/amd-epyc-processor-models-...
[3] https://www.wired.com/story/keep-pace-moores-law-chipmakers-...