nicalandia
Diamond Member
- Jan 10, 2019
- 3,331
- 5,282
- 136
I'm not sure it'll exist at all. Keep in mind this is a leak, not from Intel. What sense would there be in a leaker sandbagging on Intel's behalf?You are saying they'll be used in very small amounts but I'm saying there's a very good chance it'll exist which do not contradict each other.
Well, the much bigger question is whether they're being conservative with their timelines. No one cares about a narrowly binned SKU on the top of the stack if it's a year+ late.Which is a good trend, since it means they aren't overpromising anymore. I think the new mentality is also representative of better management because you need the conservative expectations so they can better meet their goals.
Sure, it exists, but in what volume? If they're only shipping O(10,000), then it might as well not exist as far as some of the major customers (e.g. cloud) are concerned. And they only have one SKU with 60c compared to several with 56, so I think 56 is where most of the high end volume will be.The 60 Core SKU does exist. The name is Xeon Platinum 8490H we have plenty info leaked on that CPU
Taking the claims at face value for now, that seems like a weird decision. It would still be useful to have lower end SRF options, but to eschew making a proper 400-600 core throughput monster? Just feels like wasted potential. Maybe DRAM bandwidth issues with feeding that many cores?Witaken with more info on Sierra Forest.
Sapphire Rapids-SP will support 8S system with the maximum total core count will be 480 Cores vs 192 Cores for GenoaI wonder if Intel is just banking on the fact that the are still trying to support quad and eight processor configurations for their processors with respect to having lower total core counts than AMD. I realize that those are vanishingly small markets, but they do exist...
I don't think they're really "banking" on anything at this point. If there's a market AMD's underserving, they might be happy for the win, but that's not exactly a strategy.I wonder if Intel is just banking on the fact that the are still trying to support quad and eight processor configurations for their processors with respect to having lower total core counts than AMD. I realize that those are vanishingly small markets, but they do exist...
What will that 480 core system draw ? like over 4000 watts ? I don't see that happening.Sapphire Rapids-SP will support 8S system with the maximum total core count will be 480 Cores vs 192 Cores for Genoa
Close to 3,000 Watts...What will that 480 core system draw ? like over 4000 watts ? I don't see that happening.

You Underestimate The Bribing POWER of Intel. CEO Kickbacks and outright Coercion tactics.Genoa 480 cores would be 360*2.5 or about 900 watts. Why would a data center choose to spend 3 x the power, and 3x the AC and 3x the APC power for the same or less performance ?
According to YuuKiAnS(which has 4S and 8S on the lab he works at) Intel is implementing the 8S system as two 4S System stitched/glued together. Intel has many times reported that their CPU will Support Multi-Socket Scaling with their Intel Utra Path Interconect UPI 2.0. 8S-4UPI Performance Optimized Topology.Does that system have SMP ?
What I was asking was, are those e-cores ? or p-cores with smp ? With that many, maybe its just e-cores ?According to YuuKiAnS(which has 4S and 8S on the lab he works at) Intel is implementing the 8S system as two 4S System stitched/glued together. Intel has many times reported that their CPU will Support Multi-Socket Scaling with their Intel Utra Path Interconect UPI 2.0. 8S-4UPI Performance Optimized Topology.
Those are All P cores with SMT/HT. I mentioned The Maximum Cores Possible on a 8S Systems Which is 480C/960TWhat I was asking was, are those e-cores ? or p-cores with smp ? With that many, maybe its just e-cores ?

Availability and price. Though 8S in particular would have the advantage of a larger coherency domain, niche market though that is.Edit: and for those who will pick at my numbers, lets just say a LOT more power for that 480 core Intel system. And does that system have SMP ?
I thought Genoa was limited to max 192 cores since it was 2P?Genoa 480 cores would be 360*2.5 or about 900 watts. Why would a data center choose to spend 3 x the power, and 3x the AC and 3x the APC power for the same or less performance ?
Edit: and for those who will pick at my numbers, lets just say a LOT more power for that 480 core Intel system. And does that system have SMP ?
Taking the claims at face value for now, that seems like a weird decision. It would still be useful to have lower end SRF options, but to eschew making a proper 400-600 core throughput monster? Just feels like wasted potential. Maybe DRAM bandwidth issues with feeding that many cores?
I sincerely apologize; I have regarded you highly here in the past. It will not happen again. I felt bad the day after posting it, and I didn’t see you here for a while, which made me feel worse.@ashFTW Great to see that you resort to personal attacks.
Look at what AMD has been able to achieve with just a few chips. It’s important to do designs that can be reused across a wide range of products. That’s the whole idea behind chiplets. Intel is perpetually late to market; parsimonious designs is crucial to overcoming this constant hurdle. And I was only talking about high end computing solution -- what’s covered today by SPR, PVC etc. Falcon Shores seems like the right approach to me. It can be the common platform for integration of several key future technologies Including optical interconnects, extreme bandwidth memory/caching etcLook at how much the implementation of hybrid is a fine balancing act, and it's the details that make all the difference. Ring bus frequency, voltage domains. Jasper Lake is far more efficient than Gracemont primarily because Gracemont is on the same die as P cores and uses the same resources.
It has nothing to do with future vision. Purpose oriented optimization is what gives you the final advantage to stand out. And that can be any metric, whether it be cost, performance, or power. You can never have a truly unified platform across all stacks for this very reason. Servers can be small as $200 Atom C3000 board for open source routers to $500 million HPC setups. Are you really that naive to believe those chasms can be socket unified?
If anything, platform variation continues to proliferate, not the other way around.
Coincidentally so Granite Rapids has a chance.
So uh, regarding Intel “On Demand”, you that saying that is something line “Nobody ever get’s fired for buying Intel.”?
If I hire you, and you buy one of these products. you are fired. Just an FYI. 😉
I'm really not sure how all this applies to Sierra Forest. If the argument is that so many cores would be so strong relative to the competition, Intel would instead prefer to sandbag, then I heartily disagree. Intel is clearly in no position to be holding back, and who knows where the ARM vendors or Bergamo's successor will be in 2024. Moreover, for the in-house ARM vendors in particular, Intel doesn't just have to build the better product to win over Amazon, Microsoft, etc., but to be better TCO against something with effectively zero margin requirement. That's tough to say the least. AMD is facing much the same threat, so I doubt they're sandbagging either.Probably because they don't need it?
Let me lay out few things on the table before continuing the discussion.
-AMD/Nvidia has been accused of price fixing for their GPUs.
-Microprocessor engineering is a very, very high level skill, so people who can do so are limited.
-When you hear a high profile engineer/executive leave, they end up in competitor companies. Literally you are sucking a resource from one and move it to the other.
-They are people after all, and it puts a lot of demand on their time and energy.
-And at last, they are a business here to make money.
We cheer for the companies we like and would like them to "crush" the competition with some amazing product that the competitor will never be able to touch. And do so for years and years until the other party is dust.
Remember when Andy Grove wanted to use their resources and advantage to crush AMD, Moore stopped him.
The reality is that in the big picture, competition is there to just be little better than each other. And I am not saying that necessarily in a negative light either. For whatever reason, the fantastic claims never turn out. It makes sense too, since these guys are at the top of the game already. Pushing yourself to create something no-one has ever done before in the history of mankind is hard, to put it simply.
In the big picture, the better competitor is always within reach. How many people believed that Turin will go 200, or even 300+ cores? Now it looks like they'll barely go above 128, if that. Coincidentally so Granite Rapids has a chance.
Other thing is that the better core is dwarfed by better process. The Atom team being way better is probably still not better-enough to get an actual 3-4x advantage. I assume it's enough to make a difference, like putting it strategically in hybrid configs to make Raptorlake competitive, or Sierra Forest competitive with ARM competitors, and in cloud save 30% power or $ per chip vs having Granite Rapids of equal core count.
You either run into scaling challenges(core/IO/software), or the real goal is to get it higher perf/watt and/or higher perf/$. 30% gain is NOT a trivial gain at all! It's absolutely huge!
That's why I don't believe Falcon Shores will be revolutionary as some think. Reserve judgment until it happens. Advancement is always successive. So-called "revolutionary technologies" just act as something to break barriers, and nothing more.
What ever happened to "Only the paranoid survive"?Remember when Andy Grove wanted to use their resources and advantage to crush AMD, Moore stopped him.
I don't believe in Falcon Shores just because I have zero faith in Raja's org successfully building something of such a massive scope. Maybe if it was a combined effort with the Xeon team, but that doesn't seem to be what they're aiming for. But if by some miracle they do succeed, could be quite interesting. But by then AMD and Nvidia will probably be on gen 2 at minimum.That's why I don't believe Falcon Shores will be revolutionary as some think. Reserve judgment until it happens. Advancement is always successive. So-called "revolutionary technologies" just act as something to break barriers, and nothing more.
I'm really not sure how all this applies to Sierra Forest. If the argument is that so many cores would be so strong relative to the competition, Intel would instead prefer to sandbag, then I heartily disagree.
