Raptor Lake Leaks + Intel 4 developments

[Wolverine2349]

Yes.

Intel's manufacturing has been woeful for a very long time, regularly delayed and having to cut back on features originally planned just to get something out the door or else be locked on old tech and the products suffer for it (big example being the bazillion years stuck on 14nm and the joke products that came out towards the end).

And Intel's design philosophy is horribly outdated, as Vince described. Yes, the Golden Cove core is pretty good in catching and overtaking AMD's performance lead, but it's still a clunky mess. It's huge, it's power hungry and when attached to other things to make a CPU, it just does not scale. Plus Intel still make monolithic chips so they cost a lot to make, cost a lot in wastage and just not agile enough to be profitable at the levels Intel would like (or shareholders demand).

Its pretty interesting that Intel Golden Cove core is a clunky mess yet still has 15% better IPC than AMD Zen 3, And Zen 3 is still considered good. I guess was Intel able to do it in spite of it being a clunky mess not because it was an overall good design choice. Pretty amazing and would Golden Cove have blown AMD Zen 3 IPC out of water by like 30% or more instead of just 15% if Intel had a good design choice and updated node type??

I even hear Intel is trying to update their foundry and is saying even AMD can be a customer. Lol if that ever happens Intel gives upo its CPU business cause why would they want their main competitor to be able to use them lol.

And also because of Intel's design chocie is that why they only have 8 P cores and had to use the e-cores??

 

[LePhuronn]

Its pretty interesting that Intel Golden Cove core is a clunky mess yet still has 15% better IPC than AMD Zen 3, And Zen 3 is still considered good. I guess was Intel able to do it in spite of it being a clunky mess not because it was an overall good design choice.

15% better IPC than a product that was already 2 years old don't get me wrong, Golden Cove's uplift over Cypress Cove was incredibly impressive, even ignoring the hilarious concessions Intel had to make with all the backporting nonsense, and a large part of that was finally being able to (mostly) realise the design on a manufacturing node that (mostly) worked.

Please be clear that it's Intel's manufacturing that is the weak link, not the design team. The designs are great, but Intel just can't make them.

And also because of Intel's design chocie is that why they only have 8 P cores and had to use the e-cores??

Again, it's not design choices per se, it's the ability to make the designs real, and the concessions that have to be made. But yes, Intel just can't get more than 8 Golden/Raptor Cove cores into Alder/Raptor Lake because of the power required and the heat they put out. So the Gracemont cores are a crutch to make up the performance deficit. Think of it this way: a cluster of 4 Gracemont cores takes up the same physical space as a single Golden/Raptor Cove core. The 12400F showed some pretty amazing performance in a 6+0 config, and absolutely crushed the 5600X, so what would a 10+0 Alder Lake do against the 5900X, or even 5950X? There's space on the die to do it, so why doesn't it exist? How about a 12+0 Raptor Lake? Holy smokes AMD would be in trouble, so why is it not a thing? It must be scalability. The cores take too much power, dump too much heat and the interconnect to stitch it all together is also a hungry mofo.

But then look at it this way: you yourself say the E cores are a waste, and arguably for your use case they are. But those E cores do give some impressive MT performance numbers, almost unbelievably so. So if Gracemont is surprisingly that good, why is there not a 40 core Alder Lake, or 44 core Raptor Lake? Again, scalability with that damn interconnect and the ludicrous power draw (one would assume).

So yeah, Intel's hybrid approach, whilst good on paper and good in phones, is nothing more than a fudge to get around the fact Intel just can't slap out 12, 16 (and beyond) core CPUs like AMD can. And the hybrid tech is here to stay for a while, because even if Meteor and Lunar Lake really do fix the scalability issue, Intel have invested too much in thumping this hybrid tech drum to admit "yeah, it was a short-term crutch, but we don't need it any more".

 

[Vince]

Its pretty interesting that Intel Golden Cove core is a clunky mess yet still has 15% better IPC than AMD Zen 3, And Zen 3 is still considered good. I guess was Intel able to do it in spite of it being a clunky mess not because it was an overall good design choice. Pretty amazing and would Golden Cove have blown AMD Zen 3 IPC out of water by like 30% or more instead of just 15% if Intel had a good design choice and updated node type??

I even hear Intel is trying to update their foundry and is saying even AMD can be a customer. Lol if that ever happens Intel gives upo its CPU business cause why would they want their main competitor to be able to use them lol.

And also because of Intel's design chocie is that why they only have 8 P cores and had to use the e-cores??

So the only 8 P cores issue, is a scaling issues, ringbus only scales so much in terms of core count before you get hugely diminishing returns and if you try and scale a ringbus design with a ton of cores the power requirements go to crazy levels that aren't attractive for the performance on offer. It's absolutely a catch 22, ringbus is fast for lower core count chips, there is absolutely no doubt about that and it is what intel needs/knows in terms of performance, when they transferred that to a mesh they lost so much that it wasn't really feasible with their core design so they focused heavily on ringbus.

The thing is it's not about performance, not at all, both sides have perfectly adequate performance and you wouldn't turn down a 7950x or a 13900k in the consumer side, it's not about the fan boy, the 1 or 2 fps here or there in the graphs, its not really about any of that. When you look at it objectively it's about the technology on offer and how much it costs that company to get the product into your machine and how much meat is in the bone for all those involved in that supply chain, that's how companies in this space work and operate. Right now to me at least Intel are hanging on with the old tech, they need to pivot to a model/design philosophy where they can bang these chips out ten a penny like AMD seemingly are. AMD right now have 1 core design that spans the entire product stack, a single design that competes with I wouldn't like to say how many monolithic intel designs. They make the die and just bang them out and slap them in everything from consumer to the enterprise.

 

[Wolverine2349]

15% better IPC than a product that was already 2 years old don't get me wrong, Golden Cove's uplift over Cypress Cove was incredibly impressive, even ignoring the hilarious concessions Intel had to make with all the backporting nonsense, and a large part of that was finally being able to (mostly) realise the design on a manufacturing node that (mostly) worked.

Please be clear that it's Intel's manufacturing that is the weak link, not the design team. The designs are great, but Intel just can't make them.


Again, it's not design choices per se, it's the ability to make the designs real, and the concessions that have to be made. But yes, Intel just can't get more than 8 Golden/Raptor Cove cores into Alder/Raptor Lake because of the power required and the heat they put out. So the Gracemont cores are a crutch to make up the performance deficit. Think of it this way: a cluster of 4 Gracemont cores takes up the same physical space as a single Golden/Raptor Cove core. The 12400F showed some pretty amazing performance in a 6+0 config, and absolutely crushed the 5600X, so what would a 10+0 Alder Lake do against the 5900X, or even 5950X? There's space on the die to do it, so why doesn't it exist? How about a 12+0 Raptor Lake? Holy smokes AMD would be in trouble, so why is it not a thing? It must be scalability. The cores take too much power, dump too much heat and the interconnect to stitch it all together is also a hungry mofo.

But then look at it this way: you yourself say the E cores are a waste, and arguably for your use case they are. But those E cores do give some impressive MT performance numbers, almost unbelievably so. So if Gracemont is surprisingly that good, why is there not a 40 core Alder Lake, or 44 core Raptor Lake? Again, scalability with that damn interconnect and the ludicrous power draw (one would assume).

So yeah, Intel's hybrid approach, whilst good on paper and good in phones, is nothing more than a fudge to get around the fact Intel just can't slap out 12, 16 (and beyond) core CPUs like AMD can. And the hybrid tech is here to stay for a while, because even if Meteor and Lunar Lake really do fix the scalability issue, Intel have invested too much in thumping this hybrid tech drum to admit "yeah, it was a short-term crutch, but we don't need it any more".

Well actually it was 15% better IPC than Zen 3 and Zen 3 was 1 year old coming out in November 2020 and Alder Lake Golden Cove cores on it were released exactly 1 year later November 2021.

Though you say Intel's design choice is not so good either or is it just their process node being outdated??

Cause you say they backport too many things. Like what do they bakcport and why is that bad?? And isn't a monolithic die better for performance than chiplets anyways. If only they could get it on a better node

 

[Bencher]

Miracle do Intel pay you and Dave to spout your rubbish continuously on here? Gets boring very quickly.

The only nonsense is your post

I haven't yet seen productivity results with power usage between 13900k and 7950x yet ?

Why do you think the increases are bigger each time it's because of competition !

The way you make it sound Intel should be well ahead then ? 40% increase wow should be running away with it but all I see its pretty close LOL

I dont know and i never said intel is far ahead. I just stated facts. Can you find anything that you disagree with from my post?

Intel gave us a 40% performance increase in a single year with similar power draw on the same node and with same price.

Amd gave us 40-50% performance increase in 2 years with a price increase, a die shrink and 50% increased power draw.

Is there something you actually disagree with? Please point it out and ill rectify immediately.

EG1. Actually according to techpowerups review, the 7950x literally doubles the power of the 5950x. So, make that 100% increased power draw :O

It's called squeezing every last bit out of an architecture that doesn't have a solid future and has terrible financials in terms of what it costs to build while you scramble to build something akin to what you mocked some 5 years back. I wouldn't call it a miracle so much as doing what you have to do to compete with a much, much more elegant design that costs a fraction to build while being just as fast. Make no mistake even if its market leading it wont be a success for intel due to all of these factors.

That doesnt make it bad for you, me or the average user but for intel it's still a country mile from the sort of designs they need to stop haemorrhaging financially.

I don't really care about intel or their bottomline to be fair, so yeah - whatever. The point is, it took them half the time, no node shrinks , no price increases, no power draw increases to give us a 40% performance increase. If you compare that to zen 4, it seems like some kind of dark sorcery, doesn't it?

 

[uscool]

Hopefully tomorrow can see results of productivity results and power used

 

[Vince]

I don't really care about intel or their bottomline to be fair, so yeah - whatever. The point is, it took them half the time, no node shrinks , no price increases, no power draw increases to give us a 40% performance increase. If you compare that to zen 4, it seems like some kind of dark sorcery, doesn't it?

If you read my post it clearly stated that it doesn't matter what you think. You have such an epically skewed view you cant see that there is nothing "dark sorcery" about the design, its intel doing what intel has always done. The dark sorcery you are talking about happened in 2017. It doesn't make it a bad chip, far from it, it makes it as good as the competition while being at a significant disadvantage everywhere that matters if you are intel. I'm interested to see it but regardless of your persuasion you cant really argue that this is any more or less than what was absolutely required if you were intel.

Node increase does not = gains from node maturity and given the node took how long for intel to pipe products out on you can bet your arse the node is maturing quickly hence the rapid uptick in what they can do on it now vs just 2 years ago where they failed to release a laptop size chip effectively.

 

[LePhuronn]

it seems like some kind of dark sorcery, doesn't it?

No, not even remotely. Enough with your bullcrap, it's beyond pathetic now. Just stop.

 

[Bencher]

If you read my post it clearly stated that it doesn't matter what you think. You have such an epically skewed view you cant see that there is nothing "dark sorcery" about the design, its intel doing what intel has always done. The dark sorcery you are talking about happened in 2017. It doesn't make it a bad chip, far from it, it makes it as good as the competition while being at a significant disadvantage everywhere that matters if you are intel. I'm interested to see it but regardless of your persuasion you cant really argue that this is any more or less than what was absolutely required if you were intel.

Node increase does not = gains from node maturity and given the node took how long for intel to pipe products out on you can bet your arse the node is maturing quickly hence the rapid uptick in what they can do on it now vs just 2 years ago where they failed to release a laptop size chip effectively.

I can't see how this is not dark sorcery. It took AMD 2 years, a 100% increase in power consumption and a node shrink to give us a similar performance increase that Intel did in 1 year with no increase in power consumption on the same node. If that's not impressive then I don't know what impressive actually means. Have we ever - in the history of the cpus, seen such a performance increase in that low amount of time at same power and node? I really cant remember anything similar, it is remarkable

 

[Bencher]

No, not even remotely. Enough with your bullcrap, it's beyond pathetic now. Just stop.

Putting your fingers in yours ears and screaming NO isn't an argument. I already gave my reasoning on why what Intel did is impressive, is there anything you disagree with specifically?

 

[humbug]

With or without windows bug fixed. As shown by capframex, 7950x currently runs games 10-15%slower than it should due to a windows 11 bug

Edit: oh it's not even a real benchmark you guys are hyped for fake YouTube scores lmao

Wasn't there a windows bug that hampered Ryzen 5000 series performance coincidently just as Alder Lake was coming out?

 

[cliffy]

I can't see how this is not dark sorcery. It took AMD 2 years, a 100% increase in power consumption and a node shrink to give us a similar performance increase that Intel did in 1 year with no increase in power consumption on the same node. If that's not impressive then I don't know what impressive actually means. Have we ever - in the history of the cpus, seen such a performance increase in that low amount of time at same power and node? I really cant remember anything similar, it is remarkable

13900 will use more power then 12900 i am sure. The dark sorcery you speak of is that they new AMD was ahead so when ADL repleased they made it a 16 core SKU knowing they could put an extra 8 cores on it and people like ou would be amazed. Plus if you think it only took them a year then you have little to no idea how R&D works

 

[Wolverine2349]

I can't see how this is not dark sorcery. It took AMD 2 years, a 100% increase in power consumption and a node shrink to give us a similar performance increase that Intel did in 1 year with no increase in power consumption on the same node. If that's not impressive then I don't know what impressive actually means. Have we ever - in the history of the cpus, seen such a performance increase in that low amount of time at same power and node? I really cant remember anything similar, it is remarkable

Intel had to put a bunch of e-cores in the design to get 40% multi threaded increase. What kind of increase did the same 8 P core count of Raptor Lake vs 8 P core count of Alder Lake give us?? I doubt close to 40%. Will be impressed if there is a 10% IPC increase at same clock speed for P cores only and they can clock high enough to give 20-25% better performance at same P core count with e-cores shut off and that includes gaming assuming no GPU bottleneck. Will shall see tomorrow.

 

[g67575]

It will be interesting to see how easy Bclk overclocking is on Raptor Lake, and if the performance is good (it didn't seem to work well on 12th gen CPUs in some cases, despite allowing higher frequencies).

Could allow for some nice gains on a 13700/13700f. To some extend though, it's already running at over 5ghz on all P cores, so who the hell cares?

As the last throw of the dice on LGA1700, overclocking cheaper CPUs could be the way to go.

 

[Grim5]

It will be interesting to see how easy Bclk overclocking is on Raptor Lake, and if the performance is good (it didn't seem to work well on 12th gen CPUs in some cases, despite allowing higher frequencies).

Could allow for some nice gains on a 13700/13700f. To some extend though, it's already running at over 5ghz on all P cores, so who the hell cares?

As the last throw of the dice on LGA1700, overclocking cheaper CPUs could be the way to go.

If there are cheap boards with external clock generators others no point

 

[g67575]

True.

What's considered cheap for a LGA1700 board?

Motherboards with long sounding names like 'MSI MAG B660M Mortar Max WiFi' is always a worry...

EDIT - lol, this board is still a pre-order in the UK, what a joke.

I think they are not that appealing generally, unless they can offer an affordable DDR5 board with an external clock generator.

Did any of that stuff come true about the Z790 motherboards being a big improvement over the last gen?

I remember Dave in the 25th century going on about this, as though his life depended on it

 

[Bencher]

Intel had to put a bunch of e-cores in the design to get 40% multi threaded increase. What kind of increase did the same 8 P core count of Raptor Lake vs 8 P core count of Alder Lake give us?? I doubt close to 40%. Will be impressed if there is a 10% IPC increase at same clock speed for P cores only and they can clock high enough to give 20-25% better performance at same P core count with e-cores shut off and that includes gaming assuming no GPU bottleneck. Will shall see tomorrow.

For ΜΤ performance it doesn't matter where the performance increase comes from, that's the whole point of multithreading.

13900 will use more power then 12900 i am sure. The dark sorcery you speak of is that they new AMD was ahead so when ADL repleased they made it a 16 core SKU knowing they could put an extra 8 cores on it and people like ou would be amazed. Plus if you think it only took them a year then you have little to no idea how R&D works

How can it use more power, it's locked to a PL2 of 250w.

 

[cliffy]

How can it use more power, it's locked to a PL2 of 250w.

Same way 125W to 175w (Zen 3 to 4) is twice the power as you say

 

[Bencher]

Same way 125W to 175w (Zen 3 to 4) is twice the power as you say

That's AMD's tdp, which is different to power consumption - that's not a power limit, you already know that. Techpowerup tested it, and in the same workload the 7950x consumed DOUBLE of what the 5950x does.

https://tpucdn.com/review/amd-ryzen-9-7950x/images/power-multithread.png

 

[cliffy]

That's AMD's tdp, which is different to power consumption - that's not a power limit, you already know that. Techpowerup tested it, and in the same workload the 7950x consumed DOUBLE of what the 5950x does.

https://tpucdn.com/review/amd-ryzen-9-7950x/images/power-multithread.png

they have not tested the 13900k yet though hasve they, that could use more power in the same worloads, unless of course 12th gen was already using all the power