Intel’s surprise Ryzen killer

[kitfit1]

This is why im really failing to get excited at the moment. Keller left Intel quickly and seemingly under slightly odd circumstances which is also slightly concerning, you hope we see some of his best work but I wouldn't be surprised if the premature exit could have been around him not being given the flexibility within his position to do his best work. I hope I am wrong on everything I say but genuinely how it looks right now id be surprised if we see anything special out of intel before 2023/2024 unless of course they find some momentum and start executing.

Yep, Keller leaving Intel as quick as he did tells peeps a lot more when they actually think about it. In fact, Adored TV did an inside video on the way that Intel actually works. One of the most striking things i remember from it was that very few people that "work" for Intel, actually work for them at all. Most are contract employed or are zero hour employed. None of those people would have any "yearning" to actually come up with innovative ideas, and in fact that came across loud and clear on the video. I'm not talking about factory workers, but peeps that work in R&D, engineering and software. These sort of posts need motivated, dedicated and well paid people to drive any company forward. It seems obvious to me that Intel has ignored these guiding principles for years and years. Maybe Keller thought Intel would be as innovative as AMD and drive things forward as well, it looks to me as if he found the exact opposite and just decided "sod it, i'm off"

 

[Vince]

I'm pretty sure Geona will kill any catch up Intel have in 2022, after all it's almost an entirely new architecture, and I agree 23-24 seems to be the point they'll get to 7nm and a competitive product stack, at which point AMD will be all but moving to 3nm.

I know its bizarre the massive problem for intel appears to be that amd continue delivering like they are, big wins as well, its concerning. If we keep seeing the same sort of crafty ipc upticks with their ability to make multiple tiny efficient designs side by side and just bashing loads of stuff out on the same substrate you can barely see them slowing down. You can literally see how they now have this design advantage that quite simply makes designing a new architecture much easier because you only need to focus on the cores, cache and interconnect, the rest is in the IO die which remains untouched. You literally end up separating things out more which gives you that ability to properly focus in.

 

[Troezar]

I know its bizarre the massive problem for intel appears to be that amd continue delivering like they are, big wins as well, its concerning. If we keep seeing the same sort of crafty ipc upticks with their ability to make multiple tiny efficient designs side by side and just bashing loads of stuff out on the same substrate you can barely see them slowing down. You can literally see how they now have this design advantage that quite simply makes designing a new architecture much easier because you only need to focus on the cores, cache and interconnect, the rest is in the IO die which remains untouched. You literally end up separating things out more which gives you that ability to properly focus in.

You talk a lot of sense.

 

[4K8KW10]

I know its bizarre the massive problem for intel appears to be that amd continue delivering like they are, big wins as well, its concerning. If we keep seeing the same sort of crafty ipc upticks with their ability to make multiple tiny efficient designs side by side and just bashing loads of stuff out on the same substrate you can barely see them slowing down. You can literally see how they now have this design advantage that quite simply makes designing a new architecture much easier because you only need to focus on the cores, cache and interconnect, the rest is in the IO die which remains untouched. You literally end up separating things out more which gives you that ability to properly focus in.

AMD needed 10 quarters in order to increase its servers market share from 0.8% to 5.8%.
With this growth speed, AMD will need 88 quarters or 22 years only to reach 50-50 share parity with Intel..

https://www.tomshardware.com/news/amd-vs-intel-highest-overall-x86-chip-market-share

 

[Vince]

AMD needed 10 quarters in order to increase its servers market share from 0.8% to 5.8%.
With this growth speed, AMD will need 88 quarters or 22 years only to reach 50-50 share parity with Intel..

https://www.tomshardware.com/news/amd-vs-intel-highest-overall-x86-chip-market-share

You talk like these things happen in some sort of linear fashion. I promise you many massive companies have failed in areas because they get overtaken, make poor decisions and couldn't catch back up. That wont happen to intel any time soon but never say never.

 

[Roadrunner_668]

it happened to AMD and they caught back up !

 

[smilingcrow]

Here's another review showing gaming power consumption:

That's still only two reviews but the plot thickens:
https://hexus.net/tech/reviews/cpu/145984-intel-core-i9-10850k/?page=9
and
https://www.techpowerup.com/review/intel-core-i9-10850k/18.html

 

[RavenXXX2]

Gaming is mostly driven by the RTX 2080 Ti graphics card. Reason enough we don't see a large deviation between results.

 

[smilingcrow]

"Gaming is mostly driven by the RTX 2080 Ti graphics card. Reason enough we don't see a large deviation between results."

We know that, but what is surprising is that in both those reviews the i9-10850K is consuming less than the 3600XT.
Maybe both games are just especially light on the CPU!

 

[LePhuronn]

There is no way in hell 10 Skylake Comet Lake cores draws less power than 8 Skylake Coffee Lake cores in the same game. Those 10th Gen results just don't sit right to me.

 

[Mr Oblong]

Consumer mainstream chips with 1700 pins, it looks expensive! Considering what amd are doing with just 1331 pins on the mainstream im asking myself if this design choice is smart. It appears to me that it's basically a hedt size chip to compete in the mainstream where traditionally you would have chips around that 1200 to 1400 pins. I really want them to be making compelling stuff but im not sure this is the way to go for mainstream sockets. It's interesting and im looking forward to seeing and getting my hands on it to play with but one thing it isn't looking right now is that compelling.

If the price-performance is right, I'm cool with it. Aren't you?

 

[Mr Oblong]

it happened to AMD and they caught back up !

Exactly, and Intel have so much money. That alone gives them options. Intel aren't done by a long shot.

 

[Mr Oblong]

I know its bizarre the massive problem for intel appears to be that amd continue delivering like they are, big wins as well, its concerning. If we keep seeing the same sort of crafty ipc upticks with their ability to make multiple tiny efficient designs side by side and just bashing loads of stuff out on the same substrate you can barely see them slowing down. You can literally see how they now have this design advantage that quite simply makes designing a new architecture much easier because you only need to focus on the cores, cache and interconnect, the rest is in the IO die which remains untouched. You literally end up separating things out more which gives you that ability to properly focus in.

That makes sense. AMD are getting a LOT of mileage out of their designs, and it seems like a really efficient way to go in terms of investment.

 

[humbug]

Here's another review showing gaming power consumption:

That's still only two reviews but the plot thickens:
https://hexus.net/tech/reviews/cpu/145984-intel-core-i9-10850k/?page=9
and
https://www.techpowerup.com/review/intel-core-i9-10850k/18.html

My 3600 barely uses 40 Watts and that's in a high load game like Star Citizen, in Left4Dead its about 10 Watts because the CPU isn't doing anything.

I think this is probably accurate, this is a workload where the Uncore probably uses more power than the cores, i understand the argument here is about gaming power consumption, but 10 watts vs 15 watts is not the whole story and if it were to be used like that its incredibly misleading.

In reality where the CPU is loaded up properly Intel's efficiency is very poor, using well over 200 Watt, compared with about 150 watts of the faster 12 core 3900X, and the performance per watt comparison is about to get even worse for Intel.

 

[Vince]

If the price-performance is right, I'm cool with it. Aren't you?

It's not that I am not ok with it but more pins as we know by default means more cost, or less profit for Intel if they choose that. It might mean they can be competitive short term but what it doesn't do is keep them in the game long term. Spiralling costs producing bigger and bigger monolithic dies for a large socket just doesn't seem all that sensible.

 

[Rroff]

There is no way in hell 10 Skylake Comet Lake cores draws less power than 8 Skylake Coffee Lake cores in the same game. Those 10th Gen results just don't sit right to me.

Looks like the boost is more aggressive and more responsive on the 10850, ramping quicker and spending less time at max clocks (though sometimes throttling) while the 9900K has lazier boost behaviour in that regard spending more time in a higher power state.

 

[Mr Oblong]

It's not that I am not ok with it but more pins as we know by default means more cost, or less profit for Intel if they choose that. It might mean they can be competitive short term but what it doesn't do is keep them in the game long term. Spiralling costs producing bigger and bigger monolithic dies for a large socket just doesn't seem all that sensible.

Oh I agree with you there, they're having to contort their product stack to meet the competition and it's not a good position to be in, nor does it project confidence.

However, I expect Intel will take a profit squeeze rather than lose more market share than necessary. We'll see I suppose.

 

[LePhuronn]

Looks like the boost is more aggressive and more responsive on the 10850, ramping quicker and spending less time at max clocks (though sometimes throttling) while the 9900K has lazier boost behaviour in that regard spending more time in a higher power state.

Good shout

 

[Rroff]

Good shout

Doesn't entirely explain it though. I think the 10850 must be throttling quite a bit.

 

[Armageus]

AMD needed 10 quarters in order to increase its servers market share from 0.8% to 5.8%.
With this growth speed, AMD will need 88 quarters or 22 years only to reach 50-50 share parity with Intel..

https://www.tomshardware.com/news/amd-vs-intel-highest-overall-x86-chip-market-share

Or maybe the uptake might start escalating as all of those companies who have had money sat waiting for new Intel products for the last 3 years finally get bored and are forced to buy AMD