Discussion Intel current and future Lakes & Rapids thread

Page 665 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

itsmydamnation

Platinum Member
Feb 6, 2011
2,384
2,286
136
Power, yes. But I was talking about from an iso-process kind of view. Obviously Raptor Lake will be less efficient than Raphael, but thinking longer-term, AMD has been lagging TSMC's latest by a year or two. That works quite well when Intel lags by the same or worse, but seems rather risky to rely on going forward.
I think process improvements these days are well over stated , even back in the glory days of moores law architecture superiority could still hold off a +1 gen process gap. who ever has the most consistent execution of aggressive architecture changes over time will win.
 

pakotlar

Senior member
Aug 22, 2003
721
175
116
Yes, that was the rumor. 8 P + 32 E would take roughly the same die area as 12 P + 16 E. But 8 P + 32 E (48 threads) would be faster in highly multi-threaded tasks than 12 P + 16 E (40 threads).

Lets give the P cores a benchmark that highlights the P-core strength and see what happens.
Rough math assumptions:
  • We'll use the Chips and Cheese libx264 Transcode chart since that favors P cores over E cores.
  • Assume a 125 W processor load.
  • Assume a 10 W uncore with all the various tiles (IO, iGPU, SOC, etc). This is low especially if the iGPU is used, but it gives the most advantage to the P core by giving it more power.
  • Assume that each P core is given double the power of each E core. (Feel free to change this assumption and recalculate as you wish).
12 P + 16 E Math:
  • Each block of 4 P cores uses 23 W and performs 7.7 frames/s according to the Chips and Cheese benchmark.
  • Each block of 4 E cores uses 11.5 W and performs 4.5 frames/s according to the Chips and Cheese benchmark.
  • Total power = 10 W (uncore) + 3 * 23 W (P cores) + 4 * 11.5 W (E cores) = 125 W
  • Total performance: 3 * 7.7 frames/s + 4 * 4.5 frames/s = 41.1 frames/s
8 P + 32 E Math:
  • Each block of 4 P cores uses 19.17 W and performs 6.8 frames/s according to the Chips and Cheese benchmark.
  • Each block of 4 E cores uses 9.6 W and performs 4.1 frames/s according to the Chips and Cheese benchmark.
  • Total power = 10 W (uncore) + 2 * 19.17 W (P cores) + 8 * 9.58 W (E cores) = 125 W
  • Total performance: 2 * 6.8 frames/s + 8 * 4.1 frames/s = 46.4 frames/s
The 8 P + 32 E wins over 12 P + 16 E on the benchmark that preferred P cores. Even if you subtract a few percent for Amdahl's law, 8 P + 32 E still wins.
very nice
 

Exist50

Golden Member
Aug 18, 2016
1,142
1,140
136
I think process improvements these days are well over stated , even back in the glory days of moores law architecture superiority could still hold off a +1 gen process gap. who ever has the most consistent execution of aggressive architecture changes over time will win.
Process improvements might not matter as much as they used to, but they still matter quite a bit, especially for efficiency. Getting +15% perf iso-power or -30% power iso-perf (example from N7->N5) would be really, really difficult via architecture alone. Moreover, between Intel and AMD at least, I'm not convinced the architectural gap is big enough to comfortably afford a full node's worth of buffer.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,380
3,325
136
I think process improvements these days are well over stated , even back in the glory days of moores law architecture superiority could still hold off a +1 gen process gap. who ever has the most consistent execution of aggressive architecture changes over time will win.
Ironically it matters just as much if not more. Intel 4 is 20% and Intel 3 is 18%. And that's at iso-power at a transistor level. Because it gets harder and harder to squeeze out the last bit out, the winners get rewarded even more. It's like comparing olympic sprinters versus regional ones.

Nothing has really changed. If you go too aggressive you still risk delaying and failing. When you read about microchip architects they aren't called "architects" for no reason. These are seasoned, very experienced veterans at their work and not easily displaced. The difficulty in making it work are no different from other areas such as making airplanes or buildings.

Once you understand that you understand how stupid and disconnected with the core of the company at Intel has been to fire experienced employees and engineers. It's almost comparable to ripping out foundational structures in the building you are in to sell and get extra cash!
 

Doug S

Golden Member
Feb 8, 2020
1,324
1,967
106
My guess is never. Apple is more likely to do there own fab than AMD. It takes huge volume to make it worthwhile.

Apple doesn't have near the volume to make doing their own fab worthwhile, even if they had the engineers, site and confirmed ASML orders ready to go.

There's a reason Intel is becoming a foundry, because they cannot make the numbers work fabbing only their own chips anymore.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,380
3,325
136
My guess is never. Apple is more likely to do there own fab than AMD. It takes huge volume to make it worthwhile.
Apple became what they are because they are hyper-focused at what they do. They are big so you think they could switch with a flick of a finger? But their DNA is with the pretty, expensive computers aimed at client. Even before Steve it was true. I can see why it took them until Smartphones to become successful. It just wasn't their time, just like the Newton Tablet back in the late 80s that failed miserably. They always had it. Like with art that is only appreciated hundreds of years later.

It will not be easy for them to suddenly mass manufacture server chips or fab.

It's not about technology, or patents, or ideas, but the people that make it happen.

Like what they said with Microsoft, that everything was built around having Windows succeed. And with Intel they made sure everything else was done to have their CPUs succeed. Getting out of that requires a great visionary and risk taking along with flawless execution.

Did you know it took Intel over a decade to get the server marketshare they have now? It did not happen overnight. x86(and of course Intel) servers took a small fraction slightly before Core 2 generation.

Companies represent many thousands of people but their culture is very homogenously driven, especially when they are very successful.
 
Last edited:
  • Like
Reactions: Tlh97 and ftt

DrMrLordX

Lifer
Apr 27, 2000
19,948
8,919
136
How about co-owning a fab with TSMC?
TSMC probably wouldn't do that for anyone. As you can see with Intel laying out cash for N3 wafers, TSMC has essentially forced Intel to expand TSMC's capacity in exchange for a wafer allocation. That's how TSMC prefers to do business. There are a few instances where TSMC has behaved differently (see: Sony/TSMC/Denso cooperating to make an automotive chip fab), but not many that I know of.

Speaking of which

With other companies reducing their orders, TSMC may find itself with a glut of N3 wafer capacity. Wonder if Intel will take more to pad out their CPU lineup?
 

DrMrLordX

Lifer
Apr 27, 2000
19,948
8,919
136
TSMC does joint subsidiaries like JASM in Japan with Sony. But Sony is only a minority shareholder and the fab will only work with the older 22/28nm nodes.
Yeah I think that's the arrangement with Denso. Regardless it does not appear that TSMC will allow anyone to control even partially one of their fabs with a cutting-edge node in it. Intel certainly does not have any arrangements like that with TSMC, nor does AMD nor Apple.
 

Doug S

Golden Member
Feb 8, 2020
1,324
1,967
106
How about co-owning a fab with TSMC?
What's the benefit? Apple almost exclusively needs leading edge wafers. Yeah they need stuff like PMICs, MEMS devices and so forth, but those are mostly commodity items.

If Apple co-owned an N7 fab with TSMC today, what the heck would they do with it now that they have little need for N7 wafers? You suggesting they should partner with TSMC to operate that one fab as a mini foundry?

This makes no sense, give it up. Intel is and will remain the only major CPU supplier that owns and operates its own fabs, and only because they realized a half decade ago they'd have to become a foundry to do so. After a failed attempt at starting a foundry business, they seem to be on the right path now to become competitive in the foundry market. If not for that, they'd have to sell off their fabs by the end of the decade.
 
  • Like
Reactions: Tlh97 and scannall

jpiniero

Lifer
Oct 1, 2010
12,221
3,632
136
After a failed attempt at starting a foundry business, they seem to be on the right path now to become competitive in the foundry market. If not for that, they'd have to sell off their fabs by the end of the decade.
Again, it has more to do with how well they can get the nodes working (and to an extent how much money they can weasel out of the US/Europe) than anything special about being a foundry.
 

Doug S

Golden Member
Feb 8, 2020
1,324
1,967
106
Again, it has more to do with how well they can get the nodes working (and to an extent how much money they can weasel out of the US/Europe) than anything special about being a foundry.

No they MUST become a foundry if they want to keep their fabs, because they don't sell enough CPUs to compete in the future. Every new process generation is more expensive to develop, the fabs are more expensive to build, the equipment is more expensive to buy. This math is why the foundry industry has consolidated down to two players (three once Intel is in it for real) at the leading edge.

You can only manage those ever escalating fixed costs if you run more wafers every year, and while the PC market was growing Intel did. Unfortunately the PC market isn't growing anymore, so if they stay in client/server CPUs those fixed costs become a larger and larger portion of their per wafer cost and thus per CPU cost. Meanwhile TSMC is making leading edge chips for the large majority of everyone else in the entire world and thus has a lot more wafers to spread out their roughly similar fixed costs.

To run more wafers Intel's first thought was to develop more "Intel inside" lines of business, i.e. move beyond client/server CPUs. They tried to enter the mobile SoC business, that flopped. They talked about "IoT", and people laughed at the idea anyone would want x86 for IoT. They've done NAND, then Optane, but those are commodity products they don't get the per wafer pricing CPUs do. They bought an FPGA company, which did bump their wafer count and allowed them to create a niche for FPGA capabilities in server. Lately they are dabbling in discrete GPUs (too soon to tell) and crypto (they got in right as the bubble burst)

In short, they've tried every strategy under the sun to push out more Intel branded silicon but nothing has worked, at least not to the degree it needs to work as far as increasing wafer runs. So they are forced into selling non-Intel branded silicon, i.e. becoming a true foundry.
 

Doug S

Golden Member
Feb 8, 2020
1,324
1,967
106
I was talking about AMD co-owing a fab with TSMC and not being wafer supply constrained as a result. I think Apple has enough money to challenge TSMC if they really wanted to.
Given that reports are that AMD (along with Apple and Nvidia) are cutting orders at TSMC, it doesn't seem like they are wafer supply constrained after all.
 

jpiniero

Lifer
Oct 1, 2010
12,221
3,632
136
Given that reports are that AMD (along with Apple and Nvidia) are cutting orders at TSMC, it doesn't seem like they are wafer supply constrained after all.
The AMD part is a little fishy. Even if they were cutting APU and GPU orders, you would think they would be able to shift to more Milan and consoles.

If Milan suddenly became a lot more available, that would be real bad news for Intel's server business, you would think.
 
Jun 24, 2022
33
36
46
My twenty cents:

If you upgrade every other gen, only purchase the odd number chipsets.

Z390 Z590 Z790 Z990 Z1190

Coffee Rocket Raptor Arrow Nova

I don't even know which is tik and which is tok anymore, but the even number chipsets seem to always introduce brand new tech, often times not anywhere near ready to enjoy with so many issues attached ie: PCIe 5.0 SSDs and DDR5 Alder Lake - and the odd number chipsets are always a "refinement" of the previous even number.

If it's the right time to buy, of course get whatever you need, but I placed myself on an upgrade pathway that buys every other chipset and ONLY the odd number gens.

So far it's working out extremely well.

Cheers! :)

Z390 Z590 Z790 Z990 Z1190
 
Last edited:

igor_kavinski

Diamond Member
Jul 27, 2020
4,727
2,976
106
Should Intel skip 13 and call Raptor Lake 14th gen for good/better luck? My residential building does not have a 13th floor. At least, not that I know of. Haven't really taken the stairs to investigate.
 
  • Haha
Reactions: AlltheWay LeeWhy
Jun 24, 2022
33
36
46
Is it? 11th gen was extremely poor.
RKL was extremely poor, unless you grabbed an ROG Z590 Apex motherboard and a small number of very QVL specific high-speed mem kits. ;)

Haters gonna hate, I guess. :rolleyes:

5066 CL17, 5333 CL19, 5600 CL20, 5866 CL21 all possible with ambient air.

11600K 5.5Ghz.jpg
5866Mhz CL22 1T.jpg
AIDA 5066 17 c.jpg
 
Last edited:
  • Like
Reactions: pcp7
Jun 24, 2022
33
36
46
Hint, look at their signature. Intel only and already going Raptor Lake. Intel could sell poo and some people would buy it. Like Rocket Lake.
I've ran AMD graphics cards for the last 10years. Only recently moved to an Nvidia workstation card because the AMD update from my AMD WX-4100 only uses 4lanes and 4GB of memory. And honestly, it's kinda ugly.

1656865030208.png

Whereas NVidia is offering an SFF workstation card with 12GB of memory and uses all 16PCIe lanes, also the Nvid card looks more professional to my eyes.

40931_1Z.jpeg

I was also kinda getting a bit tired of the AMD "way too blue" aesthetic after 5.5years of use, although it's an amazing workstation card, but Polaris architecture? Kinda long in the tooth. :)

Used AMD graphics for 10years and AMD workstation graphics for 5years.

Before and After pics:

IMG_2900.JPGIMG_0506.JPG
 
Last edited:

ASK THE COMMUNITY