Apple A14 - 5 nm, 11.8 billion transistors

[Eug]

A14 - First announced for iPad Air.

 

[trexfromouterspace]

They tape out chips with a hundred bugs on silicon and work on it for 15 months to bring that chip to market

If it was only 15 months they would still be competitive. Ice Lake was stuck taping out for around 30 months.

 

[dmens]

If it was only 15 months they would still be competitive. Ice Lake was stuck taping out for around 30 months.

That is what I have been saying all along. Intel design teams can only compete with a three year process lead. Without that lead, that are not even in the game.

 

[Thala]

AMD faces the same x86 cruft, that has not stopped them from making progress.

And yet AMD Ryzen 5000 is not much ahead of Intel Willow Cove with neither, performance per clock as well as frequency. So based on what metric do you think AMD is much ahead of Intel with respect to core architecture?
Evidence shows that AMD is much ahead with respect to actual products, but not so much if it comes down to core architecture.

 

[naukkis]

And yet AMD Ryzen 5000 is not much ahead of Intel Willow Cove with neither, performance per clock as well as frequency. So based on what metric do you think AMD is much ahead of Intel with respect to core architecture?
Evidence shows that AMD is much ahead with respect to actual products, but not so much if it comes down to core architecture.

Zen3 is about half the size and power use of Willow Cove. Intel's arch is very inefficient in all metrics.

 

[Thala]

Zen3 is about half the size and power use of Willow Cove. Intel's arch is very inefficient in all metrics.

Where do you have power consumption figures for a single core from? The size is supposed to be larger due to AVX512 - if when usable - Intel core significantly outperforming Zen3 cores. So not everything is black and white.

 

[dmens]

Where do you have power consumption figures for a single core from? The size is supposed to be larger due to AVX512 - if when usable - Intel core significantly outperforming Zen3 cores. So not everything is black and white.

AVX-512 is a gimmick, and it doesn't even take up that much die space. AMD has a large PPA (power/perf/area) lead. If you believe Intel's claims that their 10nm is just as good as TSMC 7nm (and that isn't really that bad of a claim, just that TSMC 7nm is actually in profitable production, and Intel 10nm is not), then Renoir beating Tigerlake on everything that matters proves AMD core is superior on all three metrics. Zen 3 against Rocketlake is at least 2x (if not more) on perf/power. In this business that is a multi-generational advantage.

 

[Thala]

AVX-512 is a gimmick, and it doesn't even take up that much die space. AMD has a large PPA (power/perf/area) lead. If you believe Intel's claims that their 10nm is just as good as TSMC 7nm (and that isn't really that bad of a claim, just that TSMC 7nm is actually in profitable production, and Intel 10nm is not), then Renoir beating Tigerlake on everything that matters proves AMD core is superior on all three metrics. Zen 3 against Rocketlake is at least 2x (if not more) on perf/power. In this business that is a multi-generational advantage.

I was asking for evidence, not for more of the same claims. I am also not asking about comparing SoCs like Tigerlake and Renoir from the perspective of what matters to you.

 

[dmens]

I was asking for evidence, not more of the same claims. I am also not asking about comparing SoCs like Tigerlake and Renoir from the perspective of what matters to you.

That's rich, asking for evidence. Go look at a die shot and figure it out. Go look at the benchmarks and compare the power expended per core to achieve said performance. On that note, when people say things like, "Intel core is bigger because all because of AVX-512", I just ignore it instead of asking for evidence. Because it is a waste of time to address ignorance.

As for SoC, you do realize it is the same core inside the laptop parts and the desktop parts, right? I have no idea what your perspective is, or what matters to you, apparently things like single-thread/multi-core perf for power expended isn't one of them? What do you care about then?

 

[Thala]

On that note, when people say things like, "Intel core is bigger because all because of AVX-512", I just ignore it instead of asking for evidence. Because it is a waste of time to address ignorance.

You might be able to understand, that i did not say this (the part in quotes) at all. All i was asking is about evidence for the claimed factor 2 discrepancy in area and power.

As for SoC, you do realize it is the same core inside the laptop parts and the desktop parts, right? I have no idea what your perspective is, or what matters to you, apparently things like single-thread/multi-core perf for power expended isn't one of them? What do you care about then?

Look, i was asking about core architecture and you bring examples of SoC multi-core perf, where the number of cores clearly are different. Renoir is not even using the core architecture we compare to. I already acknowledged that AMD products are superior at the moment - it is pointless to bring this up again and again.

What i care about? At the moment i am caring about evidence of the above mentioned factor 2 difference in area and power between Willow Cove and Zen 3 - i thought i made this very clear.

 

[naukkis]

Where do you have power consumption figures for a single core from? The size is supposed to be larger due to AVX512 - if when usable - Intel core significantly outperforming Zen3 cores. So not everything is black and white.

Die size difference isn't coming from AVX512, from published data we know that Intel Cove's have much bigger main structures like reorder buffer and register files. Increasing core size is fine if performance scales as well but as AMD can meet Cove's IPC and frequency it means that Cove's are just inefficient designs. Like Papermaster said, they could also increased performance more if they wanted by just using more transistors to problems - but that inefficient way to make cpu core. And Intel seems to be in that path.

 

[Thala]

Die size difference isn't coming from AVX512, from published data we know that Intel Cove's have much bigger main structures like reorder buffer and register files.

All i am asking is for the "published data", where you are deriving a factor 2 advantage in area and power from. I was not asking under which condition you think something is fine or if Mark Papermaster said that the expense of more gates leads to higher performance - which is trivial.
That having said, i do believe that Intel has disadvantages in area efficiency...but at the moment i am just asking for these "factor 2" numbers you are saying you have.

 

[naukkis]

All i am asking is for the "published data", where you are deriving a factor 2 advantage in area and power from.

It's well known that core+l2 for Zen2 core is 3.7mm2, and for Sunny Cove 6.91mm2. Zen3 is about same size than Zen2, Willow Cove is bigger than Sunny Cove. So there we are.

And for active power std manufacturing cpu's switching transistors per cycle is about same as power- and having main structures almost twice as big pretty much guarantees that power use in same process will be about twice as much. So switching to TSMC won't save Intel before they design more efficient uarch.

 

[Thala]

For your first statement (area) i was asking for a source - like a die analysis.

The second statement regarding power is wrong, because you assume the same activity - which is ridiculous. For starters, if a gate is not switching, there will be no switching capacity to charge.

 

[itsmydamnation]

A single zen2 core/thread in my 4700u which has a 25watt sustained TDP

will do:
GB5 / cpu-z bench / other "light" benchmarks sub 5 watts peak @ 4.2ghz , ~ 3 watt sustained
cinebench R20 , prime 95 smallest FFT sub 9 watts peak @ 4.2 , ~ 8 watt sustained
prime 95 small FFT 15watts peak @ 4.2 ~ 13 watt sustained

Zen 3 is a solid 25% better perf per watt @ 4.0ghz then Zen2

Tigerlake is about 30% better 1T perf then Renoir but pushes right up to 50watt's single core power consumption to maintain 4.8ghz
so about a 12.5% clock advantage and a 18% IPC advantage.

You only have to look at the constancy of AMD laptops CPU performance compared to the all-over the place of intels laptops performance to see this play out. laptop has bad cooling tigerlake performance ends up very disapointing , for AMD this doesnt happen for CPU performance, only really CPU+GPU perf shows the effects of bad cooling designs.

its hard to do any super close comparison because the clock controls of desktop don't exist.

 

[IntelUser2000]

Tigerlake is about 30% better 1T perf then Renoir but pushes right up to 50watt's single core power consumption to maintain 4.8ghz
so about a 12.5% clock advantage and a 18% IPC advantage.

This is wrong BTW.

RECENZE: První Test 11. generace procesoru Intel Core - Acer Swift 5 (SF514-55T)

Do redakce jsme dostali předprodukční kus ultratenkého 14'' notebooku Acer Swift 5, díky kterému bylo možné otestovat nejnovější počin v rámci procesorů od Intelu. Pod tlakem nedávných vlastních neúspěchů a nyní také silnější konkurence se musí Intel snažit, aby nepřišel o své dominantní...

notebook.cz

20W for max single thread and 17W for 4.6GHz. 50W maybe for a short period during multi-threaded workloads.

Intel is making a mistake focusing on such high frequency for Tigerlake, but that's a whole different matter.

Zen3 is about same size than Zen2, Willow Cove is bigger than Sunny Cove. So there we are.

Actually you don't know this. The CCD grew from 74mm2 to 80.7mm2. The cache does not change so the 9% growth is nearly entirely due to core size increase. The ~20% increase or so is a fantastic effort considering we get about 20% increase in performance as well. But to say its the same when we're counting millimeters is silly.

 

[itsmydamnation]

This is wrong BTW.

RECENZE: První Test 11. generace procesoru Intel Core - Acer Swift 5 (SF514-55T)

Do redakce jsme dostali předprodukční kus ultratenkého 14'' notebooku Acer Swift 5, díky kterému bylo možné otestovat nejnovější počin v rámci procesorů od Intelu. Pod tlakem nedávných vlastních neúspěchů a nyní také silnější konkurence se musí Intel snažit, aby nepřišel o své dominantní...

notebook.cz

20W for max single thread and 17W for 4.6GHz. 50W maybe for a short period during multi-threaded workloads.

Intel is making a mistake focusing on such high frequency for Tigerlake, but that's a whole different matter.

Sry took that data from Anandtech's review as its hard to find reviews that focus on 1t power consumption.

• The peak frequency of 4800 MHz is valid only when one core is loaded
• The peak frequency is 4300 MHz when more than one core is loaded
• The maximum power draw in the turbo mode is ~50 W*

So i guess isnt concise in this case

 

[IntelUser2000]

Sry took that data from Anandtech's review as its hard to find reviews that focus on 1t power consumption.

Can you tell me how the power is distributed? How much is the SoC and how much is the core out of the 8W running R20?

Intel chips have a rough 1-3W fixed power use outside of IA cores and the graphics. So even if the core is at 0W, its still using 1.xW

 

[itsmydamnation]

Can you tell me how the power is distributed? How much is the SoC and how much is the core out of the 8W running R20?

Intel chips have a rough 1-3W fixed power use outside of IA cores and the graphics. So even if the core is at 0W, its still using 1.xW

so its a bit hard as i have so much background stuff running, so my "idle" is ~10% cpu which has about 1 watt in core usage and ~3watts uncore

During R20 if i pin 1T to a single core that core is siting in the 8~9 watt range ( tile complexity?) and the uncore stays about 3watts in usage.

 

[Eug]

They are falling off a cliff faster than Kodak: at least Kodak maintained their technical dominance of traditional film to the bitter end, Intel is already second place or worse in every CPU market.

Off topic but that’s not really a valid comparison.

Film pretty much died in short order. Ironically Kodak was one of the first to digital but created either niche high end products that were horrendously expensive or else lower end products that were sub par. For some reason they never scaled down the high end stuff to mainstream prosumer levels, and their low end stuff couldn’t be scaled up and in fact eventually ended up being the low end of the mainstream low end. I suspect part of the reason was to protect its dying film business.

Meanwhile microprocessor sales is actually going up slightly. Intel may be losing ground but unlike film, it is not a dying market obviously.

What would you rather be, 2nd place in a vibrant market, or 1st place in a dead market? Or how about 1st place in a vibrant market? Cuz that latter category is where Intel still is. Intel may not be as dominant as they used to be in some regards, but they are still solidly number 1 overall. In fact, Intel’s chip revenue is still roughly twice that of TSMC.

Now I know you are trying to talk about technical prowess instead of sales revenue but what I see said about technical prowess in large part seems like vigorous hand waving. Intel in the past may have underestimated competitors such as AMD and Apple and TSMC, but I think none of those companies is stupid enough to underestimate Intel.

 

[smalM]

- Intel apparently made a decision around 10 years ago to prioritize making money over doing research and development. Everything since then has been on the cheap.

Nice narative, but not true.
Just the last 4 years Intel spent more than $12 Bn p.a. in R&D.
They are not cheap, they are inefficient.

 

[dmens]

Off topic but that’s not really a valid comparison.

Film pretty much died in short order. Ironically Kodak was one of the first to digital but created either niche high end products that were horrendously expensive or else lower end products that were sub par. For some reason they never scaled down the high end stuff to mainstream prosumer levels, and their low end stuff couldn’t be scaled up and in fact eventually ended up being the low end of the mainstream low end. I suspect part of the reason was to protect its dying film business.

Meanwhile microprocessor sales is actually going up slightly. Intel may be losing ground but unlike film, it is not a dying market obviously.

What would you rather be, 2nd place in a vibrant market, or 1st place in a dead market? Or how about 1st place in a vibrant market? Cuz that latter category is where Intel still is. Intel may not be as dominant as they used to be in some regards, but they are still solidly number 1 overall. In fact, Intel’s chip revenue is still roughly twice that of TSMC.

Now I know you are trying to talk about technical prowess instead of sales revenue but what I see said about technical prowess in large part seems like vigorous hand waving. Intel in the past may have underestimated competitors such as AMD and Apple and TSMC, but I think none of those companies is stupid enough to underestimate Intel.

x86 microprocessors is a (slowly) dying market. Supply chain as opposed to technical prowess seems to be the primary deciding factor for the big money sales in the x86 CPU business. But that is precisely where Intel is stumbling. It is just a matter of time.

 

[Eug]

x86 microprocessors is a (slowly) dying market. Supply chain as opposed to technical prowess seems to be the primary deciding factor for the big money sales in the x86 CPU business. But that is precisely where Intel is stumbling. It is just a matter of time.

Maybe, maybe not. Compared to AMD, Intel today has MUCH, MUCH higher market share than it did 15 years ago.

AMD has been gaining in the past 4 years, but right now AMD is only about where it was a decade ago or so, because AMD had been losing market share for all that time in between.

Does the past 4 years of declining market share for Intel mean that the same will be true in the next years? Possibly, but I'm not convinced.

"Past performance is not indicative of future results."

 

[Hitman928]

Maybe, maybe not. Compared to AMD, Intel today has MUCH, MUCH higher market share than it did 15 years ago.

AMD has been gaining in the past 4 years, but right now AMD is only about where it was a decade ago or so, because AMD had been losing market share for all that time in between.

Does the past 4 years of declining market share for Intel mean that the same will be true in the next years? Possibly, but I'm not convinced.

"Past performance is not indicative of future results."

AMD is currently capturing 20% of the x86 market and I expect AMD will be in a position to really increase market share when they transition to 5 nm. They will obviously have their leading edge products on 5 nm but I expect they will still be producing a lot of 7 nm CPUs which will still be very competitive for a little while at least. Then they will fill the really budget range with continued 12 nm products, perhaps porting to GF's improved 12 nm process. They'll be able to spread out volume between 3 process lines and really capture a significant portion of the market share. Intel would be able to compete in the budget segment just fine, but if 7 nm continues to give them problems, they won't really be able to stop AMD from continuing to capture the mid to high end markets in consumer, business, or server.

 

[Eug]

AMD is currently capturing 20% of the x86 market and I expect AMD will be in a position to really increase market share when they transition to 5 nm. They will obviously have their leading edge products on 5 nm but I expect they will still be producing a lot of 7 nm CPUs which will still be very competitive for a little while at least. Then they will fill the really budget range with continued 12 nm products, perhaps porting to GF's improved 12 nm process. They'll be able to spread out volume between 3 process lines and really capture a significant portion of the market share. Intel would be able to compete in the budget segment just fine, but if 7 nm continues to give them problems, they won't really be able to stop AMD from continuing to capture the mid to high end markets in consumer, business, or server.

I fully expect AMD to continue to capture additional x86 market share in the near term, and I agree that TSMC's 5 nm is key to AMD's success.

However, beyond that, the assumption that Intel will continue to slide is pretty much dependent on TSMC's ability to hit 3 nm quickly, and Intel's inability to make 7 nm (and smaller) work.

That may happen, but there are no guarantees.

Maybe I am being naive (which could very well be true since I'm not a chip geek), but I see this as more of a process node race, and not really a CPU design race. Hell, I could even see Intel using TSMC to manufacture chips if absolutely necessary.

 

[dmens]

Maybe, maybe not. Compared to AMD, Intel today has MUCH, MUCH higher market share than it did 15 years ago.

AMD has been gaining in the past 4 years, but right now AMD is only about where it was a decade ago or so, because AMD had been losing market share for all that time in between.

Does the past 4 years of declining market share for Intel mean that the same will be true in the next years? Possibly, but I'm not convinced.

"Past performance is not indicative of future results."

You are about to get a glimpse of the future tomorrow. Stay tuned.