Discussion Intel current and future Lakes & Rapids thread

[naukkis]

Since ice lake core has less transistors compared to zen2 yet higher IPC i'll take it with ridiculously compex both of you mean ridiculously good. And any modern CPU is "ridiculously complex" compared to an 8086.

As mentioned by OriAr getting 18% extra performance out of 38% extra transistors is an outstanding result.

Their 10nm process is broken, but their architecture certainly isnt. Very excited to see what they have in store with Willow and especially Golden Cove.

What? Icelake cores are about twice as big as Zen2. Zen2 has more cache which drives transistor count up, but last level cache won't consume much power and won't take a lot of silicon space. Core logic does - and core's power usage is directly comparable how many transistors are switching.

And it's different thing to get IPC up than get extra performance. Complex, big archs clocks lower than simpler ones so getting extra performance from extra complexity isn't guaranteed. And sure Icelake isn't bringing extra performance, specially multi-core performance is plain poor.

 

[lobz]

Since ice lake core has less transistors compared to zen2 yet higher IPC i'll take it with ridiculously compex both of you mean ridiculously good. And any modern CPU is "ridiculously complex" compared to an 8086.

As mentioned by OriAr getting 18% extra performance out of 38% extra transistors is an outstanding result.

Their 10nm process is broken, but their architecture certainly isnt. Very excited to see what they have in store with Willow and especially Golden Cove.

No. We meant ridiculously complex. We were talking about Ice Lake in comparison to Skylake and how it really failed to bring any meaningful efficiency improvement over it.
If you wanna drag AMD and Zen into this, you can always start another thread.

 

[OriAr]

What? Icelake cores are about twice as big as Zen2. Zen2 has more cache which drives transistor count up, but last level cache won't consume much power and won't take a lot of silicon space. Core logic does - and core's power usage is directly comparable how many transistors are switching.

And it's different thing to get IPC up than get extra performance. Complex, big archs clocks lower than simpler ones so getting extra performance from extra complexity isn't guaranteed. And sure Icelake isn't bringing extra performance, specially multi-core performance is plain poor.

The reason ICL's performance is not as good as it could have been has more to do with the state of 10nm and how much Intel manage to get out of 14nm way more than any reason laid in architecture (ICL does generally beat WHL at the same power in MT benchmarks, although obviously by less than 18%).
Sunny Cove is nowhere near double the size of Zen 2, without L3$ and FIVR the Sunny Cove core stands around 5 mm^2, which is around 38% bigger than the Zen 2 core, not exactly surprising considering SNC on average has 18% better IPC.
To claim Sunny Cove is an inefficient core is quite not true, at least it's not any more inefficient than the rest.

 

[IntelUser2000]

Is there any other info source aside from what they said at CES 2020?

No, other than the "double digit gains". Based on leaks they are keeping it extremely close to their chest. Also if it wasn't for the virus hype we'd have seen Xe arch reveal this week.

@OriAr Can't compare transistor count with Zen 2, since that's based on the whole chip. I assume Sunny Cove is quite a bit larger, but not much as some are saying. I also assume for Core chips their 10nm process have about 10-15% area deficiency compared to TSMC 7nm. When they say its competitive in area compared to foundries, its most likely with Tremont cores and other parts of the chip. Core has historically scaled 50% per process generation and no marketing talk has changed that.

 

[Adonisds]

Coffee Lake has 217 million transistors per core, while Ice Lake has 300 million. 300/217 = 1.38 > ICL has 38% more transistors per core in it than CFL.
For comparison, Zen 2 has 4 billion transistors per CCD so roughly around 350-400 million transistors per core, not that much of a difference between this and ICL.
Zen 1 has around 220 million transistors per core, again, not a big difference between AMD and Intel here.
Both companies are largely contained by physics after all.

Is Intel's design more efficient then?

 

[OriAr]

No, other than the "double digit gains". Based on leaks they are keeping it extremely close to their chest. Also if it wasn't for the virus hype we'd have seen Xe arch reveal this week.

@OriAr Can't compare transistor count with Zen 2, since that's based on the whole chip. I assume Sunny Cove is quite a bit larger, but not much as some are saying. I also assume for Core chips their 10nm process have about 10-15% area deficiency compared to TSMC 7nm. When they say its competitive in area compared to foundries, its most likely with Tremont cores and other parts of the chip. Core has historically scaled 50% per process generation and no marketing talk has changed that.

Well it's obviously a bit tricky to compare transistors count directly, so what we can do is take what we know and make some napkin math so we can roughly guess. But what's important to see here it's that it's roughly the same, certainly not in the realm of double.

And about TGL's performance, I expect something like 5% better IPC (Due the added cache) and 10% better clocks to combine for about 15% better performance. Should be a good solid iterative upgrade over ICL in performance.

 

[OriAr]

Is Intel's design more efficient then?

I'd say Intel and AMD are about equal here, no company is doing much better than the other one in that area.

 

[dmens]

Transistor count is a gimmick, the only thing that matters is performance with respect to area and power.

 

[IntelUser2000]

Talking about "grandfather Skylake". The recent security flaws all affect Skylake-based chips, not Icelake.

As a side note, I dislike the way they are being too specific with perf/clock gains. 52%! 18%! In the saner days they'd just call it 50% and 15-20%. It's like in the quest to one-up each other in the marketing contest they decided they need to be exact about it.

Transistor count is a gimmick, the only thing that matters is performance with respect to area and power.

It is not a gimmick but with 5 different ways of counting transistors its not the metric you want to focus on anymore. It has become yet another blunt tool for marketers. They don't target the general audience, but more technical people, the AT forum goers, the people working in the IT industry, enthusiasts.

 

[dmens]

It is not a gimmick but with 5 different ways of counting transistors its not the metric you want to focus on anymore. It has become yet another blunt tool for marketers. They don't target the general audience, but more technical people, the AT forum goers, the people working in the IT industry, enthusiasts.

I agree it is a marketing tool but it is absolutely a gimmick, and there is only one way of counting transistors (PMOS + NMOS). A design can have millions of transistors that do very little besides sink current and marketers will spin that as a complex design. I've worked on many projects that were hyped in that manner.

 

[mikk]

Also if it wasn't for the virus hype we'd have seen Xe arch reveal this week.

Xe online reveal should come within the next 1-2 weeks.

https://twitter.com/i/web/status/1240361839105581056

 

[mikk]

RKL-S: https://videocardz.com/newz/exclusive-intel-rocket-lake-s-features-pci-express-4-0-xe-graphics

New core architecture confirmed. Also 12 Bit AV1 support for Xe confirmed which I think is a huge feature. There is nothing about chiplet, so it's most likely a pure monolithic 8+1 port. As expected I have to say, a chiplet design for a tiny GT1 never made sense.

 

[uzzi38]

RKL-S: https://videocardz.com/newz/exclusive-intel-rocket-lake-s-features-pci-express-4-0-xe-graphics

New core architecture confirmed. Also 12 Bit AV1 support for Xe confirmed which I think is a huge feature. There is nothing about chiplet, so it's most likely a pure monolithic 8+1 port. As expected I have to say, a chiplet design for a tiny GT1 never made sense.

Grandpa Skylake being put to rest is nice. AV1 decode is very nice. PCIe Gen 4 lanes of the CPU is nice too, especially with an extra 4 on top, shame it doesn't seem like the chipset can do the same. Seeing as all new GPUs released in 2020 will support Gen 4, I'd certainly hope it had the former, if nothing else.

Gen 11 graphics is cool, I'm interested in what new overclocking features there are, and increased memory support out of the box is pretty cool too. Oh, and thank god SGX is being removed.

Past that, I'm not going to jump to conclusions.

EDIT: I'm an idiot, I meant Gen12. Thanks @Dayman1225

 

[Spartak]

Transistor count is a gimmick, the only thing that matters is performance with respect to area and power.

It's relevant in context to the claim "Sunny Cove (architecture) is ridiculously complex".

If we compare Sunny Cove to Skylake we see 18-19% extra IPC performance for 38% more transistors (217M vs ~300M), so it's a really good performance increase for the extra transistors. Not ridiculous complex.
If we compare Sunny Cove to Zen2 we see about 9% better IPC against 33% less transistors (~300M vs ~400M). Not ridiculous complex either.

Then there's the distinction between complexity and transistor count. More complexity enables less transistors for similar performance, which is opposite to what people here seem to assume. It's a trade-off that I think most if not all CPU manufacturers would prefer, as it generally means less die-space on the same process for similar performance.

 

[yeshua]

If TGL does run at 4.3GHz and doesn't catch a fire it's gonna be faster than Ryzen 4000/Zen 3 CPUs even with a frequency deficit. The only question is whether Intel can produce 8 core parts to successfully compete with the next gen Ryzen CPUs. Intel's future looks bright if/when they can solve their 10/7nm nodes.

 

[lobz]

If TGL does run at 4.3GHz and doesn't catch a fire it's gonna be faster than Ryzen 4000/Zen 3 CPUs even with a frequency deficit. The only question is whether Intel can produce 8 core parts to successfully compete with the next gen Ryzen CPUs. Intel's future looks bright if/when they can solve their 10/7nm nodes.

Yes it will reign king on top of the 4-core laptop market. I know, I know, vast majority of laptops have 4 or less cores and so on, but so did the vast majority of desktops before Zen.

 

[naukkis]

It's relevant in context to the claim "Sunny Cove (architecture) is ridiculously complex".

If we compare Sunny Cove to Skylake we see 18-19% extra IPC performance for 38% more transistors (217M vs ~300M), so it's a really good performance increase for the extra transistors. Not ridiculous complex.
If we compare Sunny Cove to Zen2 we see about 9% better IPC against 33% less transistors (~300M vs ~400M). Not ridiculous complex either.

Then there's the distinction between complexity and transistor count. More complexity enables less transistors for similar performance, which is opposite to what people here seem to assume. It's a trade-off that I think most if not all CPU manufacturers would prefer, as it generally means less die-space on the same process for similar performance.

If we compare core transistor count with L3 cache things won't look so radical. 2MB of L3 takes 175M transistors, so after declining that lefts 40M transistors for Skylake, 50M for Zen2 and 125M for Sunny Cove. L3 cache isn't problem, it packs lots of transistors in small area and won't consume much power.

 

[jpiniero]

I would be surprised if Rocket Lake-S was released this year even though it's now realistic they could do it thanks to the demand destruction from the virus.

 

[dmens]

It's relevant in context to the claim "Sunny Cove (architecture) is ridiculously complex".

If we compare Sunny Cove to Skylake we see 18-19% extra IPC performance for 38% more transistors (217M vs ~300M), so it's a really good performance increase for the extra transistors. Not ridiculous complex.
If we compare Sunny Cove to Zen2 we see about 9% better IPC against 33% less transistors (~300M vs ~400M). Not ridiculous complex either.

Then there's the distinction between complexity and transistor count. More complexity enables less transistors for similar performance, which is opposite to what people here seem to assume. It's a trade-off that I think most if not all CPU manufacturers would prefer, as it generally means less die-space on the same process for similar performance.

I don't even know where you came up with numbers but any blanket claims of relative performance without consideration of workload type and power expended to run that workload is absolutely meaningless. Zen* obviously has a specific target in mind and transistors were spent to achieve scalability.

Did I mention IPC is also a gimmick?

 

[RetroZombie]

If we compare Sunny Cove to Skylake we see 18-19% extra IPC performance for 38% more transistors (217M vs ~300M), so it's a really good performance increase for the extra transistors.

Well actually no.
With that space you can have another skylake core so, by your math 8 core skylake cpus use the same space of 6 sunny cove cpus, so the performance increase is actually negative.

Too much concentration on single core cpus in 2020 leave the market wide open for others to succeed.

 

[RetroZombie]

PCIe Gen 4 lanes of the CPU is nice too, especially with an extra 4 on top, shame it doesn't seem like the chipset can do the same.

Matching a feature of your competitor from 2017 in 2021 is not amazing.
The chipset just PCIe3, is already outdated vs x570.

 

[naukkis]

Well actually no.
With that space you can have another skylake core so, by your math 8 core skylake cpus use the same space of 6 sunny cove cpus, so the performance increase is actually negative.

Don't count transistors. Intel could easily put 2 Skylake cores instead of one Sunny Cove, or 4 Tremont cores which they have done on Lakefield.

 

[Dayman1225]

Matching a feature of your competitor from 2017 in 2021 is not amazing.
The chipset just PCIe3, is already outdated vs x570.

Where did he say it was amazing? He simply said “nice”

 

[lobz]

I would be surprised if Rocket Lake-S was released this year even though it's now realistic they could do it thanks to the demand destruction from the virus.

I thought I alone was thinking that.

 

[uzzi38]

Hey, you guys fans of curveballs on your expectations?

Got a great one for you.

https://twitter.com/i/web/status/1241754120060571648