Intel Rocket Lake 11th gen desktop CPUs confirmed for Q1 2021

[porina]

4 hours ago, NotTheFirstDaniel said:

Can anyone explain why Intel has stuck with "X Lake" for the past 5 years?

It is nothing more than a name system they have come up with. It does not indicate anything about process or architecture. "lakes" are the codenames given to mainstream desktop and mobile CPUs.

 

8 hours ago, ouroesa said:

Ouchies my brain. In future please write this as "This is kind of an official"

Yeah, that wasn't my best writing. I'll edit it in a bit.

 

3 hours ago, Bombastinator said:

Ok.  So there’s Samsung8 (omitting the nm because it’s misleading) TSMC7, intel14, and intel10.  My understanding is TSMC7 is marginally better(?) than Samsung8, and more better than intel14, whereas intel10 is supposed to be more or less on par with TSMC7, and marginally better(?) than Samsung8. Does this sound right?

As a rough generalisation I'd go with that. It is probably way more complicated but I don't know the details any better.

 

3 hours ago, schwellmo92 said:

Based off all of the above it doesn't sound like Rocket Lake would be a very competitive product so there must be something we are missing. Maybe they will have a new core architecture, or maybe 8 cores is not the top end.

Assuming they go ahead as rumoured, the biggest unknown is what they have to change to get the cores built on 14nm when it was originally designed to go on 10nm, Density will be significantly lower and the CPU will take up a bigger area. That could be something to try, take a Tiger Lake and guess how big it might be if built on 14nm. Anyway, the speculation is therefore will Intel have to cut back on some of the features in order to make it fit onto 14nm. An easy way but will lower performance is to reduce cache size for example. They may drop AVX-512 for another generation. We wont know until they say more about it, and I doubt we'll get anything significant before next year.

[schwellmo92]

52 minutes ago, porina said:

Assuming they go ahead as rumoured, the biggest unknown is what they have to change to get the cores built on 14nm when it was originally designed to go on 10nm, Density will be significantly lower and the CPU will take up a bigger area. That could be something to try, take a Tiger Lake and guess how big it might be if built on 14nm. Anyway, the speculation is therefore will Intel have to cut back on some of the features in order to make it fit onto 14nm. An easy way but will lower performance is to reduce cache size for example. They may drop AVX-512 for another generation. We wont know until they say more about it, and I doubt we'll get anything significant before next year.

I thought they were just dropping the iGPU EU's to save space. I think I recall reading somewhere the desktop parts were only going to go up to 32 EU's.

[porina]

3 minutes ago, schwellmo92 said:

I thought they were just dropping the iGPU EU's to save space. I think I recall reading somewhere the desktop parts were only going to go up to 32 EU's.

I haven't seen that, but I don't look everywhere either. That's a good point. If Intel are focusing on "gaming" then the iGPU is a lot less important and making it smaller isn't really going to impact the gaming market. 

[schwellmo92]

8 minutes ago, porina said:

I haven't seen that, but I don't look everywhere either. That's a good point. If Intel are focusing on "gaming" then the iGPU is a lot less important and making it smaller isn't really going to impact the gaming market. 

Found it, 32 EU's https://browser.geekbench.com/v5/compute/1124595

[porina]

14 minutes ago, schwellmo92 said:

Found it, 32 EU's https://browser.geekbench.com/v5/compute/1124595

A question would be, how many EUs are in existing desktop CPUs? I'm finding this a bit hard to dig out, but the link below suggests recent desktop CPUs (coffee lake, comet lake) only go to 24 EUs, so 32 would be an upgrade. Higher EU counts are only found in mobile offerings.

 

https://en.wikipedia.org/wiki/Intel_Graphics_Technology

[yian88]

13 hours ago, aisle9 said:

So releasing 14nm++++++++++++² in Q1, then a brand new architecture on 10nm in Q3 or Q4? It's like the Pointless Lake 7000 series all over again. Why the hell would you be an early adopter for a brand new CPU knowing it'll be a generation behind in 6-9 months?

 

It would make sense if it would be cheap since performance wont be amazing, but this is intel 14nm++++++++++++ and higher price than 10nm chips on the market right now, makes no sense.

[Senzelian]

Intel isn't stupid. They know exactly what they are working towards:

https://www.eetasia.com/taking-a-closer-look-at-intels-process-roadmap/

 

It doesn't really matter how "bad" Rocket Lake is. What matters is to get the new "SuperFin" process ready for production. This is Intel's actual answer to the Zen architecture and its also what will power not only their new CPUs, but also their GPUs.

 

But they're going all-in with that new process. If it doesn't work out for them, then you can really start pressing F.

[AnonymousGuy]

18 hours ago, schwellmo92 said:

At the moment it looks like Rocket Lake will use Willow Cove cores. Willow Cove is essentially a higher clocking Sunny Cove which had a 15-20% IPC increase over Skylake. If Intel can manage the same clocks as their current Comet Lake lineup we could expect 15-20% better performance per core. I'm not confident they will manage to attain 5.3GHz and Rocket Lake is also dropping down to 8 cores at the top end from 10, as far as we currently know at least.

 

Based off all of the above it doesn't sound like Rocket Lake would be a very competitive product so there must be something we are missing. Maybe they will have a new core architecture, or maybe 8 cores is not the top end.

 

I'll be looking forward to the release and finding out more information on Rocket Lake, they've been poking Skylake for a long time now so it will be refreshing to have a new core architecture in desktop.

You're the closest.  I've seen the benchmark(s) for Rocket Lake and it is something like 20% better IPC cause Willow Cove.  No idea why it's dropping down to 8 cores though from 10. Then you get Alderlake with Golden Cove in 3 quarters with it's 8+8.  Which is kinda Broadwell-ing RocketLake giving it a short lifespan.  Also no idea how good the atom cores are going to be but they're the newest latest greatest gracemont cores that I don't think have existed in any previous products.

 

I'm thinking my move is going to be Alderlake and disabling the Atom cores to free up thermal room for the golden cove big cores.

 

I think Intel is making the right moves focusing on IPC instead of AMD just dumping a bunch of cores that are going to sit idle.

[porina]

7 minutes ago, AnonymousGuy said:

I've seen the benchmark(s) for Rocket Lake and it is something like 20% better IPC cause Willow Cove.

Relative to Skylake? That'll put it roughly ahead of Zen 2 but not catching up with Zen 3.

 

7 minutes ago, AnonymousGuy said:

No idea why it's dropping down to 8 cores though from 10.

I'd still betting on die size. The cores were originally designed for 10nm, so if rumours are right building it on 14nm means they're gonna take a load of space. 

 

7 minutes ago, AnonymousGuy said:

Then you get Alderlake with Golden Cove in 3 quarters with it's 8+8.  Which is kinda Broadwell-ing RocketLake giving it a short lifespan.  Also no idea how good the atom cores are going to be but they're the newest latest greatest gracemont cores that I don't think have existed in any previous products.

 

I'm thinking my move is going to be Alderlake and disabling the Atom cores to free up thermal room for the golden cove big cores.

Separate rumour suggests there will be an offering without the small cores. I do think for performance orientated people, it'll probably just be less hassle to only look at the big cores when it comes to looking at performance.

[AnonymousGuy]

34 minutes ago, porina said:

Relative to Skylake? That'll put it roughly ahead of Zen 2 but not catching up with Zen 3.

 

I'd still betting on die size. The cores were originally designed for 10nm, so if rumours are right building it on 14nm means they're gonna take a load of space. 

 

Separate rumour suggests there will be an offering without the small cores. I do think for performance orientated people, it'll probably just be less hassle to only look at the big cores when it comes to looking at performance.

Relative to Comet Lake, Rocket Lake is expected to be 90% the performance with 2 fewer cores at the same total package power.  So based on that about 12.5% IPC boost.

 

Yeah looks like there's supposed to be an 8C+0A Alderlake part.  That'll be the one I'm interested in

 

I got my shit mixed up, Alderlake is supposed to be ~25% better IPC than Comet Lake.

 

(EDIT: and DDR5 support which seems suspiciously correlated to recent articles about Q3 2021 availability..although we're probably at a point where unless you have 64 cores getting thrashed you don't need DDR5-5400)

[porina]

1 minute ago, AnonymousGuy said:

Relative to Comet Lake, Rocket Lake is expected to be 90% the performance with 2 fewer cores at the same total package power.  

Comet Lake is still Skylake as far as peak IPC is concerned. To rephrase it, are you saying that for a multi-thread workload, 8 core Rocket Lake is 90% the performance of 10 core Comet Lake? That is 12.5% per-core improvement for Rocket Lake but we don't know how the clock compares.

[AnonymousGuy]

35 minutes ago, porina said:

Comet Lake is still Skylake as far as peak IPC is concerned. To rephrase it, are you saying that for a multi-thread workload, 8 core Rocket Lake is 90% the performance of 10 core Comet Lake? That is 12.5% per-core improvement for Rocket Lake but we don't know how the clock compares.

Yep, at the same TDP.  So it's a wildcard really of scaling of power vs. voltage vs. clock.  If this thing is running 4.8Ghz and it's consuming 125W then it's not great power efficiency but the performance per clock would be incredible vs. 10900K.  I would guess the clocks are going to be the same as Comet Lake because it's the same fab process, but it's really unknown.

[bcredeur97]

Intel: "Amd released something so we have to rush something out too"

[Fatih19]

2 hours ago, AnonymousGuy said:

I think Intel is making the right moves focusing on IPC instead of AMD just dumping a bunch of cores that are going to sit idle.

Yeah I think this isn't accurate given what AMD announce just some hours ago.

[CTR640]

Bare Lake

[porina]

5 hours ago, bcredeur97 said:

Intel: "Amd released something so we have to rush something out too"

You really think Intel is just like, well, let's do nothing. They have products ongoing too. Based on what they have openly said, that's Rocket Lake Q1 2021, and Alder Lake 2H 2021 for desktop. Semiconductor design doesn't happen overnight. They are long timescale. These products would have started development long ago. Past AMD actions may have influenced it, but what AMD just announced is way too late to influence it.

[Random_Person1234]

19 hours ago, AnonymousGuy said:

No idea why it's dropping down to 8 cores though from 10.

If it's dropping to 8 cores, will they even call it an i9?

[spartaman64]

On 10/7/2020 at 10:48 PM, Bombastinator said:

Ok.  So there’s Samsung8 (omitting the nm because it’s misleading) TSMC7, intel14, and intel10.  My understanding is TSMC7 is marginally better(?) than Samsung8, and more better than intel14, whereas intel10 is supposed to be more or less on par with TSMC7, and marginally better(?) than Samsung8. Does this sound right?

 

On 10/7/2020 at 10:55 PM, KnoT said:

Nope, you simply cant compare intel, TSMC or Samsung node as those Xnm values there are just names and names alone they could calll them "The Best Node" and it would be the same as calling it Xnm node, but if had to guess current 14nm+++++++++++ intel node is comparable with 7nm+ of TSMC but you cant rly tell as they are simply different. 

 

In the end all you should care about is Perf/watt/price interesting example here is Nvidia Turing 12nm TSMC vs AMD RDNA1 7nm TSMC

what matters is transistor density 

ofc its not the whole story but its a contributing factor to the performance of a chip

[cj09beira]

33 minutes ago, spartaman64 said:

 

what matters is transistor density 

ofc its not the whole story but its a contributing factor to the performance of a chip

something also to keep in mind is that we never got the updated specs for intel's 10nm, so i would bet the actual 10nm being produced isn't this dense

[Bombastinator]

So intel 14 along with a bunch of other processes with 14 or 10 in their names are 35-40ish.

 

the next step stems to be in the 100ish range which includes intel10, TSMC7, & samsung8 though there is a lot of variance.  Samsun8 seems to be in the 60’s range which is sort of half way between the steps

[porina]

https://videocardz.com/newz/intel-rocket-lake-s-pcie-4-0-performance-allegedly-tested-on-z490-motherboard

 

Still in the leak territory, above link claims to show Rocket Lake running on Z490 mobo with PCIe 4.0 functioning. I recall there was some fuss at the launch how well mobos could support PCIe 4.0 because the manufacturers had no way to test them. They could only design for the standard and wait.

 

Anyway, what's also interesting is that there are CPU-Z screenshot showing the cache sizes, at 0.5MB/core and 2MB/core respectively for L2 and L3. Why is this interesting? It's the same as Ice Lake, not Tiger Lake which has 1.25MB/core and 3MB/core respectively. At the launch of Tiger Lake, Anandtech stated there were no changes to the microarchitecture between Ice Lake and Tiger Lake. The IPC gains Tiger Lake showed over Ice Lake were due to the cache increases, which are apparently not going to be present on Rocket Lake. It has long been speculated that cache sizes would be an area for "optimisation" if these new cores were to be made on 14nm. 

 

As a performance prediction, this would put the CPU IPC in the ball park of Zen 2, and still some way short of Zen 3. Intel's current best offers an all core turbo up to 4.9 GHz at stock. What is a typical Zen 2 all core turbo while gaming these days? Assuming Zen 3 will be similar, Rocket Lake is unlikely to retake the gaming crown but not be far off if it retains the clock potential of current CPUs.

 

 

Above tweet includes an annotated die shot of Tiger Lake. This shows how much area the L2 and L3 caches take up. If you scale it down, that's proportionately how much area saving would be obtained from reducing their size. A first impression might be, in the whole picture, that's not a very big space saving. Why not keep it at the bigger size and take the higher performance? We don't know what other changes might be made to reduce size, so it is likely only a contributor to size reduction, and not the sole source of such.

[ManIkWeet]

Too bad I'll have bought the AMD processor by then!

[RejZoR]

I've had AMD during its golden era, 1GHz Thunderbird, the first CPU to go past 1GHz. Then I had Athlon XP 2400+ which was during AMD's era of super high efficiency. This 2400+ was 1.8GHz iirc. But could go against 2400MHz Intel easily. Remember overclocking it and motherboard recognized it as 3200+

 

Then I had "long" history of Intel CPU's, Core 2 Duo 4300 and 5200, Core i7 920, Atom N270, Atom Z8300 both in portable devices and now Core i7 5820K in desktop. Laptop has migrated to Ryzen 5 2500U and I'm hoping to build new system with Ryzen 9 5900X late this year or next year. Depending on how Big Navi turns out, I might have all out AMD system. Though RTX 3080 was in my sights until non existent availability screwed that part. Already had it paid, but decided to get a refund and buy when there will be plenty of them. I realized I never had availability issues coz I always bought stuff few months after launch. Except 6700K when it was new and pretty much unavailable. It's why I bought older but better 5820K for same price then.

 

Why is everything Intel so bloody unexciting these days? CPU's are just so boring and Xe, the only thing I'm really excited has no leaks going on that would really excite us. Come on Intel, what's going on there?!

[straight_stewie]

On 10/7/2020 at 5:12 PM, aisle9 said:

So releasing 14nm++++++++++++² in Q1, then a brand new architecture on 10nm in Q3 or Q4? It's like the Pointless Lake 7000 series all over again. Why the hell would you be an early adopter for a brand new CPU knowing it'll be a generation behind in 6-9 months?

It's not necessarily a bad thing.

Some customers collect components for a long time before final assembly (how you build a top shelf machine with very low income). What this means is that, for example, one might have some DDR4 memory which they'd rather not just throw away, but could still benefit from PCIe 4.

There is a market that will benefit from this kind of launch.

As for the "just go for AMD" response I'm likely to get from that, good luck getting full performance out of 2 of these memory kits with an AMD processor.

[Bombastinator]

4 hours ago, RejZoR said:

I've had AMD during its golden era, 1GHz Thunderbird, the first CPU to go past 1GHz. Then I had Athlon XP 2400+ which was during AMD's era of super high efficiency. This 2400+ was 1.8GHz iirc. But could go against 2400MHz Intel easily. Remember overclocking it and motherboard recognized it as 3200+

 

Then I had "long" history of Intel CPU's, Core 2 Duo 4300 and 5200, Core i7 920, Atom N270, Atom Z8300 both in portable devices and now Core i7 5820K in desktop. Laptop has migrated to Ryzen 5 2500U and I'm hoping to build new system with Ryzen 9 5900X late this year or next year. Depending on how Big Navi turns out, I might have all out AMD system. Though RTX 3080 was in my sights until non existent availability screwed that part. Already had it paid, but decided to get a refund and buy when there will be plenty of them. I realized I never had availability issues coz I always bought stuff few months after launch. Except 6700K when it was new and pretty much unavailable. It's why I bought older but better 5820K for same price then.

 

Why is everything Intel so bloody unexciting these days? CPU's are just so boring and Xe, the only thing I'm really excited has no leaks going on that would really excite us. Come on Intel, what's going on there?!

It might be the fact that Xe IS exciting is the reason there are so few leaks.  Rule#1 for watching college program advertisements on TV:  if you see a program advertised on tv it is the program very often you do NOT want to be in.  It’s being advertised because they’re having trouble filling the program.  There could be a really good reason for this.
 

 Not ALL leaks are advertisements, but a lot of them are. The sheer volume of leakage about the 3080 is what made me suspicious in the first place and still powers my suspicion. 
 

The impression I got from Xe is it’s almost but not quite good enough to shoot low end discrete GPUs in the face.   If they can make a discrete Xe and link it to the iGPU in a useful way they CAN shoot low end GPUs in the face. I think that could be really interesting. I think Xe is some of the most interesting non secret tech intel has atm.  Nvidia should be worried about Xe.  AMD should be worried about Xe.  It’s on the cusp of making a seriously big entrance.