Intel 10nm/7nm CPU/GPU Rumor Thread : Cannonlake, Icelake, Tigerlake, Sapphire Rapids, etc.

Nightfire said:
A lot to digest there. These are set to launch in 30 days. While the 10 core part may see some limited availability and mark up, I can't see the 4,6, and 8 parts having a problem. My excitement were for those parts anyhow as they are much cheaper than the 8th and 9th gen versions. While early Z490 boards are looking pricey, board manufactures will still compete for the mainstream market.

Both Rocket Lake and Zen 3 may be further away then rumored, so if you want to build something now, Comet Lake is not a bad option.

Not sure why you threw in the 10980xe in a conversation about desktop parts. Not really sure what generation of Pentium you are talking about - I'm guessing it was during the Bulldozer years.
Click to expand...


For the same reason a sudden growth in demand left the Pentium G4560 missing in action for it's first twelve months of retail. All because they added Hyperthreading to a dual-core!

G4560 mostly out of stock 3 months after launch:

https://linustechtips.com/main/topic/762525-pentium-g4560-out-of-stock-everywhere/


G4560 mostly out of stock 6 months after launch:

https://www.pcgamer.com/intels-pentium-g4560-is-not-being-discontinued/

Just because you can keep up with demand now, doesn't mean you can handle the sudden spike in demand when you give away a feature you've locked down for the last six years.

It took almost 12 months for that surge in demand to be sated, And although the Mining Boom is over, I would still expect the surge to last half as long (3-6 months), because they're releasing TWO chips based on the same die with HT now!

the 10980xe was thrown in just to highlight the fact that Intel has a long history of not meeting demand, and if their latest chip is still MIA after 6 months , it's going to be the same story for this 10 core.
 
Last edited:
kirbyrj said:
The 10900k isn't looking like it's going to be an improvement over the existing 3900x (which is ~$400)
Click to expand...

I'm wondering why go this route as well. Just get a 9900K and call it a day.
 
defaultluser said:
For the same reason a sudden growth in demand left the Pentium G4560 missing in action for it's first twelve months of retail. All because they added Hyperthreading to a dual-core!

G4560 mostly out of stock 3 months after launch:

https://linustechtips.com/main/topic/762525-pentium-g4560-out-of-stock-everywhere/


G4560 mostly out of stock 6 months after launch:

https://www.pcgamer.com/intels-pentium-g4560-is-not-being-discontinued/

Just because you can keep up with demand now, doesn't mean you can handle the sudden spike in demand when you give away a feature you've locked down for the last six years.

It took almost 12 months for that surge in demand to be sated, And although the Mining Boom is over, I would still expect the surge to last half as long (3-6 months), because they're releasing TWO chips based on the same die with HT now!

the 10980xe was thrown in just to highlight the fact that Intel has a long history of not meeting demand, and if their latest chip is still MIA after 6 months , it's going to be the same story for this 10 core.
Click to expand...

Nothing you said has any parallel to what is happening now. April 2017 was still 'Bulldozer' years for most consumers as low end Ryzen chips were not on many people's radar. It was still Intel or nothing for budget builds at that point.

A lot has changed in the last 3 years. We have had a 6/12 cpu available for $85 for some time and both companies are launching current gen 4/8 cpus for $100-$120.

After the both the 7980xe and 9980xe were stomped by TR, Intel had no real plans of selling the 10980xe in volume. It was simply a halo product to boost their portfolio.
 
Using Base clocks as a reference, there doesn't seem to be any improvement in silicon quality. The higher 56 second TAU will help with performance. Here is alittle chart on how the cpus compare. I have the PL2 turbo in grey as it really made no sense and most likely erroneous.
skylake.png


The thing to look at is per core W/Ghz. Lower is better, but it is much 'easier' to get a lower number with a lower base clock. The 8086k, 9900k, 9900ks, and 10900k would all be around 4 on this scale if set to 4.0 Ghz. The 10700k seems to be a bit worse and closer to 4.5 at 4.0 ghz, best guess. At the bottom would be the 8700k and 10600k - around 5.
 
Nightfire said:
A early look into what people can expect overclocking wise:
https://www.techpowerup.com/266741/msi-shares-fascinating-insights-into-comet-lake-binning#comments

The 10900k is definately cream of the crop. On average, it uses the same power as the 10700k at a given frequency.

The 10600k, though, is looking rough. The average 6 core cpu needed nearly 1.3v to hit 4.8 ghz vs nearly 5.1 ghz on the 10900k.
Click to expand...
I wouldn't be surprised if the 6 core die was largely unchanged since the 8700k. Maybe a minor stepping change for "hardware" Specre/Meltdown mitigations, but almost everything else remaining the same. Only way to guarantee new silicon is to get a 10core SKU.

Also of curiosity to me, if there are 6 cores harvested from the 10 cores, does the ring still have the extra stops/latency from the 10 core, even though only 6 cores are activated?
 
jeremyshaw said:
Also of curiosity to me, if there are 6 cores harvested from the 10 cores, does the ring still have the extra stops/latency from the 10 core, even though only 6 cores are activated?
Click to expand...

There is near zero chance of that. Intel does much less of this kind of harvesting than AMD, since they make actual parts for each size they sell. Intel has a 10 core die, an 8 core die, a six core die, a 4 core die, and 2 core die.

It's extremely rare to see larger die than expected with disabled cores when delidding an Intel CPU in these core count ranges. IIRC they may have done this for 6 core K parts soldered parts (used 8 core die). It saves doing settup to solder different die sizes.

Maybe they will have 10 core, with two disabled down to 8 in certain situations, but I think 10 down to 6 will be extremely unlikely.
 
Snowdog said:
There is near zero chance of that. Intel does much less of this kind of harvesting than AMD, since they make actual parts for each size they sell. Intel has a 10 core die, an 8 core die, a six core die, a 4 core die, and 2 core die.

It's extremely rare to see larger die than expected with disabled cores when delidding an Intel CPU in these core count ranges. IIRC they may have done this for 6 core K parts soldered parts (used 8 core die). It saves doing settup to solder different die sizes.

Maybe they will have 10 core, with two disabled down to 8 in certain situations, but I think 10 down to 6 will be extremely unlikely.
Click to expand...

Well actually...
https://www.anandtech.com/show/15785/the-intel-comet-lake-review-skylake-we-go-again/4

Still looking at benchmarks, but I haven't seen much of a downside to the added latency.

There are benefits to it. Bigger surface area = better cooling especially with the thinner die on the new 10th gen CPUs.
 
Stock vs stock the 10600k seems to edge out the 8700k despite lower boost clocks but this could be from having a longer tau.

The 10600k overclocking on the reviews matches up to what was expected. 5.0 ghz seems to be the norm at around 1.35v. This is around 8700k expectations. This also means 4.7-4.8 ghz "duds" are a possibility. Still, it could probably afford higher volts with the same cooling given the larger, thinner die.

It would also be neat to see it go against a delided 8700k as not many would delid a soldered cpu.

The 10900k looks good for 5.2 ghz with some getting 5.3 ghz depending on cooling and how far they wanted to push it.

The 10700k will most likely land somewhere in between.
 
Yeah that sounds about right. SL has ETAs for the 10700K and 10900K, respectively, as May 31st and June 7th. Actually a bit surprised they didn't put the top-end part as the priority, but it's only a week's difference anwyay.
 
Nightfire said:
Well actually...
https://www.anandtech.com/show/15785/the-intel-comet-lake-review-skylake-we-go-again/4

Still looking at benchmarks, but I haven't seen much of a downside to the added latency.

There are benefits to it. Bigger surface area = better cooling especially with the thinner die on the new 10th gen CPUs.
Click to expand...

I stand corrected. Sound like an extension of what they did for 9th generation. They only want to run one die through the solder line for the K series.
 
Nightfire said:
Using Base clocks as a reference, there doesn't seem to be any improvement in silicon quality. The higher 56 second TAU will help with performance. Here is alittle chart on how the cpus compare. I have the PL2 turbo in grey as it really made no sense and most likely erroneous.
View attachment 241917

The thing to look at is per core W/Ghz. Lower is better, but it is much 'easier' to get a lower number with a lower base clock. The 8086k, 9900k, 9900ks, and 10900k would all be around 4 on this scale if set to 4.0 Ghz. The 10700k seems to be a bit worse and closer to 4.5 at 4.0 ghz, best guess. At the bottom would be the 8700k and 10600k - around 5.
Click to expand...

My 9900k hasn't run at "base clock" for any great length of time, and certainly runs under 95w when it is.

It is REALLY difficult to quantify CPUs in these terms anymore, the only way you can really do it is using task energy analysis, and you need to make sure that the same settings are applied to all chips, and that the motherboards are as close to each other in specification as is possible.

If you have an embarrassingly parallel workload all the time running exactly the same settings/clock then you can compare - but many workloads don't look like this (even parallel ones) so again, it's really difficult.

Aside from hardware, there are 2 key factors that influence this:
  1. Bios settings - each manufacturer runs different bios settings, Gigabyte for instance disables speedshift and most C-states (when the settings are set to auto) when the clocks are changed from defaults. This behaviour can change from bios version to bios version too!
  2. Windows power plan settings (per my recent thread on QuickCPU), my 9900k idles around 2-5W, typically works around 30-40W when I'm doing something that is parallel but is not embarrasingly so. If you look at other setups they're idling around 30-40W on the cpu.
Why this matters is simple, if 85% of the time the cpu load is at 100%, and the other 15% it is not being used, the processor has the capacity to shift it's multiplier to the lowest state in 15ms or even switch itself off. This will save considerable energy.

These chips are efficient, if you turn the features on that they are designed with
 
Last edited:
Apart from IceLake (mobile), we've still yet to see desktop releases for Sunny Cove - pretty disappointing. Even if the server release (-SP) is executed and on schedule for late 2020, doesn't that mean we're still waiting til 2021 for enthusiast parts?
That lines up more with Sapphire Rapids.

https://en.wikichip.org/wiki/intel/microarchitectures/ice_lake_(client) - 9W to 28W mobile only.
https://en.wikichip.org/wiki/intel/microarchitectures/ice_lake_(server)
https://en.wikichip.org/wiki/intel/microarchitectures/sapphire_rapids
 
Mav451 said:
Apart from IceLake (mobile), we've still yet to see desktop releases for Sunny Cove - pretty disappointing. Even if the server release (-SP) is executed and on schedule for late 2020, doesn't that mean we're still waiting til 2021 for enthusiast parts?
That lines up more with Sapphire Rapids.

https://en.wikichip.org/wiki/intel/microarchitectures/ice_lake_(client) - 9W to 28W mobile only.
https://en.wikichip.org/wiki/intel/microarchitectures/ice_lake_(server)
https://en.wikichip.org/wiki/intel/microarchitectures/sapphire_rapids
Click to expand...

I don't put much stake in wikichips.
Rocket Lake is Willow Cove, which is suppose to be better than Sunny cove.

We will probably never get 10nm on desktop, Tiger Lake included, which is Willow Cove.

While 10nm is competitive with lower freqs of mobile and server, they just don't cut it when pushing to higher freqs.

It's possible that mobile will have dual 11th gen cpus, just like the 10th gen (Ice lake and Comet Lake). Tiger Lake featuring efficiency and great integrated graphics and Rocket Lake for balls out performance.
 
The only annoying thing about Ice Lake on mobile is that they didn't get VRR in the GPU; that's a requirement for me, even with weak Ultrabook graphics, because it just looks so much better with the kind of games you can play well on them -- and those games tend to make tearing even more pronounced!

On the desktop, more cores and more per-core performance would be nice, as would lower power usage to lower the associated noise. Right now AMD has them edged out there for most use cases, so it would be nice to see Intel get their desktop house in order too.
 
IdiotInCharge said:
The only annoying thing about Ice Lake on mobile is that they didn't get VRR in the GPU; that's a requirement for me, even with weak Ultrabook graphics, because it just looks so much better with the kind of games you can play well on them -- and those games tend to make tearing even more pronounced!

On the desktop, more cores and more per-core performance would be nice, as would lower power usage to lower the associated noise. Right now AMD has them edged out there for most use cases, so it would be nice to see Intel get their desktop house in order too.
Click to expand...
Doesn't Ice Lake have VRR on the IGP? I don't know of any Ice Lake laptops with VRR displays, admittedly, whereas there are plenty of AMD laptops with VRR displays (even office laptops, like some generations of Thinkpads).

EDIT: https://www.amd.com/en/products/freesync-laptops
Given the Surface Laptop 3 is listed, it would make very little sense for MSFT to not have it on the Ice Lake versions, though who knows with MSFT? 60-120Hz range sounds strange for a 60Hz display and there is no 13" AMD model, so in retrospect, probably shouldn't put too much faith in the AMD webpage.
 
Last edited:
Snowdog said:
I stand corrected. Sound like an extension of what they did for 9th generation. They only want to run one die through the solder line for the K series.
Click to expand...

Even more complicated than thought:
https://www.techpowerup.com/267757/...circulation-only-one-comes-with-stim#comments

So the i5 lineup has Q0 and G1 stepping. Q0 stepping is based on 10 core dies and has solder with the new thin die. G1 stepping is more like what the 8th gen 6 core parts had.

Thr 10400/f has both Q0 and G1 while the 10500/10600 are G1 only and the 10600k are Q0 only.

So which is better?
On the surface, it looks to be the Q0. However, if using a stock cooler, G1 may be better as it would be easier to delid. It would be interesting to see the latency effects as well.

BTW both ASRock and ASUS now support BFB or 'Ape' on B460 which is pretty cool.
 
jeremyshaw said:
Doesn't Ice Lake have VRR on the IGP? I don't know of any Ice Lake laptops with VRR displays
Click to expand...
I don't think so. I would love to be wrong, but at the same time, I haven't seen an Ice Lake laptop ship with VRR using the IGP yet either. I think it's something along the lines of the necessary IP not flowing to retail parts due to the need to backport features from 7nm and so on.

jeremyshaw said:
whereas there are plenty of AMD laptops with VRR displays (even office laptops, like some generations of Thinkpads
Click to expand...
Yup, and realistically this is where I'd want to go in the near-term, if say Dell could shoehorn them into a modern XPS chassis without making too many compromises. As it stands, I'm not sure that AMD is up to Ice Lake efficiency standards as can be currently implemented. They're close, though.
 
defaultluser said:
For the same reason a sudden growth in demand left the Pentium G4560 missing in action for it's first twelve months of retail. All because they added Hyperthreading to a dual-core!

G4560 mostly out of stock 3 months after launch:

https://linustechtips.com/main/topic/762525-pentium-g4560-out-of-stock-everywhere/


G4560 mostly out of stock 6 months after launch:

https://www.pcgamer.com/intels-pentium-g4560-is-not-being-discontinued/

Just because you can keep up with demand now, doesn't mean you can handle the sudden spike in demand when you give away a feature you've locked down for the last six years.

It took almost 12 months for that surge in demand to be sated, And although the Mining Boom is over, I would still expect the surge to last half as long (3-6 months), because they're releasing TWO chips based on the same die with HT now!

the 10980xe was thrown in just to highlight the fact that Intel has a long history of not meeting demand, and if their latest chip is still MIA after 6 months , it's going to be the same story for this 10 core.
Click to expand...
Yeah, they weren't to hard to find though, I ended up with 2 of them, one for each of my daughters PC's and I didn't pay more than msrp. Still running strong. Yes, manufacturers are in the (bad) habit of releasing things without having much stock. But, on the flip side, it's either that or hold off on letting anyone get it. From their perspective, better to release it and have people wanting it and slowly trickling to customers (and reviewers) than to sit on stock that could be selling. Frustrating though, that's for sure!
 
More random leaks for RKL:
https://www.techpowerup.com/268511/intel-willow-cove-backported-to-14nm-is-cypress-cove

So not exactly Willow Cove as nothing transfer perfectly. Perhaps a lower cache keeps or something else keeps 'Cypress Cove' slightly lower in IPC gains over SKL than Willow cove, though not surprisingly clocks will be higher than Willow.

It's all but confirmed that both RKL and Zen 3 will be early 2021, save perhaps flagship paper launches. Today, AMD mentioned it was due to 'lack of competiton'. Ouch.

Cypress Cove is 'technically' capable of 10 cores but power consumption was already an issue with 8 cores.

The RKL core deficit roughly matches the ipc gains so it will be interesting to see if the flagship RKL can match the flagship CML in MT performance.
 
Snowdog said:
I never believed it was coming this year anyway. It usually takes intel months to get stock up, so they will barely have stock up on 10th gen desktop, and then flip over to 11th?
Click to expand...
I'm sure they have plenty of stock... for Dell and HP priority customers :)
 
Tiger Lake with XE graphics reportedly coming along nicely:
https://twitter.com/ryanshrout/status/1273352056208850944
Ryan Shrout
Perks of the job! Took a prototype Tiger Lake system for a spin on Battlefield V to stretch its legs. Impressive thin and light gaming perf with Xe graphics! Early drivers/sw, but it’s the first time I’ve seen this game run like this on integrated gfx. More later this year!
Click to expand...
 
Snowdog said:
Tiger Lake with XE graphics reportedly coming along nicely:
https://twitter.com/ryanshrout/status/1273352056208850944
Click to expand...
As slow as AMD is taking their APUs... they might miss the boat. AMD needs to be swinging for the fences in the ultrabook space, that's where volume meets margins.

And if Intel can get a competent IGPU running with high-refresh-rate VRR?

I already use graphics with an 8th-gen CPU just fine...
 
Intel has pretty much admitted it won't surpass AMD until 5nm which is late 2022 at the earliest but most likely 2023...geez...how the mighty have fallen
 
polonyc2 said:
Intel has pretty much admitted it won't surpass AMD until 5nm which is late 2022 at the earliest but most likely 2023...geez...how the mighty have fallen
Click to expand...
Surpass in what, core counts?

AMD is still <1/10 the size and is limited by TSMC output.
 
polonyc2 said:
overall performance...right now Intel leads in gaming on paper with benchmark numbers and even then the difference is maybe 6%...
Click to expand...
AMD has more or less reached parity in per-core performance and provides more cores; both made possible by Intel's process stumbles coinciding with TSMCs process successes. However, AMD will remain a small fraction of the market because TSMC simply cannot match Intel for supply, not by a long shot, and AMD is still not TSMC's highest-profile customer.

That means that AMD is going to remain 'behind' for likely five years or so if Intel does nothing, and Intel is doing a lot more than nothing.

What AMD has ensured is their survival in a competitive position, a place they haven't been in since Intel started shipping Core 2. They still have a very large mountain to climb before they can actually lead.
 
IdiotInCharge said:
AMD has more or less reached parity in per-core performance and provides more cores; both made possible by Intel's process stumbles coinciding with TSMCs process successes. However, AMD will remain a small fraction of the market because TSMC simply cannot match Intel for supply, not by a long shot, and AMD is still not TSMC's highest-profile customer.

That means that AMD is going to remain 'behind' for likely five years or so if Intel does nothing, and Intel is doing a lot more than nothing.

What AMD has ensured is their survival in a competitive position, a place they haven't been in since Intel started shipping Core 2. They still have a very large mountain to climb before they can actually lead.
Click to expand...

I don't see even the slightest hint that AMD is supply constrained. There seems to be lots of supply and lots of discounts...

IMO, AMD share has grown slowly because:

1: Until the new 7nm APU, AMD laptop chips just weren't as good as Intels. Laptops are most of the PC market.
2: Server market is conservative, moves slowly.
3: Inertia.
 
Snowdog said:
I don't see even the slightest hint that AMD is supply constrained. There seems to be lots of supply and lots of discounts...
Click to expand...
Not that AMD is currently supply constrained, as in they're running out of product to sell (as you state, we don't see evidence of that), but more that they would be supply constrained if they were to take even 5% more of the CPU market from Intel. There's just a huge gap between the two.
Snowdog said:
2: Server market is conservative, moves slowly.
Click to expand...
In general, yes, but this is becoming less and less the case with cloud stuff.
Snowdog said:
3: Inertia.
Click to expand...
Agreed, but this is more a placeholder for stuff like the cost to redesign platforms, test codebases, or even just adjust the relevant marketing materials and educate purchasing managers. Skylake's everywhere and is a well-known quantity be it in a database server or an ultraportable, or even a network appliance.
 
IdiotInCharge said:
AMD has more or less reached parity in per-core performance and provides more cores; both made possible by Intel's process stumbles coinciding with TSMCs process successes. However, AMD will remain a small fraction of the market because TSMC simply cannot match Intel for supply, not by a long shot, and AMD is still not TSMC's highest-profile customer.

That means that AMD is going to remain 'behind' for likely five years or so if Intel does nothing, and Intel is doing a lot more than nothing.

What AMD has ensured is their survival in a competitive position, a place they haven't been in since Intel started shipping Core 2. They still have a very large mountain to climb before they can actually lead.
Click to expand...

Intel's 'stumbles' are not going away anytime soon...AMD has Zen 3 coming this Fall while Intel has another round of 14nm chips...like I said earlier Intel has a long road back...2023 at the earliest...who knows how far ahead AMD will be by then...when did Ryzen debut?...2017?...so it'll be at least 6 years of AMD dominance
 
polonyc2 said:
Intel's 'stumbles' are not going away anytime soon...AMD has Zen 3 coming this Fall while Intel has another round of 14nm chips...like I said earlier Intel has a long road back...2023 at the earliest...
Click to expand...
Really, this is just a function of Intel getting their production lines running.
polonyc2 said:
who knows how far ahead AMD will be by then...when did Ryzen debut?...2017?
Click to expand...
AMD is limited by TSMC, so AMD can get as far 'ahead' as TSMC does. Or not.
polonyc2 said:
so it'll be at least 6 years of AMD dominance
Click to expand...
AMD only really reached broad parity -- that means faster in some cases, slower than others -- with Zen 2. And they still represent a small fraction of the market, and an even smaller fraction of the install base. They'd need another five years to even start 'dominating'.
 
IdiotInCharge said:
Really, this is just a function of Intel getting their production lines running.

AMD is limited by TSMC, so AMD can get as far 'ahead' as TSMC does. Or not.

AMD only really reached broad parity -- that means faster in some cases, slower than others -- with Zen 2. And they still represent a small fraction of the market, and an even smaller fraction of the install base. They'd need another five years to even start 'dominating'.
Click to expand...

dominating in terms of performance which has nothing to do with TSMC or install size
 
polonyc2 said:
dominating in terms of performance which has nothing to do with TSMC or install size
Click to expand...
Well, they're not dominating in terms of performance, and they're not dominating in terms of sales -- and both of those metrics are completely dependent on TSMC. Benefits of AMD being 'fabless'.
 
Why are people still buying Intel?

People always ask why do some folks still buy Intel if AMD is so good?...Well lets talk about it...

 
I have intel because my ryzen 1700 had some compatibility issues with memory and the software I used at the time. In addition to this, when I upgraded to the 9900k the competing processor was the 2800x which wasn’t as fast.

Ryzen still needs fast memory to work optimally (cause of their infinity fabric) and that is a consideration in price also.

The intel processors, for the average person, aren’t that much worse than the AMD chips, if you don’t look at the 14+++ vs 7nm arguments, or clock for clock comparisons.

Also, you can’t overclock AMD chips worth a damn as they clock themselves, intel chips typically can be overclocked at least 10% on their AVX side, and some on their cache side too. This can make them monsters for single (or few) thread performance.
 
You must log in or register to reply here.
Back
Top