Discussion Intel current and future Lakes & Rapids thread

[eek2121]

Even 5GHz with 7% better ipc than Skylake is not that shabby at all. Intel must stop the bleeding anyway they can.

Well, for some people it will be an upgrade. However, if the rumors are correct and there is not a 10 core, 10900K folks have to upgrade to Zen 3.

7% better performance does little good when your core count drops by 20%.

 

[Markfw]

Well, for some people it will be an upgrade. However, if the rumors are correct and there is not a 10 core, 10900K folks have to upgrade to Zen 3.

7% better performance does little good when your core count drops by 20%.

And the competition's IPC goes up by 20% on top of the 7%....

 

[Glo.]

And the competition's IPC goes up by 20% on top of the 7%....

Maybe we will wait for the reviews, before we will claim anything?

 

[inf64]

Zen3 has a massive IPC advantage over Skylake, around 27-28%, so if CypressCove has around 10% higher IPC it needs to clock to enormous 5.5Ghz in order to match stock 5800X with 4.7Ghz Turbo in ST. I don't think that 5.5Ghz is impossible but that chip will be just worse across the board than 5800X, the only saving grace would be much lower price.

 

[Exist50]

Zen3 has a massive IPC advantage over Skylake, around 27-28%, so if CypressCove has around 10% higher IPC it needs to clock to enormous 5.5Ghz in order to match stock 5800X with 4.7Ghz Turbo in ST. I don't think that 5.5Ghz is impossible but that chip will be just worse across the board than 5800X, the only saving grace would be much lower price.

It should be more like 22%, give or take. Zen 2 was a few percent over Skylake, plus 19% for Zen 3. If Rocket Lake is backported Sunny Cove/Willow Cove, it could be around Zen 3 single thread performance if they maintain the clocks. Of course, efficiency will be a joke, but that's a given at this point.

 

[HurleyBird]

Zen3 has a massive IPC advantage over Skylake, around 27-28%, so if CypressCove has around 10% higher IPC it needs to clock to enormous 5.5Ghz in order to match stock 5800X with 4.7Ghz Turbo in ST. I don't think that 5.5Ghz is impossible but that chip will be just worse across the board than 5800X, the only saving grace would be much lower price.

Gaming is likely to be another saving grace. The important thing for Intel isn't to beat AMD convincingly but to win, or at least be in contention for, something. The rate at which mindshare and market perception shifts when your opponent has the "everything crown" is much more significant than when they have the "majority of things crown."

 

[inf64]

It should be more like 22%, give or take. Zen 2 was a few percent over Skylake, plus 19% for Zen 3. If Rocket Lake is backported Sunny Cove/Willow Cove, it could be around Zen 3 single thread performance if they maintain the clocks. Of course, efficiency will be a joke, but that's a given at this point.

That is incorrect. General purpose IPC advantage that Zen2 has over Skylake, as per AT Deep Dive Ryzen 3000 article is ~7%



Zen2 VS SKL : 1.65/1.54~=1.071 or 7%

This is matched perfectly by a thorough computerbase review:

5.3Ghz 10900K ~100
4.6Ghz 3900X ~93

Normalized per Ghz 3900X is ~7% faster per clock (non gaming).



In conclusion, Zen3 has ~1.07x1.19=1.27 or 27% higher IPC than Skylake, that is a massive advantage.

 

[DisEnchantment]

Any ideas what the top LGA1700 SKU is bringing.
So many contacts added. Another DDR channel? PCIe lanes for sure will go up.
Some additional supply lines?
Either it is planned to have some contacts for future use or Intel has plans for a nice increase in PCIe/DDR/etc connectivity. Very intriguing.

 

[jpiniero]

Any ideas what the top LGA1700 SKU is bringing.
So many contacts added. Another DDR channel? PCIe lanes for sure will go up.
Some additional supply lines?
Either it is planned to have some contacts for future use or Intel has plans for a nice increase in PCIe/DDR/etc connectivity. Very intriguing.

It has DDR5 and possibly PCIe 5. But it also may have some portion of PCH functions as well on package.

 

[Exist50]

That is incorrect. General purpose IPC advantage that Zen2 has over Skylake, as per AT Deep Dive Ryzen 3000 article is ~7%
...

Thanks for the data.

It has DDR5 and possibly PCIe 5. But it also may have some portion of PCH functions as well on package.

DDR5 and PCIe 5.0 don't take inherently more pins. And Alder Lake does not integrate the PCH.

I assume some of the growth is power delivery, maybe bumping up the display support to 4, and probably a lot reserved for future uses. If they actually intend to keep this socket for 3 generations, then it makes sense to budget for, e.g. adding USB4 support, more PCIe, etc.

 

[cortexa99]

Intel Alder Lake-S CPU photo leaks

Exclusive: Intel Alder Lake-S CPU pictured - VideoCardz.com

Alder Lake-S smiles for the camera. Intel Alder Lake-S CPU photo leaks Intel’s first desktop processors based on 10nm SuperFin architecture will officially launch in the second half of 2021. Our sources believe that the launch of Alder Lake will be closer to the fourth quarter, rather than the...

videocardz.com

So again a new socket. And more important, only ~6 months gap between ADL and RKL! Would it become a most messy business timeframe that Intel ever made???

 

[ondma]

Intel Alder Lake-S CPU photo leaks

Exclusive: Intel Alder Lake-S CPU pictured - VideoCardz.com

Alder Lake-S smiles for the camera. Intel Alder Lake-S CPU photo leaks Intel’s first desktop processors based on 10nm SuperFin architecture will officially launch in the second half of 2021. Our sources believe that the launch of Alder Lake will be closer to the fourth quarter, rather than the...

videocardz.com

So again a new socket. And more important, only ~6 months gap between ADL and RKL! Would it become a most messy business timeframe that Intel ever made???

Better that than another iteration of Skylake. My concern is that ADL will be delayed.

 

[ksosx86]

Better that than another iteration of Skylake. My concern is that ADL will be delayed.

Delayed slightly, covid & some retooling.

 

[mikk]

1 or 2 months months ago fanlesstech told ADL-S is Q3 2021. Videocardz sounds like it might slip into Q4.

 

[Zucker2k]

That is incorrect. General purpose IPC advantage that Zen2 has over Skylake, as per AT Deep Dive Ryzen 3000 article is ~7%

View attachment 31759

Zen2 VS SKL : 1.65/1.54~=1.071 or 7%

This is matched perfectly by a thorough computerbase review:

5.3Ghz 10900K ~100
4.6Ghz 3900X ~93

Normalized per Ghz 3900X is ~7% faster per clock (non gaming).

View attachment 31760

In conclusion, Zen3 has ~1.07x1.19=1.27 or 27% higher IPC than Skylake, that is a massive advantage.

Normalized per GHz, huh? How does that work out in the real world, though?

 

[HurleyBird]

Normalized per GHz, huh? How does that work out in the real world, though?

Somewhat awkwardly since it seems Skylake scales better as you approach 5 GHz than Zen 2 does.

 

[inf64]

Normalized per GHz, huh? How does that work out in the real world, though?

Zen2 doesn't clock that well past 4.4Ghz on all cores and due to latency gaming performance is somewhat lower than SKL. Zen3 fixes all that

 

[Thala]

Zen2 doesn't clock that well past 4.4Ghz on all cores and due to latency gaming performance is somewhat lower than SKL. Zen3 fixes all that

Has AMD even disclosed in what tests the 19% IPC uplift can be achieved?

 

[inf64]

Has AMD even disclosed in what tests the 19% IPC uplift can be achieved?

They have not, but they stated it's a geo-mean of 25 workloads (which is a lot). Historically they have been very accurate with Zen claims, ie. they claimed ~50% higher IPC than PD/EX, ~3.5% higher IPC for Zen1+ Vs Zen1 and ~15% higher IPC for Zen2 VS Zen1+, all of which was correct and verified by AT. Interestingly CB20 does have slightly lower IPC jump with Zen3, around ~17.8%, so some tests are obviously higher than 19% and some are lower.

 

[coercitiv]

They have not, but they stated it's a geo-mean of 25 workloads (which is a lot).

Geekbench 5 score or it didn't happen. /s

 

[itsmydamnation]

They have not, but they stated it's a geo-mean of 25 workloads (which is a lot). Historically they have been very accurate with Zen claims, ie. they claimed ~50% higher IPC than PD/EX, ~3.5% higher IPC for Zen1+ Vs Zen1 and ~15% higher IPC for Zen2 VS Zen1+, all of which was correct and verified by AT. Interestingly CB20 does have slightly lower IPC jump with Zen3, around ~17.8%, so some tests are obviously higher than 19% and some are lower.

They also claim 24% better perf per watt, given that it's the same process that's pretty awesome, cezanne is gonna be a beast.

 

[jpiniero]

DDR5 and PCIe 5.0 don't take inherently more pins. And Alder Lake does not integrate the PCH.

41% more pins is a lot. Last time they increased it that much was with Nehalem when they moved the memory controller ondie and eventually got integrated graphics on package.

 

[Shivansps]

Maybe LGA1700 is quad channel? It is both a marketing trick and what is needed to win the IGP wars on DDR5.

 

[DrMrLordX]

So again a new socket. And more important, only ~6 months gap between ADL and RKL! Would it become a most messy business timeframe that Intel ever made???

Rocket Lake-S is supposed to be a January 2021 launch. Supposed to be. Remember that Comet Lake-S was originally supposed to be December 2019 and it was May 2020 instead. But. Assuming Rocket Lake-S does show up in January, Alder Lake-S is said to be 2H 2021 which could be as late as December 2021. And I would not be surprised if that is the case. So more like 11 months. Assuming Intel can pull it in that quickly, which is not guaranteed.

Geekbench 5 score or it didn't happen. /s

Don't forget the IPC chart.

 

[dullard]

Maybe we will wait for the reviews, before we will claim anything?

Remember any rumors that show AMD in a good light in an Intel thread are okay here. All other rumors require waiting for reviews.