Intel to Clock "Rocket Lake-S" High, Evidence of an ES with 5.00 GHz Boost

[btarunr]

Intel's 11th Generation Core "Rocket Lake-S" desktop processors in the LGA1200 package could come with clock speeds that are of the norm these days. Intel appears unwilling to dial down clock speeds in the wake of increased IPC with the new generation "Cypress Cove" CPU cores that drive these processors. Twitter handle "leakbench," which tracks interesting Geekbench results, fished out a database listing for a "Rocket Lake-S" engineering sample with clock speeds of 3.40 GHz base, and 5.00 GHz boost.

The listing has all the telltale signs of "Cypress Cove," such as 48 KB L1D cache, 512 KB per core L2 cache, and 16 MB shared L3 cache for this 8-core/16-thread chip. "Cypress Cove" is rumored to be to be a back-port of Intel's "Willow Cove" CPU core design from its original 10 nm+ node to the 14 nm++. VideoCardz compared this "Rocket Lake-S" ES benchmark result to that of a retail Core i7-10700K, and found its single-threaded performance to be roughly 6.35 percent higher despite a 200 MHz clock-speed deficit, although for some reason, its multi-threaded performance is trailing by over 15 percent.



View at TechPowerUp Main Site

 

[cucker tarlson]

I think they'll have to choose to either limit them to 8c but squeeze out every mhz or dial down the clock and release 10+ core versions

 

[GoldenX]

Stock water coolers incoming.

 

[watzupken]

If the 10% IPC uplift is accurate, then Intel have no choice but to push Rocket Lake to run as fast as it can. To be honest, it does not bode well for Intel if information in these leaks are true. With only a max of 8 cores, and 10% IPC uplift, they are at risk of losing whatever performance crown that they can cling on to. This news is gonna hit Intel's stock price again considering that their outlook in the next 2 to 3 years looks really gloomy.

I think they'll have to choose to either limit them to 8c but squeeze out every mhz or dial down the clock and release 10+ core versions

I wonder if Intel can do a better Rocket Lake if they decided back then to omit the iGPU. I believe Rocket Lake comes with Xe graphics, which I think takes up quite a chunk of the die space.

 

[cucker tarlson]

if it's a good igpu I wont complain
I always buy the igpu versions cause where else are you going to get emergency situation display for just a couple of bucks but if Xe igpu could actually play less demanding titles at 1080p low then it'd be sweet.
I'll be dumping my 2070S soon before rtx3000 and that means no gaming for a few weeks.With a,let's say 1050,level gpu I could still play some stuff

 

[Caring1]

8 Cores is more than enough, but Intel has to stop this fascination with 5GHz clocks and aim for high all core clocks.

 

[Chrispy_]

8 Cores is more than enough, but Intel has to stop this fascination with 5GHz clocks and aim for high all core clocks.

Yeah, the "boost clocks" really aren't something people can achieve in real-world scenarios on either AMD or Intel unless you buy high-end cooling and a high-end motherboard.

If you take either Intel or AMD and put typical $50 air-cooling on a mid-tier motherboard, and then give it a medium load (like gaming or multitasking between a Suite/Browser/Voice) then you'll find that R9 is running around 4.2GHz and i9 is running around 4.6GHz. Don't forget, that's with cooling and a board that already exceeds the 'stock' values, given how motherboard vendors like to squeeze for a competitive advantage.

Sure, they can both hit 500MHz higher on a single core, but when was the last time your OS actually paused all background threads, outside of a synthetic benchmark that artificially grabs all cores and parks all of them except the one running the single-threaded test? If you actually want the advertised boost clocks for either platform across all cores, you are going to need a herculean motherboard, some impressive cooling and plenty of OC experience to dial the voltages down before your socket melts.

 

[Raendor]

Doesn’t look from the numbers like anything much to offer compared to existing lineup. I was hoping for more from a new arch even despite backport.

 

[watzupken]

if it's a good igpu I wont complain
I always buy the igpu versions cause where else are you going to get emergency situation display for just a couple of bucks but if Xe igpu could actually play less demanding titles at 1080p low then it'd be sweet.
I'll be dumping my 2070S soon before rtx3000 and that means no gaming for a few weeks.With a,let's say 1050,level gpu I could still play some stuff

Rumor is that it is a 32 EUs part, so I am not expecting GTX 1050 kind of performance.

 

[tabascosauz]

If the 10% IPC uplift is accurate, then Intel have no choice but to push Rocket Lake to run as fast as it can. To be honest, it does not bode well for Intel if information in these leaks are true. With only a max of 8 cores, and 10% IPC uplift, they are at risk of losing whatever performance crown that they can cling on to. This news is gonna hit Intel's stock price again considering that their outlook in the next 2 to 3 years looks really gloomy.

I wonder if Intel can do a better Rocket Lake if they decided back then to omit the iGPU. I believe Rocket Lake comes with Xe graphics, which I think takes up quite a chunk of the die space.

Not sure where you're getting this 10% number. Willow Cove offers a generational (10-15%) improvement over Sunny Cove, which is well known to have met or exceeded Intel's stated IPC improvement over Coffee Lake (can't remember if 14 or 18%). The only reason Ice Lake didn't show it was because maximum attainable clocks on 10+ were abysmal and negated most of those improvements.

Please don't tell me that "10%" came from Moore's Law is Dead.

That said, Willow Cove is supposed to feature more/updated L3 as its headlining feature. All the leaks around Cypress Cove show that the 8-core ES has exactly the same amount of L3 as existing 8-core SKUs from Coffee 8-core and Comet 10-core dies. Intel did say it was a "reworked cache structure"...

 

[cucker tarlson]

Rumor is that it is a 32 EUs part

no idea what that means

 

[Vya Domus]

but Intel has to stop this fascination with 5GHz clocks

If they do, they lose those last few % that keeps them relevant.

 

[cucker tarlson]

If they do, they lose those last few % that keeps them relevant.

No,on the contrary they are able to keep performance almost identical but lower the power draw significantly.
Locked skus like 10700 are the best all round propositions in the 10 series,but you dont know about that,as was clear in the previous thread that you spammed not even realizing 10700 was an 8/16

 

[Dredi]

if it's a good igpu I wont complain
I always buy the igpu versions cause where else are you going to get emergency situation display for just a couple of bucks but if Xe igpu could actually play less demanding titles at 1080p low then it'd be sweet.
I'll be dumping my 2070S soon before rtx3000 and that means no gaming for a few weeks.With a,let's say 1050,level gpu I could still play some stuff

With just dual channel memory and intels horrid memory controllers (go over 2666mhz and you void your warranty!) I find it extremely unlikely that the performance could be over that of a GT 1030 (that has similar memory performance to intel 10th gen processors). For 1050 level performance you’d need to double your memory speed.

 

[cucker tarlson]

I dont think this warranty stuff matters

 

[Dredi]

I dont think this warranty stuff matters

Maybe price/performance does? Ddr4 memory kits with 6000MHz speeds are likely going to be more expensive than a gtx 1050.

 

[cucker tarlson]

Maybe price/performance does? Ddr4 memory kits with 6000MHz speeds are likely going to be more expensive than a gtx 1050.

the what kits ?

 

[Dredi]

the what kits ?

The kits that could enable gtx 1050 level of performance from an igpu.

 

[cucker tarlson]

yeah right

 

[Dredi]

So you think that intel, a new player in performance gpu space, can make equally performant gpus as nvidia, but with just half of the memory throughput? :’D

 

[cucker tarlson]

So you think that intel, a new player in performance gpu space, can make equally performant gpus as nvidia, but with just half of the memory throughput? :’D

I think if we had apus matching 1030 in 2018 we can expect 1050 4 years later

 

[Dredi]

I think if we had apus matching 1030 in 2018 we can expect 1050 4 years later

With DDR5 it’s easily doable as you can get the necessary throughput, but with DDR4 the max we can expect is 1030 level of performance. This is likely the reason why AMD reduced the number of CUs in renoir. Even if you get more compute power, it does not matter if the memory bottleneck is roughly the same as before.

 

[FreedomEclipse]

So its back to the old clock speed wars for Intel then? Just like in the P4 vs Athlon 64 days.

Intel grasping at straws to try and relive their glory days when they could boast about how much faster clock speeds their CPUs had over AMD offerings.

 

[windwhirl]

Rumor is that it is a 32 EUs part, so I am not expecting GTX 1050 kind of performance.

no idea what that means

EU = Execution Units.

 

[cucker tarlson]

EU = Execution Units.

same as sm or ncu,but still unknown how many shader units per eu and performance

a 32 ncu/sm navi/turing would be very fast. 2060 super/5700 level.