Intel Rocket Lake-S CPU Benchmarked: Up to 22% Faster Compared to the Previous Generation

[AleksandarK]

Just a few days ago, Intel has decided to surprise us and give out information about its upcoming Rocket Lake-S platform designed for desktop users. Arriving early next year (Q1) the Rocket Lake-S platform is yet another iteration of the company's 14 nm node. However, this time we are getting some real system changes with a new architecture design. Backporting its Golden Cove core to 14 nm, Intel has named this new core type Cypress Cove. What used to be the heart of Ice Lake CPUs, is now powering the Rocket Lake-S platform. Besides the new core, there are other features of the platform like PCIe 4.0, new Xe graphics, and updated media codecs. You can check that out here.

Today, we have gotten the first benchmarks of the Intel Rocket Lake-S system. In the Userbenchmark bench, an unknown eight-core Rocket Lake CPU has been compared to Intel's 10th generation Comet Lake-S processors. The Rocket Lake engineering sample ran at 4.2 GHz while scoring a single-core score of 179. Compared to the Core i9-10900K that runs at 5.3 GHz, which scored 152 points, the Cypress Cove design is 18% faster. And if the new design is compared to the equivalent 8C/16T Compet Lake CPU like Core i7-10700K clocked at 5.1 GHz and scoring 148 points, the new CPU uarch is up to 22% faster. This represents massive single-threaded performance increases, however, please take the information with a grain of salt, as we wait for the official reviews.


View at TechPowerUp Main Site

 

[londiste]

Skylake and derivatives > Sunny Cove (in Ice Lake) > Willow Cove (in Tiger Lake) > Golden Cove (in Alder Lake)
Rocket Lake's Cypress Cove is between Sunny Cove and Willow Cove, a modified/improved Sunny Cove backported to 14nm.

As I said in some other thread, this result is suspicious. If Rocket Lake in this test is at 5GHz, the result is very respectable. If the measured 4.2GHz is true and 4.2GHz Rocket Lake does 18% over 5.3GHz Comet Lake, that would be ~40% single-thread performance increase. That benchmark does measure boost clock for single-thread test and results for both 10700K and 3800K scale as expected with the measured clock. Maybe it is just unable to read the frequency from ES Rocket Lake?

 

[Crackong]

179 / 4.2 = 42.619
152 / 5.3 = 28.679

equals to 48.6% IPC improvement.

I guess the research days in area 51 finally pays off.

 

[btarunr]

Don't expect RKL under your tree. It won't be here before late-March 2021.

 

[Caring1]

Don't expect RKL under your tree. It won't be here before late-March 2021.

And by then AMD will be kicking their butt again.

 

[Vayra86]

Sorry but even if they do get fast again, that logo is just too ugly a sticker to put anywhere. Hard pass.

Jokes aside... that looks promising, much like the last few years were full of promises

 

[-The_Mask-]

Intel was talking about double digits IPC gains, this is around 50% more. This sounds highly unlikely, otherwise Intel would of course hinted to something like this and not to 10%+.

 

[AleksandarK]

179 / 4.2 = 42.619
152 / 5.3 = 28.679

equals to 48.6% IPC improvement.

I guess the research days in area 51 finally pays off.

That is not a correct way to calculate IPC, yeah it scores higher results at lower clocks, however, we have to wait for benchmarks to see how much IPC increase is there actually.

Plus take into account that this may be a heavily optimized benchmark so not a real-world performance representation.

 

[ZILZAL]

Looks like intel is trying hard urging their users not to turn red after ryzen reveal and wait 6-8 months.

 

[Vya Domus]

179 / 4.2 = 42.619
152 / 5.3 = 28.679

equals to 48.6% IPC improvement.

I guess the research days in area 51 finally pays off.

The clock speeds reported by these programs are often catastrophically wrong. I seriously doubt this chip is boosting only to 4.2 Ghz, even for a engineering sample.

 

[Dredi]

179 / 4.2 = 42.619
152 / 5.3 = 28.679

equals to 48.6% IPC improvement.

I guess the research days in area 51 finally pays off.

Highly likely that the clock speeds are not recorded correctly. Rocket lake has very similar core design to tiger lake and that has only 18% better IPC than skylake (according to intel). That 18% also benefitted from memory speeds going up from 2666 to 4266MHz, so the gains should be lower with rocket lake. The performance gain in the userbenchmark(lol, only intel marketing would use such a bullshit benchmark) result is 17%, and that to me indicates that it is more likely that the clock speeds are more or less the same as before.

 

[ZoneDymo]

compared to competition, competition, renoir, amd's flagship 4800u, and of course we cant ignore the competition, on the right is AMD 4800u the current flagship cpu from the competition and on the left is Intel...Eleventh gen.

 

[Kokotas]

And what cooling solution are they using for that increase? What is the tdp? Unless we see that the rest is quite meaningless.

 

[bug]

The elephant in the room is that even if Intel catches up to AMD in performance, 14nm means they will still lag in power draw.

 

[micropage7]

and it's back to start where many push the clock speed to get better result, actually i expect more. like in lower speed and lower power consumption they can give better than this

 

[londiste]

That is not a correct way to calculate IPC, yeah it scores higher results at lower clocks, however, we have to wait for benchmarks to see how much IPC increase is there actually.
Plus take into account that this may be a heavily optimized benchmark so not a real-world performance representation.

Why is that not a correct way? Performance divided by frequency gives a pretty good idea for single-core performance per frequency (which is what "IPC" is used as shorthand for these days). If you look at other results - for example 10700K across the frequency range or 3800X across frequency range, the results seem to scale and fairly linearly.

Btw, the results used for comparison in the news post are incorrect. 159 points seems to be right for Comet Lake at about 5.1 GHz (both 10700K or 10900K), 5.3GHz gets 165 points.
That Rocket Lake score is 12-13% faster than 5.1GHz Comet Lake (10700K max boost) and 8-9% faster than 5.3GHz Comet Lake (10900K max boost).
Assuming 4.2 GHz Rocket Lake, then frequency-normalized this is still ~38% that sounds way too much.

 

[Hattu]

All these lakes and coves makes me very confused.

 

[seth1911]

Looks like intel is trying hard urging their users not to turn red after ryzen reveal and wait 6-8 months.

AMD Drivers are still crap

My primary OS is Debian:
CPU = sometimes it cant boot, sometimes the OS crash if the SSD is on the Asmedia Chip
GPU = Vulcan dont work right, RX 5700 have a Hardwarebug with Audio via HDMI

There is only one reason to switch from Intel/Nvidia to AMD, u wanna pay a lot of cash to be a Beta tester.

 

[Vya Domus]

Why is that not a correct way? Performance divided by frequency gives a pretty good idea for single-core performance per frequency (which is what "IPC" is used as shorthand for these days).

Because if you run a benchmark that uses for instance AVX, then you will observe practically no increase in IPC compared to any other CPU with the same number of AVX units because better branching, larger instruction windows and other changes which improve IPC do not really affect that kind of code. IPC is highly depended on the kind of code that is tested.

Not to mention that there are all kinds of odd instructions that execute either much faster or much slower going from one architecture to another.

 

[londiste]

Because if you run a benchmark that uses for instance AVX, then you will observe practically no increase in IPC compared to any other CPU with the same number of AVX units because better branching, larger instruction windows and other changes which improve IPC do not really affect that kind of code. IPC is highly depended on the kind of code that is tested.

It is all in context of this single benchmark anyway. And it does not actually seem to be a bad one looking at other CPUs and historical results.

 

[ador250]

I saw a tweet yesterday that this is some AVX-512 benchmark which RocketLake fully supports and that's why in that particular workload it beat CometLake like 40%+. So in ur usual cases it's more likely +-10% IPC uplift.

 

[R0H1T]

179 / 4.2 = 42.619
152 / 5.3 = 28.679

equals to 48.6% IPC improvement.

I guess the research days in area 51 finally pays off.

These are probably single threaded benchmarks, not measuring IPC unless they're specifying constant clock speed. Also we don't know if the clocks were fixed, then again when Intel says double digit IPC I'm thinking closer to 10% than say 20% over SKL.

AMD Drivers are still crap

My primary OS is Debian:

CPU = sometimes it cant boot, sometimes the OS crash if the SSD is on the Asmedia Chip

GPU = Vulcan dont work right, RX 5700 have a Hardwarebug with Audio via HDMI



There is only one reason to switch from Intel/Nvidia to AMD, u wanna pay a lot of cash to be a Beta tester.

Sure, if you say so

Maybe the EU, Lawrence Livermore National Laboratory & any other Supercomputer operator should get your memo

 

[TheLostSwede]

All these lakes and coves makes me very confused.

It's a cove in a lake, obviously.

AMD Drivers are still crap

My primary OS is Debian:
CPU = sometimes it cant boot, sometimes the OS crash if the SSD is on the Asmedia Chip
GPU = Vulcan dont work right, RX 5700 have a Hardwarebug with Audio via HDMI

There is only one reason to switch from Intel/Nvidia to AMD, u wanna pay a lot of cash to be a Beta tester.

This message was brought to you by Intel.

 

[ZoneDymo]

All these lakes and coves makes me very confused.

This, I am actually a fan, hence im on this forum but man I cant keep up with all of this….

 

[RedelZaVedno]

Never underestimate what A LOT of money thrown into the problem and 111.000 pretty smart employees can achieve. It's not gonna win any efficiency rewards, but never the less, it's very impressive what Intel can still squeeze out of 14 nm ++++++ node. Q1 2021 might be more interesting desktop CPU wise than we thought. But this really is the last 14nm iteration Intel can compete with. They'll be F after Zen4 launch if they can't start mass producing 10 nm or 7nm dies.