Rocket Lake-S could be substantially faster for games than the current Comet Lake-S
Click here to post a comment for Rocket Lake-S could be substantially faster for games than the current Comet Lake-S on our message forum
asturur
The amount of cache would go finally on par with ryzen.
D1stRU3T0R
AMD had 5GHz on 32nm, while 32nm was pretty early and bad.
Why Intel wouldn't do 5.5GHz on a REALLLLYYYY matured 14nm?
Undying
Could be is a key word here. Also, 8core i9? Is that the best you can do intel?
Andy Watson
So Intel ticktock has gone from alternative smaller process and architectural changes to instead throwing more cache at it and trying to going for GHz, in memory of the old Pentium days.....
Having said that if it gives the performance then does it matter how it is done, as long as the price is ok.
southamptonfc
Hmm 5.5gz on 14nm. That would be 250W+ TDP then...
Core i9 EE
Richard Nutman
If this is finally Sunny Cove architecture, then it will certainly be faster than their existing Skylake architecture. Faster than Zen3? Possibly..
The extra IPC may affect the clocks they are able to achieve with it however...
asturur
I personally do not care for the Ghz. The important is the architecture, software people will find a way to exploit it to its best.
Matt26LFC
No its not Sunny Cove, this should be based on Willow Cove, the successor to Sunny Cove, so its better than you thought 😉
NightWind
Besides high clocks, what also matters that it's gonna a be a new...sorry, a good and stable 14nm technology.
https://i.pinimg.com/originals/70/56/a5/7056a5f1b3335c148247757f7bd66961.gif
wavetrex
IPC increase is the result of more transistors. More execution units, more parallelization, more cache, wider data paths and so on. The entire thing gets "wider".
When stuck on the same node, more transistors mean more power used at the same clock, and obviously more heat generated.
Increasing transistor count (IPC) AND clocks on the same node while keeping the chip within allowed power envelope is pretty much impossible, and it's not like they haven't already optimized everything to the max.
If this Rocket Lake has indeed (much) higher IPC, we might see a significant reduction in attainable clock speeds as well (potentially back to 4 Ghz or so), as lower clocks also bring more efficiency per transistor.
kviksand81
You can't really compare things the way you are trying to here, as there are much more details to it than just "32nm" and "14nm", let alone that the former was a planar process and the latter is a relatively early commercially available FinFET (3D or Gate All-Around) process technology.
Based on your assessment in your statement, you may wonder why it is easier said than done for the 7nm TSMC FinFET process to push through the 5 GHz barrier and why performance doesn't scale to perfectly linear towards infinity.
Hint: Very few things are perfectly linear in their scaling behaviour from negative infinity to positive infinity.
The above is not said in a provocative way or as a smart-ass comment, just to be clear.
Richard Nutman
Ahh! So Sunny Cove will never make it to desktop due to their problems with 10nm I guess.
Seems Willow Cove is identical to Sunny Cove except for 3 parts. Details here;
https://www.anandtech.com/show/15971/intels-11th-gen-core-tiger-lake-soc-detailed-superfin-willow-cove-and-xelp/3#:~:text=Willow Cove: +10-20% Performance Over Sunny Cove&text=The microarchitecture of a Willow,performance uplift over Sunny Cove.
Though they claim one of the main improvements is "SuperFin Frequency", which "First, where most of the performance uplift comes from, is the process node. Moving to 10SF and the new SuperFin transistor has enabled Willow Cove to scale better with respect to voltage and frequency, allowing for better metrics across the board".
But there was never a high performance Sunny Cove for desktop, so that is pretty much meaningless, especially since this news states Rocket-Lake is still using 14nm.
bobblunderton
Increasing beyond 8 cores / 16 threads is where most game-engines falter, or just don't see any advantage. You definitely don't need more than that with 90~95% of games. Sure, BeamNG.Drive will use more, and mass battle simulators/games will use more, but most game won't.
While I type this from a computer with 16 cores / 32 threads and sound like a hypocrite - entirely yes - but games just don't NEED more than that, even with streaming. So you waste TDP envelope on an extra 2~4 cores when you could just settle with an 8-core and get higher clocks (which will help dx11 games a lot as they bind on single-threaded speed at 1080p and somewhat at 1440p) with the same TDP as the higher-core count chip at lower clockspeed.
There's always room for CPU configurations higher than 8-cores in HEDT platform arena. You have better memory bandwidth to keep things running smoothly on HEDT, too.
I think 8 cores is good enough for the majority of gamers today - most people could live with an R5 3600 and be fine.
My games didn't perform better except BeamNG when I went from 3700x to 3950x, but BeamNG's traffic ran smoother then, but besides zip/render models/graphic work etc, not much is helped with going beyond 6~8 cores.
That money is currently better spent on a GPU.
Luc
Rocket Lake is similar to Tiger Lake, despite the last should have more cache L2 and the first more L3.
The problem is that TL hasn't been deeply reviewed by everyone, and if I must trust someone, Ian Cutress from Anandtech have a good point telling there isn't any improvement on IPC over Sky Lake and that Intel only wins in AVX or AI. There is an improvement in compute related to SL performance too, thanks to higher turbo and power consumption been well balanced thanks to new 10 nm superfinn lithography, but still much slower than twice the cores from AMD.
I expect Intel to show Rocket Lake the same way, faster than AMD in AVX and AI, while been hotter because of 14 nm, and much slower in compute.
If Rocket Lake wins, I think it will be thanks to software been optimiced for Intel AVX, but I don't hope so 🙄
Luc
Actually you don't need more than 8 cores, nor AVX 512, but it could change in the future, it will depend on optimisation, as always.
Richard Nutman
There should be a substantial jump in IPC from Skylake to Sunny/Willow Cove architecture. More execution ports, more instructions retired per clock, and just generally increased execution buffers.
Trying to release all those extra transistors at 14nm though is going to prove tricky to keep clocks as high as they are with CometLake chips (Skylake architecture) (10 series).
Intel's naming of architecture and chip releases is confusing as hell though!
SeriousSkeletor
GHz is the ONLY e-peen enlargement method. Its scientifically proven kappa.
Luc
In Intel words, they improve every time on every front, but always in a black box mode, so it's really difficult for me to belive them until I see it, let's hope there is a real upgrade this time, at least in the security side.
asturur
Considering that i do not get which is the latest tock. Was it skylake and everything after where just meaningless ticks?
Richard Nutman
For desktop yes. The latest 10 series are still on Skylake architecture.
Sunny/Willow Cove only exist in mobile chips.
