Intel’s 10th Gen Comet Lake for Desktops: Skylake-S Hits 10 Cores and 5.3 GHzby Dr. Ian Cutress on April 30, 2020 9:00 AM EST
- Posted in
- Comet Lake
So here’s the big question – how does Intel’s hardware stack up against the Zen 2 processors from AMD. For this, we’re going to do some price-to-price comparisons.
At ~$430, the Core i9-10900F goes up against the R9 3900X
|Battle at ~$430|
Ryzen 9 3900X
|10C / 20T||Cores||12C / 24T|
|2.8 GHz||Base Frequency||3.6 GHz|
|65 W||TDP||105 W|
|5.1 GHz||Favored Core (TB3)||4.6 GHz|
|2 x DDR4-2933||DRAM Support||2 x DDR4-3200|
|PCIe 3.0 x16||PCIe Support||PCIe 4.0 x24|
In this instance, Intel has the higher turbo favored core and lower TDP, but AMD has the much higher base frequency, PCIe 4.0 support, and faster memory.
At ~$180, the Core i5-10500 and i5-10400F go up against the popular Ryzen 5 3600:
|Battle at ~$180|
Ryzen 5 3600
|6C / 12T||6C / 12T||Cores||6C / 12T|
|3.1 GHz||2.9 GHz||Base Frequency||3.6 GHz|
|65 W||65 W||TDP||65 W|
|4.5 GHz||4.3 GHz||Favored Core (TB3)||4.2 GHz|
|2x DDR4-2666||2x DDR4-2666||DRAM Support||2x DDR4-3200|
|PCIe 3.0 x16||PCIe 3.0 x16||PCIe Support||PCIe 4.0 x24|
The Core i5-10500 has the higher turbo frequency, but don’t forget this is Zen 2 vs Skylake, and Zen 2 has the higher IPC, so that turbo deficit in frequency might actually still be a win for AMD. The fact that the base frequency is in AMD’s favor considerably, plus the DDR4 support and PCIe support, means that the AMD chip is likely the option here. The i5-10400F is in a similar boat, but at least the deficits it does have come with a price reduction.
How about some halo against halo comparison? The Ryzen 9 3950X and 3900X vs the Core i9-10900KF ?
|Halo vs Halo|
Ryzen 9 3900X
Ryzen 9 3950X
|10C / 20T||Cores||12C / 24T||16C / 32T|
|3.7 GHz||Base Frequency||3.8 GHz||3.5 GHz|
|125 W||TDP||105 W||105 W|
|5.2 GHz||Favored Core (TB3)||4.6 GHz||4.7 GHz|
|4.8 GHz||All-Core Turbo (TB2)||4.0 GHz||3.9 GHz|
|250-350W ?||All-Core Turbo Power||136 W||125 W|
|2x DDR4-2933||DRAM Support||2 x DDR4-3200||2 x DDR4-3200|
|PCIe 3.0 x16||PCIe Support||PCIe 4.0 x24||PCIe 4.0 x24|
Some users will state that the 3900X is the better comparison, only being $40 cheaper, so I’ve included it here as well. Ultimately the thing mainly going for the new hardware is that turbo frequency, up to 5.2 GHz on favored core or 5.3 GHz when under 70ºC. Just looking at the raw CPU data on paper, and some might consider the 10900 series a raw deal.
It should be noted that Intel has different PL2 recommendations for each of the overclockable processors:
- Core i9-10900K: TDP is 125 W, PL2 is 250 W, Tau is 56 seconds
- Core i7-10700K: TDP is 125 W, PL2 is 229 W, Tau is 56 seconds
- Core i5-10600K: TDP is 125 W, PL2 is 182 W, Tau is 56 seconds
Normally the recommended PL2 value is 1.25x the TDP, but in this case Intel is increasing the recommended values. This won’t stop the motherboard manufacturers from completely ignoring them, however.
Also, PL2 and Tau are based on a comparative power load that is defined as a function of a power virus, typically 90-93% or so. This means a complete power virus will go beyond this.
Intel is caught between a rock and a hard place. With its main competitor offering sixteen cores on its mainstream platform and on a better process node, Intel’s struggles with its 10nm process means that the company has to rely on old faithful, 14nm, another time. Unfortunately old faithful is showing its age, especially combined with the fifth generation of Skylake, and all Intel can do is apply new optimizations to get the best out of the chip.
This is to be fair, if I was in Intel’s shoes, what I would probably be doing as well. Rearchitecting production lines to start testing for favored cores isn’t as straightforward as users might think, and then adding in more control logic for Thermal Velocity Boost also means expanding out the firmware and driver support too. Adding in things like DMI/PEG overclocking, per-core HT selection, and VF curves, help with keeping the platform interesting.
In an ideal world, on the desktop Intel would be on its second generation of 10nm hardware by now. We would also be on Ice Lake or a post-Ice Lake microarchitecture, and this would be the suitable entry point for PCIe 4.0 connectivity. As it stands we need to wait, and now we have a new motherboard line with partial PCIe 4.0 support for a product that doesn’t exist yet. Unfortunately this is where I think Intel has made its biggest mistake, in having a new socket/chipset combination straddle the generations between PCIe 3.0 and PCIe 4.0. This is going to create a lot of confusion, especially if some of the new motherboards that are designed to meet ‘PCIe 4.0 specification’ end up not working all that well with the future Rocket Lake product. It’s not a hurdle I would like to come across if I was in the target market for this hardware. I would have, if possible, used the previous socket for another generation and then made the change over for PCIe 4.0 and a new socket with Rocket.
While Intel is announcing the hardware, the exact time it will be on shelves is unknown. Typically with these launches we will have a sense of when review samples will be arriving and when the hardware will go on shelves. At this point I still have open questions with Intel as to when that is – I guess that the online retailers will know when their stock is in place and it will be shown on their websites today.
Post Your CommentPlease log in or sign up to comment.
View All Comments
Lord of the Bored - Monday, May 4, 2020 - linkPentium 4. How's that for broken?
I could name more, but why? You don't care.
Spunjji - Monday, May 4, 2020 - link"UNBROKEN" - except for FDIV, the defective 1.13Ghz Coppermine PIII, the first Wilamette P4 that required costly RDRAM to provide lower performance at a higher price than the Tualatin PIII, the gloriously overpriced Emergency Editions, the Pentium D "dual-core" CPUs that communicated over an ageing FSB for lower performance at higher power than the Athlon X2, the busted Cougar Point chipset Sandy Bridge shipped with, and the total failure to deliver Cannon Lake on time and in quality with even remotely functional hardware (which they're now pretending never happened). Oh and those little Spectre and Meltdown doodads, but those were NBD right? I know I'm missing a few in there too.
AMD have had extremely competitive products for *three years* now. You're an idiot.
Spunjji - Monday, May 4, 2020 - link*in quantity, not in quality; although technically the quality of the Cannon Lake hardware was shockingly poor too. If your partners have to bring in a bottom-run AMD GPU just to make a functional system out of your latest *dual-core no-turbo* product so that you can pretend to shareholders that you are "definitely shipping" to "select partners", you're just about hitting rock bottom.
alufan - Tuesday, May 5, 2020 - linkexplain athlon 64 then please ? or the fact AMD was the first to 1ghz or the fact all modern intel CPUs have massive issues with security because they were designed almost 10 years ago (hence 10th gen name) so yes inlet are consistently bad or poor value AMD just went the wrong way with a chip called Bulldozer but the theory behind it was good as is seen from the modern Ryzen multithreads and the Intel copies released since
Korguz - Tuesday, May 5, 2020 - linkalufan, he cant and wont, and wont post any proof to his BS personal option anti amd claims. when you provide him with any facts and proof, he runs away with his tail between his legs, and either doesnt reply, or resorts to insults, condescending remarks or name calling, the guy cant get his own facts straight, let alone anything else
Santoval - Friday, May 1, 2020 - link"But from now on Intel has no other moves, regarding 14nm and Skylake architecture."
The key word above is "and". Rocket Lake-S will *still* be fabbed at 14nm+++++ but it will (finally) sport a brand new μarch. So it's one out of the two above (an "or"). It is yet unknown how well it will clock though, since this first kind of "backporting" Intel will try. Sunny/Willow Cove are a bit wider designs, so they might not be able to break past 5 GHz. In any case though the power efficiency of Rocket Lake-S is going to be atrocious. Intel cannot beat the laws of physics, and this is why the estimated all-turbo TDPs of their high end parts are more than double than those of AMD.
Those who want both efficiency and performance (i.e. the sane people), and still want to stick with Intel, might find Tiger Lake more appropriate for them. I hear Intel will release Tiger Lake parts up to the -H series and up to 8 cores (earlier I heard Tiger Lake-H would top at 6 cores, but later on the max cores turned to 8). Tiger Lake will also sport the much faster Xe iGPUs, and the -H series should have decent clocks for both the cores and the iGPU (or not).
Those who are not satisfied with anything less than -S series for desktop, and still want to stick with Intel, will need to wait for Alder Lake-S. That ... might take a while. I don't think it will be released before 2H 2021, more likely Q4 2021. Therefore it will be targeted against ... Zen 4. You know, the one that will be fabbed at TSMC's 5nm, will have twice the L2 cache (1 MB), even more L3 cache, DDR5, possibly PCIe 5.0 *and* AVX-512 (Intel's last bastion).
Santoval - Friday, May 1, 2020 - linkp.s. Alder Lake will sport Golden Cove cores and will probably be the last 10nm CPU series of Intel. Its successor is (tentatively) called Meteor Lake and will sport the long awaited Ocean Cove cores Jim Keller and his team have long been working on. This should also be Intel's first 7nm μarch and CPU series, and should also be the first real threat against AMD in terms of both efficiency and performance - unless Intel screw up their 7nm node again. I predict a Q1 2023 (Q4 2022 if all the gods bless Intel with good luck) release, so it should be targeted against Zen 5.
Deicidium369 - Saturday, May 2, 2020 - linkI am expecting Rocket Lake S to be 4-4.5GHz - at least initially.
Tiger Lake will not be a socketed desktop part - NUC11 will be Tiger Lake
There will be ZERO desktop parts with PCIe5. None.
Kevin G - Sunday, May 3, 2020 - linkActually PCIe 5.0 will be wanted for one thing one desktops: single lane controllers for 10 Gbit NICs.
In a few years it does look like NVMe controllers will begin to saturate a PCIe 4.0 M.2 connector so the move to PCI3 5.0 will happen at the right time.
It’ll happen, just a matter of when.
Deicidium369 - Monday, May 4, 2020 - linkNo. PCIe3 and 4 are already capable... Yeah maybe 2027 PCIe5 on desktops.