Intel Core i9-10900 10-core CPU Pictured

[AleksandarK]

Intel's desktop Comet Lake-S lineup is close to being released and we are getting more leaks about the CPU models contained inside it. Perhaps one of the most interesting points for Comet Lake-S series is that it brings a boost in frequency and boost in core count, with the highest-end Core i9 processors going up to 10 cores. Thanks to Xfastest, a Hong Kong-based media outlet, we have first pictures of what appears to be an engineering sample of the upcoming Core i9-10900 processor.

Being a non-K version, this CPU is not capable of overclocking and has a fixed TDP rating of 65 Watts. Compared to 125 W of the K models like the upcoming Core i9-10900K, this CPU will output almost half the heat, thus requiring a less capable cooling solution. The CPU is installed in LGA1200 socket, which is a new home for Comet Lake-S CPUs and provides backward compatibility for coolers supporting LGA1151. In the sample processor pictured below, we can see a marking on the CPU that implies 2.5 GHz base clock. Previously rumors were suggesting that this CPU version has 2.8 GHz base clock, however, it can be an early engineering sample given that no official imprints are found on the CPU heat spreader.


View at TechPowerUp Main Site

 

[maxitaxi96]

Well to be honest, everytime Intel added 2 Cores to their best Desktop CPU it somehow worked out for them...
Allthough i have my doubts about that now looking at their competitiors line-up

 

[Vayra86]

2.5 Ghz base for a desktop part? Wtf... I remember them coming with 3.4 base. Intel is just moving goal posts for higher boost figures, and they don't seem to know when to stop.

LOL. Soon you're better off sticking a laptop CPU in there instead.

 

[ncrs]

2.5 Ghz base for a desktop part? Wtf... I remember them coming with 3.4 base. Intel is just moving goal posts for higher boost figures, and they don't seem to know when to stop.

LOL. Soon you're better off sticking a laptop CPU in there instead.

They have to do this in order to keep their own definition of what TDP is. Adding more cores requires the base clock to go down since it's basically still the same 14nm process Skylake.

 

[laszlo]

i don't think they're stupid to add 2 cores with overall lower clocks as they'll gain nothing...

 

[ratirt]

2.5 Ghz base for a desktop part? Wtf... I remember them coming with 3.4 base. Intel is just moving goal posts for higher boost figures, and they don't seem to know when to stop.

LOL. Soon you're better off sticking a laptop CPU in there instead.

I'm sure Intel will bump the clocks up a bit.

 

[ncrs]

i don't think they're stupid to add 2 cores with overall lower clocks as they'll gain nothing...

It's just the base clock. The turbo will be higher of course and will break the TDP by a lot.

Adding more cores even if they are overall lower speed when working together will gain them MT performance which is what Zen excels at. Intel's ST "advantage" will remain basically the same since it can turbo one core very high.

 

[Vayra86]

It's just the base clock. The turbo will be higher of course and will break the TDP by a lot.

Adding more cores even if they are overall lower speed when working together will gain them MT performance which is what Zen excels at. Intel's ST "advantage" will remain basically the same since it can turbo one core very high.

That is just the thing though. Is this really the projected use case for these chips? For a laptop chip I can understand a high boost and lower base, because lots of work is bursty. For a high performance desktop MSDT part... no. This is what happens: Intel looks good in non-sustained single core load scenario's, and in any sustained load scenario it will noticeably lose performance much faster because as soon as that high turbo exceeds TDP over 'X' time, the CPU will use base clock.

So this comes down to the question: do you want your many core CPU to run its crunching workloads at 2.5~2.8 Ghz?

To me that seems abysmal, and the only reason is that >5Ghz 'sells'. I think we can count on them doing at least that.

 

[notb]

That is just the thing though. Is this really the projected use case for these chips? For a laptop chip I can understand a high boost and lower base, because lots of work is bursty. For a high performance desktop MSDT part... no. This is what happens: Intel looks good in non-sustained single core load scenario's, and in any sustained load scenario it will noticeably lose performance much faster because as soon as that high turbo exceeds TDP over 'X' time, the CPU will use base clock.

Why would it be different? These CPUs are sold to the same clients (maybe putting gamers aside).
It's the same boost-idle-boost-idle cycle.

Actually it's the other way round (Intel vs AMD in expectations). AMD looks great in Cinebench or batch encoding. People buy them, run a few benchmarks, post results on forums - great. And one day they notice that their office laptop boots quicker, opens websites faster and actually is perfectly fine for everything they need. So why did they buy this huge desktop? And how to use 12 cores?

So this comes down to the question: do you want your many core CPU to run its crunching workloads at 2.5~2.8 Ghz?

LOL on crunching workloads. How many people here actually do some heavy computing on their uber fast PCs? And I mean concious useful activity, not running benchmarks and distributed computing projects.

Also, you would have to manually limit the CPU to force it to run at those 2.8GHz (which will happen in SFF OEM machines). Leave it alone, provide decent airflow - it'll boost all day long if needed.

 

[ncrs]

That is just the thing though. Is this really the projected use case for these chips? For a laptop chip I can understand a high boost and lower base, because lots of work is bursty. For a high performance desktop MSDT part... no. This is what happens: Intel looks good in non-sustained single core load scenario's, and in any sustained load scenario it will noticeably lose performance much faster because as soon as that high turbo exceeds TDP over 'X' time, the CPU will use base clock.

So this comes down to the question: do you want your many core CPU to run its crunching workloads at 2.5~2.8 Ghz?

To me that seems abysmal, and the only reason is that >5Ghz 'sells'. I think we can count on them doing at least that.

Is this a high performance desktop part? This will go into OEM builds like Dell OptiPlex. Most DIY builders will use a K part that is not bound by a 65W TDP.

As I wrote before, I'm pretty sure the low base clock is just the result of pressure to keep the TDP low - it's 65W after all. My guess is that it's the big OEMs requirements.

Also, you would have to manually limit the CPU to force it to run at those 2.8GHz (which will happen in SFF OEM machines). Leave it alone, provide decent airflow - it'll boost all day long if needed.

A bit of a stretch here, Intel will not boost "all day long" if their own recommendations for power limits is followed. Every OEM follows this and only select gamer boards disable it:
From 7th Generation Intel ® Processor Families for H Platforms:

Compared with previous generation products, Intel Turbo Boost Technology 2.0 will increase the ratio of application power towards TDP and also allows to increase power above TDP as high as PL2 for short periods of time.

 

[Ruru]

65W and 10 cores with Skylake 5.0 architecture and 14nm++++++++ means probably that it uses its turbo clocks for a blink of an eye so HWInfo and similar software shows that it had peaked at those turbo clocks..

 

[Vario]

Well i5-8400 had a low base clock and yet had great performance in turbo. However with 10 cores 20 threads, its going to be blazing hot.

Its predecessor, the i9 9900K was bad enough. To keep the i9 9900K under an aircooler I lowered the all core turbo to something like 4.5 and the two core turbo to 4.8, which got the prime95 temps around 75*C and reported maximum TDP under 150W dropping to 100W long TDP limit, from what was 105*C+ and probably 250+W!

 

[GlacierNine]

2.5 Ghz base for a desktop part? Wtf... I remember them coming with 3.4 base. Intel is just moving goal posts for higher boost figures, and they don't seem to know when to stop.

LOL. Soon you're better off sticking a laptop CPU in there instead.

7700K was 4.2GHz base and 6700K was 4GHz base.

8700K started the trend of reducing base clocks, with 3.7GHz. 9900K continued it. This continues it further.

 

[trparky]

How many people here actually do some heavy computing on their uber fast PCs?

But what about the older single-threaded games that require a high clock speed? If it can't sustain the boost clock for any decent amount of time those older games are going to suffer performance penalties.

 

[ncrs]

But what about the older single-threaded games that require a high clock speed? If it can't sustain the boost clock for any decent amount of time those older games are going to suffer performance penalties.

I doubt this will be a problem. Single or even 2-3-4 core turbo speeds should be at least as good as the older generations. It's the all core turbo that will suffer with core count increases.

 

[trparky]

It's the all core turbo that will suffer with core count increases.

Oh most definitely. I come back to my old joke of a question... Are we all going to have to have our very own nuclear power station-style cooling tower to cool these suckers?

 

[cucker tarlson]

ring or mesh ?
if this is a 5ghz ring 10 core,however inefficient in cinemark,it's gonna kick butts and take names in gaming.

 

[GlacierNine]

But what about the older single-threaded games that require a high clock speed? If it can't sustain the boost clock for any decent amount of time those older games are going to suffer performance penalties.

Skylake's 6700K was a 4GHz all core part, 4.2GHz singlecore. It was faster in games than anything that came before it with any number of cores.

Intel's IPC hasn't changed at all since that time, so we can directly compare the clockspeeds.

As long as the new chips aren't throttling below 4GHz on 4 core workloads, or 4.2GHz single core, then they'll still be as fast or faster than the hardware that was top of the line when those games came out. I really don't think anyone needs to worry about their 6700K outperforming their 10900K as a result of lost clockspeeds - there's just no way a 4 core load is going to be so impossible to cool that it'll need to run at 3.9GHz across each core.

 

[trparky]

there's just no way a 4 core load is going to be so impossible to cool that it'll need to run at 3.9GHz across each core.

How do you figure? The 9900K as it stands right now is a bitch to cool with its eight cores, add another two cores and it's going to exacerbate an already bad cooling situation.

 

[GlacierNine]

How do you figure? The 9900K as it stands right now is a bitch to cool with its eight cores, add another two cores and it's going to exacerbate an already bad cooling situation.

Sure, but you said single threaded. Those extra cores don't matter. They're not under load, therefore they produce negligible heat.

At the end of the day, these are still 14nm parts. A single 10900K core can be considered "pretty much" the same as a single 6700K core. They have the same architecture and IPC. At 4GHz, both parts will perform identically.

That means that with one core, you're dealing with "pretty much" the same amount of heat, over the same amount of area, at the same clockspeeds and voltages. Add a core, you double it, add a core, you triple it, add a core, you quadruple it. You've now built a 6700K. Now add 6 more of those cores, you've built a 10900K.

Now granted, a 10900K core is going to do this at lower voltage and with less heat, because of the refinements of the manufacturing process, but that only works in the favour of the later chip.

If you load 4 of those more efficient, later production 10900K cores, you'll get a reasonable amount less than 6700K heat. If you load all 10 cores you get 10900K heat. If you load one single core then you'll get substantially less heat than either of those circumstances, which means cooling a single threaded workload is simply not an issue - if you're only pursuing the same clocks, anyway. Intel always tries to use as much of the available headroom as possible, which is why the single core boost always goes up, from 6700K to 7700K, 8700K, 9900K, and now 10900K. They're not really producing more heat when in single threaded workloads. They're just producing lots more in multi-threaded workloads.

Single core boost will always go up as long as manufacturing keeps improving. The battle is in maintaining high all-core boost clocks as you add more and more cores into the same space.

 

[JackCarver]

How do you figure? The 9900K as it stands right now is a bitch to cool with its eight cores, add another two cores and it's going to exacerbate an already bad cooling situation.

There are users in the forum who run a 9900KS @ 5.2GHz all core Turbo and corsair 115i platinum AiO, never exceeding 75 degrees in Aida64. So sure you would need a very good cooling solution but it's not impossible.

 

[trparky]

They're not under load, therefore they produce negligible heat.

Technically they're always under load, they're not doing nothing; there are always threads being run on it. They might be light loads but they're loads none the less.

 

[phanbuey]

There are users in the forum who run a 9900KS @ 5.2GHz all core Turbo and corsair 115i platinum AiO, never exceeding 75 degrees in Aida64. So sure you would need a very good cooling solution but it's not impossible.

Possible for sure, but I am amazed at how just two more cores makes quite a huge difference in terms of heat.

You and I are running probably one of the best chips in terms of cooling and performance -- my 8700K @5.1 no avx offset 1.39v on a single thick 120mm rad and not break 76 according to AIDA/BIOS the chip pulls about 170W during an avx load. A 9900KS with just 2 more cores is a comparative nightmare to cool and would absolutely bake that rad at those same settings - add two more cores to a 9900KS and it would be some sort of fire code violation.

I think if they came out with a 9700K version of this series - a 10 core with no HT for a reasonable price it would be a winner.

 

[GlacierNine]

Technically they're always under load, they're not doing nothing; there are always threads being run on it. They might be light loads but they're loads none the less.

Now you're just being picky in favour of preserving a practical inaccuracy. If task manager reports 0% utilization, the core is not under load. Being pinged to 1% or so every few seconds to run a background process isn't the same thing as being pegged at 100% by P95

 

[ncrs]

Now you're just being picky in favour of preserving a practical inaccuracy. If task manager reports 0% utilization, the core is not under load. Being pinged to 1% or so every few seconds to run a background process isn't the same thing as being pegged at 100% by P95

Task Manager is far too inaccurate for this. HWiNFO will show you both package and core deep C-state residency. Only at the highest one the core is "off" (not really truly off, but power gated enough for the power usage not to matter).