Why does a high core processor have a low clock speed?

[BThamS]

yeah, probably you may know what I'm talking now. The AMD THREADRIPPER. It has three varieties, namely 3990x, 3970x and 3960x. Differences in these CPUs are the core count and the clock speed. These CPUs got an inverse proportionality between the core count and the clock speed. But why they got that? Please anyone clear my doubt.

[-rascal-]

6 minutes ago, BThamS said:

yeah, probably you may know what I'm talking now. The AMD THREADRIPPER. It has three varieties, namely 3990x, 3970x and 3960x. Differences in these CPUs are the core count and the clock speed. These CPUs got an inverse proportionality between the core count and the clock speed. But why they got that? Please anyone clear my doubt.

 

Power and heat generation.

 

They are all rated at 280W.

64-cores (3990X) running at 3.8 GHz uses A LOT more power, and generates A LOT more heat than 24-cores (3960X) running at 3.8 GHz.

3990X running at 3.8GHz full-time would probably be 400W+, rather than the 280W rating.

 

Similar situation with the new Intel 10-core i9-10900K.

Running at 3.7 GHz on all 10-cores / 20-threads, it's pretty close to the 125W rating.

Overclock the damn chip, so all 10-cores / 20-threads now run at 5.2GHz+, it's now almost 300W+.

[J-BANANZA]

Probably mainly due to cooling. Less cores, more surface area for each core to cool out. More cores, less surface are to spread the heat. That just means that maintaining the cores cool is harder.

[Kisai]

8 minutes ago, BThamS said:

yeah, probably you may know what I'm talking now. The AMD THREADRIPPER. It has three varieties, namely 3990x, 3970x and 3960x. Differences in these CPUs are the core count and the clock speed. These CPUs got an inverse proportionality between the core count and the clock speed. But why they got that? Please anyone clear my doubt.

That's to hit a TDP limit.

 

You'll notice the high core counts have lower clock speeds because if they all ran at the maximum clock speed of the fastest part with less cores, it would exceed the TDP. 

 

eg 2Ghz 16 cores = 4Ghz 8 cores of TDP capacity if it scaled linearly. But it doesn't, CPU's actually exponentially consumes more power past a point.

 

https://www.anandtech.com/show/14605/the-and-ryzen-3700x-3900x-review-raising-the-bar/19

 

[Jurrunio]

because doubling the power draw and heat output is not acceptable even if you have double the cores. You can blame physics, cores at the middle of the die will always run the hottest. Giving it more cores = the bigger the center "hotspot" is and the more significant it gets. since the aim is to keep hotspot temperature at least close between all of them under the same cooling condition (since TDP rating is the same), power draw must go down which leads to voltage and clock speed to drop.

[trag1c]

DC-DC conversion is another massive issue. It's no small feat designing a circuit (specifically a buck converter) that can step down the 12 volts that the power supply gives you into the 1ish volt and several hundred amps of peak draw that are needed for the CPU. Now you normally can't do that with only 1 buck converter so you split it into phases which drives complexity even higher as each phase is responsible for a portion of the duty cycle. You also have to make sure the current and the voltage ripple are almost non existent as even 10s of milivolts could be the difference between a working and stable CPU or a dead or unstable CPU. The ripples only get worse with higher current loads as well.