Intel Criticizes AMD for “Adding Core Count Just Because You Can”

[Grimlakin]

VMware has been down this road. For a time you simply assigned how many mhz or ghz you wanted a vm to have. But multithreaded apps would suck without separate threads to run in parallel on. So now you tell it how many cores you want to give the os at the ghz of your cpus (yes this can be tweaked.). It's a half *** way of testing faster ghz over mti threaded. But in this example. It. Was better to let the guest os manage threads at the virtual hardware level then be assigned resources via the hypervisor rather than the hypervisor try to emulate a 100 ghz zpu across 50 2 ghz threads.

 

[Riccochet]

The thing I take issue with is that he's acting like the end user doesn't have choices. If they don't need a 3900X or 9900K then they can buy a lower part with lower core count. You can buy what you want, which may not always be what you need. An i5 or 3600 will get the job done. An i9 or 3900 is going to last you a good bit longer as everything becomes more threaded, games included.

I know a lot of people that are buying less NAS's and building less "task specific" PC's and servers to opt for a single solution to do everything. Storage is cheap, modern motherboards can handle a lot of storage, the HTPC is becoming obsolete while Plex and Emby with Roku's are taking over. No one needs a server, gaming PC, NAS, HTPC anymore when one powerful PC can do it all.

 

[Brian_B]

... higher single-threaded boost clocks ... As a result, they'll pull slightly more power even if your just using one core.

But again, that's using the performance of the chip and isn't directly related to core count - that's clock speed and more related to the architecture in general, not how many cores are on there. The power only goes up if you use it, you could always underclock a higher binned chip to a lower bin if you were concerned about that particular issue.

The statement that higher core count = higher power is what I'm refuting.

For most desktop users, there is no benefit to having that many cores and simply adding them increases costs, power consumption and generates more heat.

Thermal/Electrical Power relates to the amount of productivity performed (i.e. IPC and frequency) for a given architecture (i.e - Zen2 is better performance per watt per core than Zen1 than Bulldozer) - the number of cores you stack into a package past that --- sure you would be allowed to reach higher thermals, but only if you actively engage those additional cores because you have to follow laws of thermal dynamics. It doesn't mean that a 12C will automatically draw more power than a 6C, unless you actually start to use more than 6C, which at that point - duh.

Either typical desktop users don't see benefit because they can't engage more cores - that's valid. But they would only see more power consumption and heat if they can actually use those cores - at which point, well, they are getting used.


There is also the consideration of Race to Idle - is it more efficient to have a higher peak but execute the task faster on a 12C, or lower peak but longer duration on a 6C -- that I'm not getting into at this point, although the industry trend seems to favor Race to Idle right now

 

[Dan_D]

But again, that's using the performance of the chip and isn't directly related to core count - that's clock speed and more related to the architecture in general, not how many cores are on there. The power only goes up if you use it, you could always underclock a higher binned chip to a lower bin if you were concerned about that particular issue.

The statement that higher core count = higher power is what I'm refuting.



Thermal/Electrical Power relates to the amount of productivity performed (i.e. IPC and frequency) for a given architecture (i.e - Zen2 is better performance per watt per core than Zen1 than Bulldozer) - the number of cores you stack into a package past that --- sure you would be allowed to reach higher thermals, but only if you actively engage those additional cores because you have to follow laws of thermal dynamics. It doesn't mean that a 12C will automatically draw more power than a 6C, unless you actually start to use more than 6C, which at that point - duh.

Either typical desktop users don't see benefit because they can't engage more cores - that's valid. But they would only see more power consumption and heat if they can actually use those cores - at which point, well, they are getting used.


There is also the consideration of Race to Idle - is it more efficient to have a higher peak but execute the task faster on a 12C, or lower peak but longer duration on a 6C -- that I'm not getting into at this point, although the industry trend seems to favor Race to Idle right now

I didn't word my post very well. However, I never said anything about idle cores consuming anything. All they do is drive up costs. What I did say, or more properly meant to say is that in AMD's case, the higher single core clocks on the higher end parts will pull more power than their lower clocked counterparts in single threaded applications. A 3600X for example won't pull what a 3900X does because the latter clocks higher. You are comparing 4.7GHz to 4.4GHz in boost clocks. This is splitting hairs as the 3900X will only do this in bursts but watching Ryzen Master, it hits 1.55v more often than the lower end CPU's do while boosting. I doubt it amounts to much, but it is something.

The thing is, Ryzen requires considerably more power as you near the clock speed wall.

As for multi-core use, your points about race to idle are valid.