Still waiting for my Batch 2 here and since the first community members seem to get their hands on their 12th gen Frameworks, i was wondering about what can be set in the bios.
Mainly regarding the CPU.
Can you still disable cores? If so, can you specifically disable P-Cores or E-Cores?
Can you disable turbo boost for only P-Cores or only E-Cores?
thanks a lote. Looking forward to play with my FW laptop too!
Have you tried to disable turboboost and see how the system behave ?
Passed 2.4GHZ it s been 10 years that CPU uses a lot of power…
Now that we have so many cores , and BIG ones crunching numbers, this should let use be below 2.4.
On my current 2 cores laptop I have them saturated while working, so I need the max boost out of it.
Other thing I am planing to test is deactivating the E-cores. Because for now with cuurent kernel optimisation the net results is that 12th gen is consumming more power with its E-cores… So if I don t have a server to run and my P cores still have headroom, why would I use E-cores .
I can’t help but wonder if one could sacrifice performance for maximum battery life by simply disabling as many P-cores as possible (maybe even disabling hyper-threading? I wonder if 2 P-core threads on a single P-core would be more efficient than 1 P-thread + 1 E-thread…)
Basically the idea is to force as much as possible to run on the E-cores instead.
Doesn t seam that easy, all the new latope using 12th gen have worste battery life than 11th gen. WHo is the culpride? Maybe the OS kernel …but maybe not.
I am also curious to play with it. How do you make a wall new architecture with “efficient cores” and you end up with less battery life ? Whether it is marketing …or else. I am aswell skeptik and willing to test this.
It’s more cores. Given the same battery capacity, more cores…and just slightly less battery runtime. That is actually more efficient per core, and overall as well. Higher efficiency doesn’t necessarily give you longer run time…because now you have more cores.
A van able to carry 1000KG.
10 kids, each weighing 100KG. = 1000KG in total.
They went to crossfit training for 1 months, loss some fat, gained some muscles. Overall, 80KG each, lean and mean.
Now, same van, able to carry 12 kids (10 + 2 friends). The original kids are fitter now than they were before.
more transistors = more power consumption (assuming a given voltage of course)
And, generally, 4 E-cores takes up the same amount of space on the processor die as a single P-core.
(fun fact: according to AMD, this is a big reason, forgive the pun, why Intel even has to go down the “P/E core” route because Intel’s general non-Atom-based CPU cores are simply too large to adequately increase core counts without making power consumption and the die-size impractical)
But, in non-latency sensitive applications, an E-core has around 50% to 75% the performance per-GHz as a P-core, hence my comment of sacrificing performance. The E-cores also do not clock as high, but that actually helps make the E-cores even more efficient because, at a given voltage, increasing clockrate linearly increases power consumption (e.g. 2GHz @ 1.3v consumes twice as much power as 1GHz @ 1.3v), not to mention that hitting higher clockrates generally requires more voltage which increases power exponentially.
Just a small misconception I’d like to address, E-cores to my understanding are not more efficient, they just operate at a lower power-envolope. More efficient would imply a higher performance-per-watt statistic, but if I remember correctly P-cores still lead in this department.
It seems Intel largely focused on reducing the footprint of E-cores instead of making them as efficient as possible.
Edit: Changed lower for higher performance-per-watt, sorry haha it’s late.
At least on the 12900K (non-S) which has the same amount of P-cores and E-cores, the E-cores were around half as slow:
Yet earlier in the very same review, we can see that fully loading only the E-cores results in 48w while fully-loading only the P-cores results in 239w:
Doing some basic math, this would imply that the E-cores are getting around twice the performance-per-watt.
However, it’s unclear if they were running 8 threads on the P-cores or 16 threads on the P-cores, e.g. if they were taking advantage of hyperthreading or not.
Additional tests on the following page shows that some workloads definitely see substantial performance improvement with hyperthreading, but sadly we do not know what the power cost is of that improved performance:
Furthermore, it being the 12900K rather than the non-k can imply that the P-cores have stock clocks that are way past the sweet-spot on their performance-per-volt efficiency curve (e.g. if you underclocked by just 10% you could very well be able to drop voltage by 30% and, as mentioned, voltage increases power consumption exponentially so you’d see substantial power savings).
If only I had access to a 12th gen Framework laptop then I could totally test this out myself and report back on this whole P vs E-core stuff since I even have a wall power usage meter (the idea is that you disconnect the battery and run only off of AC power), and a P laptop SKU would obviously have the P-cores be configured much closer to the sweet-spot on their performance-per-volt efficiency curve than a K desktop SKU would.
If you look at energy required for X GHz, and energy required to reach X+Y GHz, you’ll usually see that it’s an exponential increase in energy requirement. It’s not linear.
As such, if you look at E-cores, they’re clocked slower…so relatively speaking, from P cores to E cores, the energy efficiency is actually greater with E-cores.
Now, when it comes to performance-per-watt, there’s no indication (that I know of) that E-cores has lower IPC than P-cores. So, energy required to complete a given length of task / program / set of instructions is actually less.
Similar to this: If you had to travel the distance of 1 light-year, it’s actually more energy efficient to travel at 0.5c than it is to travel at 0.9999c… You’ll still get the job done (get to your destination). It’s just slower.
So, with that said, I’m not sure if it’s a misconception at all.
The trade-off is time and / or energy (two sides of the same coin).
This is different for all architectures though, I assume E-cores have a different Voltage-Clock speed graph compared to P-cores, this is because of how a archetecture works at the base, and what largely differentiates P-Cores and E-Cores. Cores cannot run beyond a certain clock speed if that time is shorter then the shortest amount of time a electrical signal can pass through a certain pipeline stage, E-Core has less pipeline stages.
Here is a definetive answer to this! P-Cores are obviously running at a place in the voltage frequency graph that gives them tiny voltage-clock gains (at the upper end you need a exponentially higher voltage to sustain higher clock speeds), but if a P-core were clocked lower (say at non-boost clock), it would be far more efficient then a E-Core.
Haha at this point it’s just me arguing about the intracacies of proccessor design so I’ll stop now!