Intel Confirms APO Feature Not Coming to 13th Gen and 12th Gen Core Processors

[InVasMani]

It seems like a lot of people are kind of hopping on the Intel hate train w/o really any of us being briefed fully on how APO operates and why it can or isn't being supported for 12th gen and 13th gen. I suspect their gate keeping pretty hard highhandedly against 13th gen while 12th wouldn't provide the same tangible uplift we see with 14th gen for APO that could probably be nearly identical for 13th gen as well.

I saw that based upon how I suspect APO operates based on post I had made on TPU around E cores shared cache and intelligently utilizing it to reduce latency, but also has a side benefit of dropping power consumption and temperatures as well that translates to power savings and/or higher performance in tandem with boosting algorithms. It could provide higher peak performance for a individual cooling restraints or longer boost duration for higher sustained performance. In either case it's a net positive.

Basically suspect Intel has mimicked disabling of CCX cores on AMD side to provide uplift within the E core cluster. It's a bit like treating each core within a cluster of E cores as a CCX if it operates as I suspect it does and then cleverly dynamically enabling or disabling them based upon the workload. It's not going to to provide performance uplift for the same reasons, but in lighter workloads that are more latency and thermal sensitive it can show positive gains while also increasing efficiency.

 

[lemonadesoda]

The problem with opening this can of worms "processor optimisations for individual games" is that EVERY POSSIBLE PERMUTATION of E- and P- core and cache and memory speeds and memory capacity, as well as OCs will demand a different profile to be optimised. And then depending on GPU the equation will change again! Maybe even summertime and wintertime profiles depending on thermal throttling! Then if the game engine is updated, those optimisations may no longer be as optimal as they could be. The matrix of CPU version x memory x OC x GPU x game engine is unmanageably enormous. The resources needed to maintain and expand that matrix is just not economic. Let AI do it. Let the thread scheduler build its own logbook of performance counters and tweaks and self-optimise. It would require the user to have to accept a short period of suboptimal operation, that would tune-up over time. And changes to memory, OCs, GPU etc. would set it back to square one. The algorithms initial coefficients could be seeded by grabbing optimal profiles of CPU vs GPU vs game engine from a shared internet database. What a monster. But far better than having to download the latest hand-tweaked profiles from Intel that will cover only a very narrow spectrum of PC setups.

 

[thestryker6]

But far better than having to download the latest hand-tweaked profiles from Intel that will cover only a very narrow spectrum of PC setups.

It uses DTT to operate which means it's likely using AI for everything, but the model has to be made still. It's also likely that the benefits will vary depending on game so it's significantly better for Intel to control the variables. This of course also will allow them to dump it easier if the ROI isn't there which is a pretty big downside.

It should have just been pitched as a tech demo until it had broader hardware and software support.

 

[AusWolf]

It seems like a lot of people are kind of hopping on the Intel hate train w/o really any of us being briefed fully on how APO operates and why it can or isn't being supported for 12th gen and 13th gen. I suspect their gate keeping pretty hard highhandedly against 13th gen while 12th wouldn't provide the same tangible uplift we see with 14th gen for APO that could probably be nearly identical for 13th gen as well.

I saw that based upon how I suspect APO operates based on post I had made on TPU around E cores shared cache and intelligently utilizing it to reduce latency, but also has a side benefit of dropping power consumption and temperatures as well that translates to power savings and/or higher performance in tandem with boosting algorithms. It could provide higher peak performance for a individual cooling restraints or longer boost duration for higher sustained performance. In either case it's a net positive.

Basically suspect Intel has mimicked disabling of CCX cores on AMD side to provide uplift within the E core cluster. It's a bit like treating each core within a cluster of E cores as a CCX if it operates as I suspect it does and then cleverly dynamically enabling or disabling them based upon the workload. It's not going to to provide performance uplift for the same reasons, but in lighter workloads that are more latency and thermal sensitive it can show positive gains while also increasing efficiency.

Considering that 14th gen is the same as 13th gen hardware-wise, it doesn't take much speculation to see why APO isn't enabled on older or lower-spec CPUs.

 

[fevgatos]

That's fine, I'll go AMD for the next build

You should, AMD's APO works equally well on all generations across all of their CPUs. Give intel the middle finger.

/s

Fundamentally different but still with P and E cores and games have problems with proper utilisation of them. Also I do not trust Intel with their claims they will have high performance CPU in 20A next year.

It has nothing to do with Ecores, people need to stop this misinformation. Every CPU ecores or no ecores would perform better with proper game specific scheduling. Especially the dual ccd zens

 

[A Computer Guy]

Every CPU ecores or no ecores would perform better with proper game specific scheduling. Especially the dual ccd zens

I feel the same way. If games and applications could simply inform the scheduler of a threads core preference this problem of scheduling could be solved pretty quickly with actively maintained software.

 

[ratirt]

Considering that 14th gen is the same as 13th gen hardware-wise, it doesn't take much speculation to see why APO isn't enabled on older or lower-spec CPUs.

Considering only top processors are benefiting from it, I think that this APO is not for the users per se but for Intel to compete with AMD.

 

[fevgatos]

I feel the same way. If games and applications could simply inform the scheduler of a threads core preference this problem of scheduling could be solved pretty quickly with actively maintained software.

That's currently a problem mainly for high end CPUs with lots of cores. Some games just hog resources even when they don't need them and push power draw to stupid amounts without any performance differences. You can clearly see that on a 14900k or a 7950x, were they draw double the power of their smaller siblings for not much performance gains, or in the case of the 7950x, none at all. I've seen a 7950x hit 140-150w in cyberpunk while performance is almost identical to the 7700x running at 70w. What the heck is that?

That's why the first thing I did on my 14900k was to turn off HT. Dropped power draw by a crapton

 

[AusWolf]

That's why the first thing I did on my 14900k was to turn off HT. Dropped power draw by a crapton

How did it affect performance?

 

[fevgatos]

How did it affect performance?

You mean in games? It did nothing to a slight improvement in some games.

In Mt workloads, went from 41k to 37k in cbr23 but with a huge drop in power and temperatures.

 

[AusWolf]

You mean in games? It did nothing to a slight improvement in some games.

In Mt workloads, went from 41k to 37k in cbr23 but with a huge drop in power and temperatures.

That sounds cool! How much power are you saving this way?

 

[aktpu]

The fact that APO isn’t supported on the 14700k just makes this another non-starter for me.

14700k is supported

 

[BoggledBeagle]

The problem with opening this can of worms "processor optimisations for individual games" is that EVERY POSSIBLE PERMUTATION of E- and P- core and cache and memory speeds and memory capacity, as well as OCs will demand a different profile to be optimised.

Why??? They can optimise for some standard configuration, and if you change your CPU settings, as turning HT off or changing P and E-cores frequencies and RATIO between these frequencies, you may start undoing part of the improvement.

User himself can test if his settings have some negative impact and if this impact is worth it compared to what is he trying to do with these settings (as lowering power draw by changing frequencies, disabling some e-cores, etc.).

 

[fevgatos]

That sounds cool! How much power are you saving this way?

In cbr23? If I told you around 160w you wouldn't believe it right?

CPU hits around 380w on stock and then thermal throttles down to around 320 that my cooler can handle. With ht off and a lower llc (since it doesn't need that much voltage with HT off) power draw stats at a constant 200 to 220 watts.

 

[BoggledBeagle]

In cbr23? If I told you around 160w you wouldn't believe it right?

CPU hits around 380w on stock and then thermal throttles down to around 320 that my cooler can handle. With ht off and a lower llc (since it doesn't need that much voltage with HT off) power draw stats at a constant 200 to 220 watts.

You did something wrong. 8 P cores (without HT) and 16 E cores running at 5600 MHz and 4400 MHz can never consume just 220W in Cinebench R23.

 

[fevgatos]

You did something wrong. 8 P cores (without HT) and 16 E cores running at 5600 MHz and 4400 MHz can never consume just 220W in Cinebench R23.

Okay

 

[BoggledBeagle]

Okay

Are you sure that you were running all cores at that frequencies???

 

[fevgatos]

Are you sure that you were running all cores at that frequencies???

Yes, but where did the 5.6 come from? It's running 5.7

 

[BoggledBeagle]

Speculation based on specs.

 

[fevgatos]

Speculation based on specs.

I've seen some people report that their 14900k boosts to 5.6 but mine does 5.7 straight out of the box. Dunno

 

[noel_fs]

intel never changes

 

[AusWolf]

CPU hits around 380w on stock and then thermal throttles down to around 320 that my cooler can handle. With ht off and a lower llc (since it doesn't need that much voltage with HT off) power draw stats at a constant 200 to 220 watts.

That's only about 20-30% higher than what my 11700 non-K eats with power limits removed while still scoring 2.5x higher in Cb. That's awesome!

 

[unwind-protect]

It seems like a lot of people are kind of hopping on the Intel hate train w/o really any of us being briefed fully on how APO operates and why it can or isn't being supported for 12th gen and 13th gen.

Well, being briefed - we still don't have good information on what exactly Thread Director is doing. And I've been looking.

The only good source of information is the Linux kernel, but it aims for completely different goals. Well, probably. We can't know.

 

[BoggledBeagle]

We are off topic here, but I tried turning HT off on 14700K and the result was rather disappointing, the multithread drop was pretty significant and perfectly proportional to drop in power draw.

 

[Lycanwolfen]

Seems everyone is now making software to try and make their hardware work better. Why not make the hardware better.