ARROW lake overclocking

[Bobaganoosh]

There should be already some experience with overclocking these?

I tried to add 200 MHz to both types of cores on 265K and it seems to work.

BTW I tried to install XTU, which wanted something else installed, and in spite of that something else appearing to be installed correctly, XTU still did not work.

check out SkatterBencher. He's got guides up on overclocking all the different parts of Arrow Lake. I'm not sure about the XTU issue...it seems there were several parts to this new gen that were undercooked when served. It may take time for Intel (and Microsoft) to figure out how to fix it.

Another one that's been interesting is the testing CapFrameX has been doing to see if there's an issue with the spread-out P-core arrangement as well as interconnect OC and setting up prioritization for P-cores in the software:

It seems like RAM and interconnect OC can have a huge improvement, but issues with the scheduler don't seem to be the primary issue (although I've seen some applications where it looks like it is). If only it was easier for them to test APO settings for games (or even have a default state that works 90% of the time) and make Windows run that as default...

It seems clear that disabling e-cores does not help. Prioritizing P-cores in Cyberpunk helps quite a bit though, but short of games that have that option built in, APO support, or some kind of task-lasso app, it's not a one-size-fits-all solution.

From what I can tell watching videos from Jay, Der8aur, and reading some of the SkatterBencher stuff (briefly), it seems like there's not a lot of room for P-core clock increases and it won't have the largest gains there anyway. The best bet is to get temperatures down (because it won't do any overclocking otherwise), try to run both P and e-cores with adaptive voltage and some V/F curve manipulation allowing slightly higher voltages for top clocks and dropping all-core voltages as much as possible, setting up per-core ratio OCing where it boosts depending on thread-usage...then also pushing the interconnect as much as possible and getting the fastest RAM you can. You may not get much higher on P-core boost clocks, but if temps are under control the sustained all-core should be able to go higher. E-cores can go 200-400MHz faster for most chips, which will help mainly in Cinebench, but maybe also in games? The interconnect and RAM OC seem to help the most with latency which is where games will see the most improvement.

Most of the "old-timers" or OCers that have been doing it a long time still like to go for override voltage and all-core OC. SkatterBencher is one of the only ones since at least 11th gen that really shows the benefit of per-core OC and dialing in V/F curves with adaptive voltage. I've found that really helpful on 11th, 12th, and 13th gen (didn't buy 14th...didn't initially plan to buy 12th or 13th, but that's a long story). I think it will come even more into play here, but it's all about what you're trying to achieve.

So after all that, you might see quite an improvement in games...which is a lot of work and expensive parts...and if you're only after gaming improvements AMD will still beat it with X3D (especially with PBO/UV/CO).

Speaking of improvements, what did you see as the outcome of your 200MHz boost to P/e cores?

 

[BoggledBeagle]

I just realised that never saw the P cores run quicker than 5200 MHz, when it has max turbo frequency 5500 MHz. I wonder why it does not want to run quicker.

My "overclock" to 5400 MHz is no overclock at all.

 

[MxPhenom 216]

check out SkatterBencher. He's got guides up on overclocking all the different parts of Arrow Lake. I'm not sure about the XTU issue...it seems there were several parts to this new gen that were undercooked when served. It may take time for Intel (and Microsoft) to figure out how to fix it.

Another one that's been interesting is the testing CapFrameX has been doing to see if there's an issue with the spread-out P-core arrangement as well as interconnect OC and setting up prioritization for P-cores in the software:

It seems like RAM and interconnect OC can have a huge improvement, but issues with the scheduler don't seem to be the primary issue (although I've seen some applications where it looks like it is). If only it was easier for them to test APO settings for games (or even have a default state that works 90% of the time) and make Windows run that as default...

It seems clear that disabling e-cores does not help. Prioritizing P-cores in Cyberpunk helps quite a bit though, but short of games that have that option built in, APO support, or some kind of task-lasso app, it's not a one-size-fits-all solution.

From what I can tell watching videos from Jay, Der8aur, and reading some of the SkatterBencher stuff (briefly), it seems like there's not a lot of room for P-core clock increases and it won't have the largest gains there anyway. The best bet is to get temperatures down (because it won't do any overclocking otherwise), try to run both P and e-cores with adaptive voltage and some V/F curve manipulation allowing slightly higher voltages for top clocks and dropping all-core voltages as much as possible, setting up per-core ratio OCing where it boosts depending on thread-usage...then also pushing the interconnect as much as possible and getting the fastest RAM you can. You may not get much higher on P-core boost clocks, but if temps are under control the sustained all-core should be able to go higher. E-cores can go 200-400MHz faster for most chips, which will help mainly in Cinebench, but maybe also in games? The interconnect and RAM OC seem to help the most with latency which is where games will see the most improvement.

Most of the "old-timers" or OCers that have been doing it a long time still like to go for override voltage and all-core OC. SkatterBencher is one of the only ones since at least 11th gen that really shows the benefit of per-core OC and dialing in V/F curves with adaptive voltage. I've found that really helpful on 11th, 12th, and 13th gen (didn't buy 14th...didn't initially plan to buy 12th or 13th, but that's a long story). I think it will come even more into play here, but it's all about what you're trying to achieve.

So after all that, you might see quite an improvement in games...which is a lot of work and expensive parts...and if you're only after gaming improvements AMD will still beat it with X3D (especially with PBO/UV/CO).

Speaking of improvements, what did you see as the outcome of your 200MHz boost to P/e cores?

Im glad some people are zeroing in on the ring bus frequency when looking at overclocking and performance as that imo seems to be pretty underclocked in these SKUs considering the tile arrangement.

I kind of view these chips as Intels new Sandy Bridge where it sets the stage for them for several future generations based on similar design principals, but with various refinements. I suspect the IMC will get moved to the tile containing the CPU cores in the next generation of CPUs. That alone will probably bring gaming on these cpus back. But if RAM speed and ring bus freq can help for now on Arrow Lake it might be okay, considering non-gaming loads on these chips is pretty sweet.

I think scheduling improvements in Windows (Considering the most recent 24H2 update or whatever its called totally hoses these CPUs to poverty town in performance even further) and any Intel microcode updates that coincide with those updates will help things as well.

 

[Bobaganoosh]

I just realised that never saw the P cores run quicker than 5200 MHz, when it has max turbo frequency 5500 MHz. I wonder why it does not want to run quicker.

My "overclock" to 5400 MHz is no overclock at all.

What are your temps? Did you do any voltage adjustments? What settings did you change to set the increase?

 

[BoggledBeagle]

What are your temps? Did you do any voltage adjustments? What settings did you change to set the increase?

It did not boost, because on Asus board TVB was on Auto by default. It needs to be enabled. Now it boosts to 5.5 GHz, but during heavy one thread load just sometimes.

 

[MxPhenom 216]

It did not boost, because on Asus board TVB was on Auto by default. It needs to be enabled. Now it boosts to 5.5 GHz, but during heavy one thread load just sometimes.

That is usually how these boosts work. The max boosted advertised is almost always on a single core under single core loads. All core boost is usually few 100mhz lower

 

[D007]

Per core voltage is not new. E-cores are still bundled 4 at a time and OC all core individually.

8P overclocking vs older 8P overclocking took an 8 threads kick to the nuts, no more HT. This means any benchmarks 8c vs 8c will get completely clobbered by anything with HT previously released.

So 8 P-core by themselves is useless benchmarking at this point, competitively speaking of course.

These transcripts you copy pasted seem funny to me.

16.6 Megahertz? No, it's a 166mhz divider.

100mhz
133mhz
166mhz
200mhz
So forth.

And DVLr is already in Raptor Lake chips, but said to be disabled? Needs more information research.

So far, nothing here makes Arrow Lake appealing.

They are expensive. Have less threads. Lower clocks, and doesn't game as well as previous gen chips.

I feel there's nothing to Save Intel at this time. They flushed themselves down a toilet already, and Arrow Lake doesn't make up for it.

Also, noting 250w for a 285K means no thermal headroom for overclocking, just like Raptor. So nobody will really be doing any "overclocking" with these with their current cooling.

What makes these even uglier to look at, need a new motherboard again to actually loose performance. Nothing good to note here besides the e-cores are a little faster and you get a few extra. Too bad this doesn't make up the loss of P-core performance. Sad face.

lol agreed. This would be comparable to cutting off your nose to spite your face if you’re a gamer.
It’s a hard pass for me.