Arrow Lake Refresh

[chameleoneel]

Ok looks like the reasonable limits of this 250K so far, are indeed 5.6 Pcore and 4.9 Ecore.

5.0 Ecore works. but, 5.1 did not. And I don't trust 5.0 for daily driving.

I can't get DDR5 8400 working. Even though it worked fine on the 265K.

Asrock has not updated the BIOS since launch of Arrowlake Refresh. So, maybe they can improve the memory compatibility someday...

(I am also going to re-seat the CPU. I have been trying a Thermal Grizzly Kryosheet (a graphene pad). And it seems like the temps are worse than they should be, with my usual Thermal Grizzly Phasesheet (PTM 7950 copy). So I'm going to switch back to that, and re-seat the CPU in the process. Maybe the memory connection will improve).

Until then, I am running DDR5 8000. And have the latency tweaked to 66ns. Which is right about the best I could get with the 265K.

The 250K should have a little better inherent latency, due to a 100mhz faster base clock for the Ring/cache. And one less cluster for the Pcores. So that's one less 'stop' in the Ring.

As it stands, with the current settings, I am very close to the 265K's stock multicore performance------with two less Pcores and same amount of Ecores. That's pretty dang good.
And while my 265K would do 5.5ghz all day on the Pcores-------it hotter/louder than I would like in gaming. And wouldn't hold 5.5 during cinibench. But this 250K holds 5.6 in cinibench.

I haven't tested gaming yet. I wanna reseat the CPU and switch back to Phasesheet, first.

 

[Below Ambient]

I freaking finally setup my 250K and did some preliminary testing.

For reference, my 265K would do 5.5 Pcore and 4.9 Ecore (I actually never tried 5.0 Ecore. It would not do 5.6 Pcore stable). Letting it run fully open for about 240 watts, was a bit too much power and heat, for my 240 AIO set for balanced noise/performance.
I would usually limit it to 180 watts and limit the core current to 200 (limiting the current helps with micro-throttling, and resulted in better multicore scores).

and it would do DDR5 8400 no problem. Did not tolerate ANY overclock to the ring/cache. 200s boost equivalent settings for D2D and NGU was stable. Although about once a week, it might fail a POST. But, never any errors in testing or any crashes in games.

I don't remember the SP rating on the cores. Not terrible. But, also not high 90's.

-------------------------

For this 250K. The SP rating for the P-cores is 70.................wow that is low (lower SP rating is considered worse).

However, its maybe not as bad as we think? This probably won't be a 5.9ghz chip. But so far....


.....5.6Ghz Pcores. 4,9Ghz Ecores. With no voltage adjustment (~1.22v) - (265K would auto adjust itself to 1.4-ish, for 5.5Pcore).
So far, can't get it to POST with DDR5 8400, which is honestly pretty stupid.
It does do 8000


Fully open, it does about 180 watts. And narrowly beats the 265K at 5.5P 4.9E when limited to 180 watts.
No need to limit the core current. Doing so actually hurts This 250k's performance.

Only game test I have done so far, is the FF14 benchmark. Which is notorioiusly difficult for Arrowlake. And.....

.......There is basically no improvement. Despite IBOT optimizations specifically for FF14. and Tomshardware benchmarked FF14 and showed notably improved FF14 performance with IBOT (I'm not sure if they use the FF14 benchmark, or ran an instance in the real game).

Depends on the motherboard...did you try more vsa? Mine wouldn't budge past 8400 until I messaged with vsa...I'm stuck at 8600 ATM vsa is on 1.5 and vdd vdpp are at 1.53 using 8200 team group 48gb

 

[chameleoneel]

Depends on the motherboard...did you try more vsa? Mine wouldn't budge past 8400 until I messaged with vsa...I'm stuck at 8600 ATM vsa is on 1.5 and vdd vdpp are at 1.53 using 8200 team group 48gb

In this same board, my 265K runs this 8400 non-Cudimm RAM no problem. Simply set XMP and go.

This 250K wont POST with XMP. I have to set XMPfor the timings----but manually adjust it to 8000, down from 8400.

After that, it will POST. And then I can tweak the timing further----just as well as I can with the 265K.

This 250K simply wont tolerate 8400.
I also maxed the System Agent to 1.4v and it still wont do it.

 

[chameleoneel]

Also, I got iBOT to improve FF14 benchmark.

You gotta load APO/IBOT, before you load the main menu of the benchmark.

It might have a greater affect on the real game. But for the benchmark, it basically adds about 5-10 fps to any given moment of the benchmark. Thats with a 5080 with a modest overclock.

 

[chameleoneel]

Weird behavior on this 250K

Gaming performance is basically the same at 5.6P and 4.9 E ---- as it is at stock speeds (5.1P with 5.3 boost on two cores. And 4.6E). In a couple of games its a teeensy bit less. But, overall, the numbers are the same.

Monster Hunter Wilds benchmark is exactly the same.

------------------------------

But, if I do any other OC combo. such as only OC the P cores. Or only OC the E cores. or do some lower OC combo, such as 5.5P and 4.7E-------Everything has less gaming performance, than stock.

There must be some hidden multipliers or something, which is affecting some micro latencies, at these in between clock speeds. And somehow the 5.6/4.9 max OC is brute forcing past it, for a VERY small overall gaming improvement.

Its really curious, because my 265K had pretty linear improvements, with core speeds. Elden Ring in particular, enjoyed extra Pcore speed.

 

[chameleoneel]

Gaming performance is great, overall. With a 5080 and this RAM set to 8000 and some tweaked timings for 66ns memory latency.

Expedition 33, MH:Wilds Benchmark, Oblivion Remastered, RE:Requie----are all VERY CLOSE or the same performance, as my 9800X3D.

FF14 Benchmark is still massively behind.

Elden Ring still has a pretty big difference. Although not as massive as FF14. Especially if you are just running around and actually playing. But, tight corridors and staring at walls can show a pretty big swing in the X3D's favor.


Personally, I think people with 270K which have high SP ratings----should be focusing on latency improvements. So, cache/ring speed, D2D, NGU, etc.

My 265K will boot with pretty high cache speed. and would show some nice improvements. But, wasn't actually ever stable with ANY cache speed increase.

This 250K so far won't even boot with the smallest cache increase. I haven't tried increasing D2D and NGU beyond 200S boost multipliers of 32.

 

[chameleoneel]

Ok, the cache/Ring on this 250K seems pretty great. I've got 4.4ghz cache speed, and it seems solid.

So, I've got 4.4 cache/ring. 3.2 NGU, 3.5 D2D. Various RAM tweaks, including 128,000 TREFI (my RAM has BEEFY heatsinks).

Gonna work on core tuning some more. As noted earlier, the performance behavior is weird. And overall, just running stock core speeds seems just as good, in games. And only a little worse in multicore.

One thing which is really great on these CPUs compared to a 9800X3D: shader compilations go a lot faster. MH:Wilds benchmark shader comp was maaaybe a minute. Compared to like 4 or 5 minutes on 9800X3D.

 

[DogsofJune]

I was wondering about the 250k. I almost grabbed one instead of the 270k.

I flogged the 270k this past weekend. It's a nice performer. Figures now, a week later, that the price dropped.

 

[The Lurker]

Have mine sitting in a basement ready to go into a server. Been collecting parts for nearly 2 months. As you expect, the ram was the hardest to find.

 

[Below Ambient]

In this same board, my 265K runs this 8400 non-Cudimm RAM no problem. Simply set XMP and go.

This 250K wont POST with XMP. I have to set XMPfor the timings----but manually adjust it to 8000, down from 8400.

After that, it will POST. And then I can tweak the timing further----just as well as I can with the 265K.

This 250K simply wont tolerate 8400.
I also maxed the System Agent to 1.4v and it still wont do it.

1.45 is pretty much the starting point... you need more than 1.4

 

[Below Ambient]

Ok, the cache/Ring on this 250K seems pretty great. I've got 4.4ghz cache speed, and it seems solid.

So, I've got 4.4 cache/ring. 3.2 NGU, 3.5 D2D. Various RAM tweaks, including 128,000 TREFI (my RAM has BEEFY heatsinks).

Gonna work on core tuning some more. As noted earlier, the performance behavior is weird. And overall, just running stock core speeds seems just as good, in games. And only a little worse in multicore.

One thing which is really great on these CPUs compared to a 9800X3D: shader compilations go a lot faster. MH:Wilds benchmark shader comp was maaaybe a minute. Compared to like 4 or 5 minutes on 9800X3D.

blackbird pc on youtube did a tuned 9950x3d vs tuned 270k, the 270k won with better 1% lows

 

[arestavo]

blackbird pc on youtube did a tuned 9950x3d vs tuned 270k, the 270k won with better 1% lows

I could see that. The 9800X3D is better than a 9950X3D (I own both), and it would be the one I'd pick for a gaming head to head. Or the 9850X3D.

Unless they disabled the second CCD?

 

[LigTasm]

I could see that. The 9800X3D is better than a 9950X3D (I own both), and it would be the one I'd pick for a gaming head to head. Or the 9850X3D.

Unless they disabled the second CCD?

My 270K is far more consistent on frame rate than my 9800X3D. The averages are like 20% less but the gap between the average and 0.1/1% lows are much less varied on the Intel build. Basically it is more comfortable to play on in many games.

 

[arestavo]

My 270K is far more consistent on frame rate than my 9800X3D. The averages are like 20% less but the gap between the average and 0.1/1% lows are much less varied on the Intel build. Basically it is more comfortable to play on in many games.

Good for you!

 

[LigTasm]

Good for you!

Its been noted in many reviews as well. It makes for a good budget option I guess, although the dead socket kind of sucks. Intel really needs to give up on this 1 chip generation per socket thing.

 

[arestavo]

Its been noted in many reviews as well. It makes for a good budget option I guess, although the dead socket kind of sucks. Intel really needs to give up on this 1 chip generation per socket thing.

OK? Thanks?

Take the compliment and run...

 

[LigTasm]

OK? Thanks?

Don't post in a public forum if you want to be a dickhead when people reply? Thanks.

 

[arestavo]

Don't post in a public forum if you want to be a dickhead when people reply? Thanks.

Thaaaaaaanks! Have a blessed day.

 

[chameleoneel]

1.45 is pretty much the starting point... you need more than 1.4

Can you link me some info on that?

From what I have seen, 1.4v is supposed to be the max safe voltage for Arrowlake's System Agent.

And like I said, my 265K doesn't need an SA voltage boost for 8400. And my 265K is no golden sample.

 

[The Lurker]

Its been noted in many reviews as well. It makes for a good budget option I guess, although the dead socket kind of sucks. Intel really needs to give up on this 1 chip generation per socket thing.

This has been the normal since the beginning of time. It was simply a given that you change the board with the cpu. Personally, it hasnt bothered me then, it doesn't bother me now. I guess, if the board has everything you need then it would be nice not to have too change it, but I never really thought of it as an issue.

 

[Below Ambient]

Can you link me some info on that?

From what I have seen, 1.4v is supposed to be the max safe voltage for Arrowlake's System Agent.

And like I said, my 265K doesn't need an SA voltage boost for 8400. And my 265K is no golden sample.

go hang out at overclock.net theres a few huge threads on it in there