Testing ARROW lake - power limited

[BoggledBeagle]

I got myself a testing PC with 265K, Strix E board and ordinary RAM (32GB, 6000 MHz, CL36) and tested the CPU with two power limits 100 and 160W, and compared it to 13900KS with 4 E cores disabled, both with HT on and off.





As you can see, the results are pretty consistent, both apps in both power limits mean 24% performance improvement of Arrow lake compared to Raptor lake with HT disabled and 11% with HT enabled.

Most of the power inefficiency of Raptor lake is gone when it is so much power limited - Raptor lake is extremely efficient with 100W power limit. I believe that the result is pretty positive for Arrow lake with 20 threads against Raptor lake with 28 threads.

13900KS could be a little bit better than 14900K, but that is down to silicon lottery, I do not think it has a significant advantage compared to standard 14th gen CPUs.

 

[Onasi]

I mean, even any architectural changes aside, I would kind of hope for this being the case considering, you know, the massive node jump to TSMC 3N.

 

[Macro Device]

the massive node jump to TSMC 3N.

Yes, it's absurd.

NVIDIA, 2014: same node, give or take 40% more power efficiency (GTX 670 VS GTX 960).
Intel, 2024: 3x node shrink, up to 24% more power efficiency.

 

[Tralalak]

Could I ask you to test the CPU-Z 2.11 benchmark? The benchmark (2.09, 2.10, 2.11) can now be ran on a single coreset/cluster.

• Compare P-cores Golden/Raptor Cove vs P-cores Lion Cove
• Compare E-cores Gracemont vs E-cores Skymont

source: https://www.techpowerup.com/forums/threads/share-your-cpuz-benchmarks.216765/post-5286485
source:

Share your CPUZ Benchmarks!

Hi, Can you tested also the CPU-Z 2.09 benchmark on a single coreset/cluster ? Single-Thread / Multi-Thread • All Cores : 901 / 10 793 • P-Cores : ??? / ???? • E-Cores : ??? / ???? Thanks

www.techpowerup.com

Thank you in advance.

 

[N/A]

Intel, 2024: 3x node shrink, up to 24% more power efficiency.

That is if the CPU has the ideal voltage curve. But it might be able to get a bit more performance per watt with some fine tuning.

Also, the input voltage DLVR is 1.5 volts and must be stepped down to 1.2 as a resistor, so there is a 33% power loss in voltage conversion, as explained by Derbauer.

 

[Ayhamb99]

Could I ask you to test the CPU-Z 2.11 benchmark? The benchmark (2.09, 2.10, 2.11) can now be ran on a single coreset/cluster.

• Compare P-cores Golden/Raptor Cove vs P-cores Lion Cove
• Compare E-cores Gracemont vs E-cores Skymont

source: https://www.techpowerup.com/forums/threads/share-your-cpuz-benchmarks.216765/post-5286485
source:

Share your CPUZ Benchmarks!

Hi, Can you tested also the CPU-Z 2.09 benchmark on a single coreset/cluster ? Single-Thread / Multi-Thread • All Cores : 901 / 10 793 • P-Cores : ??? / ???? • E-Cores : ??? / ???? Thanks

www.techpowerup.com

Thank you in advance.

In addition @BoggledBeagle if possible could you try to test the performance of some games with the E-cores disabled. I have a distinct feeling that Arrow Lake's gaming performance suffered during reviews due to scheduling issues. Appreciate the efforts

 

[BoggledBeagle]

I am afraid I am not planning to do any advanced or time consuming testing.

I also have no spare graphic card I can use, and no quick or special RAM kit, so gaming testing is out of question anyway. These CPUs will perform just fine for normal realistic gaming scenario, when people are GPU limited. The "Arrow lake gaming disaster" is hugely overblown IMO.



Even with the most powerful GPU (how many people have such a GPU?), the CPUs are single digit % apart each other when you play in a normal resolution.

 

[Hyderz]

Overclock your e-cores and you shall see some higher fps… it won’t beat 7800x3d but apparently thats where the lost performance is, Microsoft needs to update to prioritize the cores better

 

[Cowboystrekk]

Interesting! I'm a bit surprised that the efficiencygain vs raptor is so small at 100W moving from 10nm to 3nm. Nvidia gained a lot more going from 8nm to 4nm, for instance 4090 being almost 100% more efficient than 3090.

 

[Outback Bronze]

special RAM kit,

Jay has apparently tested the cudimm memory:

 

[BoggledBeagle]

Interesting! I'm a bit surprised that the efficiencygain vs raptor is so small at 100W moving from 10nm to 3nm. Nvidia gained a lot more going from 8nm to 4nm, for instance 4090 being almost 100% more efficient than 3090.

How I wrote, Raptor lake is extremely efficient when you run it at low frequencies when power limited.

I believe that the efficiency comparison of 4090 and 3090 is not very valid, because 4090 could draw much more power, and at its 450W(?) power limit it is already pretty power limited which means it has a big efficiency advantage over 3090 on top of the manufacturing process advantage.

Also, do not graphic cards tend to have much larger gen to gen performance improvements than CPUs?

Jay has apparently tested the cudimm memory:

I just checked and our largest retailer does not have a single stick of this RAM in stock, not even listed.

BTW how extreme is the overclock in the video, 5.1 GHz E cores seems pretty high, and also is not the cache overclock dangerous?

 

[Outback Bronze]

also is not the cache overclock dangerous?

Got no idea at this stage mate. What I'd really like to know if these cudimm's are compatible with Z790 or are they just for Z890. They run under XMP 3.0 and from what I understand Z790 supports XMP 3.0.

Would be interesting what they can do on Raptor Lake. Time should tell I suppose...

Congrats on your purchase! Must be nice to play around with the shiny new Z890

 

[BoggledBeagle]

Must be nice to play around with the new Z890

I like that board a lot, I am honestly a bit pissed off that I will most likely end up returning it and continue using my Raptor lake CPU.

It works fine so far, even with the ordinary PCIE gen 4 SSD and cheap RAM kit.

 

[pressing on]

What I'd really like to know if these cudimm's are compatible with Z790 or are they just for Z890.

Just for Z890 as the motherboard needs to support the clock driver, the memory will be XMP certified. Should be available later next month (November).

 

[bug]

Interesting! I'm a bit surprised that the efficiencygain vs raptor is so small at 100W moving from 10nm to 3nm. Nvidia gained a lot more going from 8nm to 4nm, for instance 4090 being almost 100% more efficient than 3090.

I'm still not sure why people conflate fab process with efficiency. Yes, individual transistors need less power to operate. But chip design is what actually dictates efficiency.
I mean, look back at Athlon XP vs P4. P4 was actually built on a more advanced node.

 

[Cowboystrekk]

I'm still not sure why people conflate fab process with efficiency. Yes, individual transistors need less power to operate. But chip design is what actually dictates efficiency.
I mean, look back at Athlon XP vs P4. P4 was actually built on a more advanced node.

I think of the potential of savings and here AMD has done a superb job vs Intel even though they use an inferior node.

 

[bug]

I think of the potential of savings and here AMD has done a superb job vs Intel even though they use an inferior node.

You're also still missing the point (the fab node is a small factor in the efficiency equation), but that's your right I guess.

 

[R0H1T]

the fab node is a small factor in the efficiency equation

Not really when you consider that Intel's been leading AMD on the "fab" front for a good 10-15 years till zen4(?) & now back in the lead again. This needs to be mentioned because while Dozer was a major mishap, its issues were compounded by the fact that it ran "hot" & used too much power ~ imagine the same chip using what 50% less power?

Intel's lead on the x86 front was often way too exaggerated & at times directly as a result of their superior nodes. You can say the same about Apple right now but let's forget the fruit seller atm

 

[TheinsanegamerN]

You're also still missing the point (the fab node is a small factor in the efficiency equation), but that's your right I guess.

You could easily argue the opposite too. go look at the efficiency improvements moving from samsung 8nm to TSMC 4nm for nvidia. The gap was MASSIVE. I mean, the argument you gave was P4 vs athlon XP. Those chips are old enough to drink now.

With Arrow lake, moving to tsmc 3nm, which as you seem to think shouldnt matter, intels power consumption was cut in HALF. That's a pretty big deal, and no, its not just core design. If it was, then intel would be using their own node instead of paying TSMC.

Not really when you consider that Intel's been leading AMD on the "fab" front for a good 10-15 years till zen4(?) & now back in the lead again. This needs to be mentioned because while Dozer was a major mishap, its issues were compounded by the fact that it ran "hot" & used too much power ~ imagine the same chip using what 50% less power?

it still would have been hot and slow, even at half power the FX 8000s pulled nearly twice what i5s pulled under similar loads while getting beaten by i3s.

That core design was just disastrously bad.

 

[nomdeplume]

Just for Z890 as the motherboard needs to support the clock driver, the memory will be XMP certified. Should be available later next month (November).

Supposedly

Raptor lake mobo will see a Gear 2 BIOS update for 9200 or above CUDIMM. How impactful that proves for general system improvements in gaming or OC is more interesting outside of Arrow Lake thread. Just clarifying the point here and moving on.

 

[Tralalak]

Intel 1 P core 16 E core | Ultra 9 285​

source: https://www.reddit.com/r/intel/comments/1gf33lz

 

[ShrimpBrime]

compared it to 13900KS with 4 E cores disabled, both with HT on and off.

4 E-cores disabled? Why? Is that to make it closer to even playing field 20 cores? (I guess that's why, just wasn't explained at OP)

Would be nice to see the temp differences on this comparison as well. (my 2 pennies)

 

[BoggledBeagle]

Intel 1 P core 16 E core | Ultra 9 285
source: https://www.reddit.com/r/intel/comments/1gf33lz

Perhaps not only Windows, but also games themselves are confused by that PP-EEEE-PPPP-EEEEEEEE-PP (or PP-EEEEEEEE-PPPP-EEEE-PP) and PP-EEEEEEEE-PPPP-EEEEEEEE-PP core ordering. I wonder if Intel really had to do this, possibly it was necessary to improve thermals? Are 265K always missing E cores in the upper stack, or it varies?

4 E-cores disabled? Why? Is that to make it closer to even playing field 20 cores?

Yes, 8+12 vs 8+12 cores.

 

[ShrimpBrime]

Yes, 8+12 vs 8+12 cores.

Cool, that's what I figured!

How where the temps on each CPU? In the future, can you include this information?
Think it would really help to understand the cooling methods needed for each also!

These where done on liquid?? I'm a dummy, there's a picture of an Air cooler HAHA. NVM dumb question.

 

[bug]

4 E-cores disabled? Why? Is that to make it closer to even playing field 20 cores? (I guess that's why, just wasn't explained at OP)

Would be nice to see the temp differences on this comparison as well. (my 2 pennies)

Don't worry about it, I'm sure TPU will have in depth core and RAM scaling reviews up in no time.

Perhaps not only Windows, but also games themselves are confused by that PP-EEEE-PPPP-EEEEEEEE-PP (or PP-EEEEEEEE-PPPP-EEEE-PP) and PP-EEEEEEEE-PPPP-EEEEEEEE-PP core ordering. I wonder if Intel really had to do this, possibly it was necessary to improve thermals? Are 265K always missing E cores in the upper stack, or it varies?

Games have no idea how the cores are arranged. If anything, having a shorter physical path between them, makes thread migration a tad smoother. Lowering thermals because P cores are not all close together anymore was just a bonus, I guess.