News Intel's performance-enhancing IPO program debuts in gaming PCs across China — overclocked performance with full warranty

Presumably, these CPUs will be sold at a premium. If all Arrow Lake CPUs were capable of those speeds, they'd just ship that way. Therefore, the IPO-capable models seem like they must be binned.
 
If all Arrow Lake CPUs were capable of those speeds, they'd just ship that way.
Yes and no.

As I've been slowly making my way through how ARL OC works I can say the disclosed settings here are pretty mild. The ring OC for ARL is extremely hit and miss, but this is only +100Mhz which should work across the board. The D2D/NGU going to a 31x is also just going to work as what impacts stability here is DRAM speed with low latency. The E-cores OC pretty consistently to 4.9/5Ghz without much trouble. The P-core OC seems like it should be turbo ratio based so that shouldn't particularly be problematic either as limited clock scaling and all core tend to be the issues here. I do wonder if Intel has a method for picking which cores clock best that they aren't sharing though because this is much more important with ARL than RPL.

Overall these are very safe OC figures, but I wouldn't be surprised if the CPUs now have RPL efficiency (note the 280/350 PL1/PL2) to ensure they work across the board. It's not a tradeoff I would make, but a lot of people don't really seem to care much about CPU efficiency if it goes faster.
 
Last edited:
  • Like
Reactions: PixelAkami
From what I saw, they're increasing the die-to-die speed from 2.1 GHz to 3.1 GHz, which sounds like a lot.
It does sound like a lot, but the only instability issues with high D2D clocks I've seen are with high speed memory (over 8000) with tight timings (under 9.5ns). My 265K runs 32x for both D2D and NGU without touching voltages with DDR5-7200 CL34 with tweaked subtimings so it matches lower capacity kits. This is also one of the things Splave mentioned in the ARL OC article he did for Tom's.
 
Why on Earth are they overclocking the cores instead of the various interconnect settings. I guarantee there’s much more performance in optimizing the interconnect settings than there is in a 200MHz clock speed bump. 200MHz isn’t much with the click speeds CPUs are doing these days. I don’t personally own Arrow Lake but everyone that does says the performance gains are in the interconnect speeds rather than core clocks.
 
Why on Earth are they overclocking the cores instead of the various interconnect settings. I guarantee there’s much more performance in optimizing the interconnect settings than there is in a 200MHz clock speed bump. 200MHz isn’t much with the click speeds CPUs are doing these days. I don’t personally own Arrow Lake but everyone that does says the performance gains are in the interconnect speeds rather than core clocks.
If you look at the source you'd see they're tweaking D2D, NGU and Ring clocks as well as CPU/DRAM.

It's also mentioned in the article:
IPO targets the CPU (P-cores, E-cores, Ring-bus, NGU, D2D interconnects, PL1 and PL2) and the RAM (Transfer speeds and timings).
 
I do wonder if Intel has a method for picking which cores clock best that they aren't sharing though because this is much more important with ARL than RPL.
https://www.intel.com/content/www/us/en/support/articles/000021587/processors.html
I don't think they changed this, they would have released a statement.
From what I remember reading they check which cores need the lowest amount of voltage and power and use those since those will have the highest headroom.
Intel® Turbo Boost Max Technology 3.0 Technology identifies the best performing core(s) on a processor and provides increased performance on those cores through increasing frequency as needed by taking advantage of power and thermal headroom.

Due to production differences, processor cores vary in maximum potential frequency. Intel® Turbo Boost Max Technology 3.0 identifies up to two of the fastest cores on your CPU, known as “favored cores”. Then it applies a frequency boost to those cores (or that core) and directs critical workloads to them.
 
Why on Earth are they overclocking the cores instead of the various interconnect settings. I guarantee there’s much more performance in optimizing the interconnect settings than there is in a 200MHz clock speed bump. 200MHz isn’t much with the click speeds CPUs are doing these days.
They're not mutually-exclusive. Not everything is totally bottlenecked on L3 cache or memory, in which cases a core bump will indeed increase performance. Keep in mind the P-Core limit is 3.8% faster and E-Core limit is 6.5% faster. These are on par with the difference between Raptor Lake (Gen 13) and its refresh (Gen 14), which probably stands as a good example of what's feasible without silicon-level changes.

As for interconnect, I already pointed out that they said the die-to-die interconnect was boosted from 2.1 to 3.1 GHz, which seems massive to me. Like, it's hard for me to believe they even had so much headroom.

Uncore also got a substantial boost. From what I'm reading, that should affect communication with the memory controller. However, Skatterbencher's experiments with this showed very little impact on Y-cruncher scores.

Here's the list of claimed improvements.

Limit​
New​
Original​
Increase​
P-Core (GHz)
5.4​
5.2​
3.8%​
E-Core (GHz)
4.9​
4.6​
6.5%​
Ring (GHz)
4.0​
3.9​
2.6%​
Uncore (GHz)
3.1​
2.6​
19.2%​
Die-to-Die (GHz)
3.1​
2.1​
47.6%​
PL1 (W)
280​
125​
124.0%​
PL2 (W)
350​
250​
40.0%​

While the PL1 increase seems massive, consider that most gamers were probably already doing 250 W (or unlimited Tau - same thing), in which case it's only a 12% increase.

Source:
 
Last edited:
https://www.intel.com/content/www/us/en/support/articles/000021587/processors.html
I don't think they changed this, they would have released a statement.
From what I remember reading they check which cores need the lowest amount of voltage and power and use those since those will have the highest headroom.
TB 3.0 only covers up to two core boost while Intel has turbo multipliers for every step. I suppose they could just be giving a +200Mhz boost to the stock settings which only cover 2 core and all core. It's just not what I think of when overclocking as the individual turbo optimization is right there.
 
Have you found a good place to look these up? I found a way to query it on Linux (and I'm sure Windows overclocking tools can do this, as well), but that only helps if you're already running the CPU you want to know about.
I don't think proper turbo information is available anymore outside of the CPUs themselves. Intel does just do two and all core multipliers for stock behavior so once you know those they're the same for all of the same SKU. The only thing that I think changes is the V/F curve on a per CPU basis.

TPU reviews are the only ones I can think of off hand that do per thread clock speed monitoring. I've actually referenced those before picking what to buy.
 
I don't think proper turbo information is available anymore outside of the CPUs themselves. Intel does just do two and all core multipliers for stock behavior so once you know those they're the same for all of the same SKU. The only thing that I think changes is the V/F curve on a per CPU basis.
What I mean is this. On Linux, you can run the command: sudo turbostat --Summary 2>&1 | grep "MHz max turbo"

For the N305, you get the following output:
Code:
30 * 100.0 = 3000.0 MHz max turbo 8 active cores
31 * 100.0 = 3100.0 MHz max turbo 7 active cores
32 * 100.0 = 3200.0 MHz max turbo 6 active cores
33 * 100.0 = 3300.0 MHz max turbo 5 active cores
35 * 100.0 = 3500.0 MHz max turbo 4 active cores
35 * 100.0 = 3500.0 MHz max turbo 3 active cores
37 * 100.0 = 3700.0 MHz max turbo 2 active cores
38 * 100.0 = 3800.0 MHz max turbo 1 active cores

No matter what you run or what you increase the power limits to, this is the max clock speed the CPU firmware will allow your CPU cores to run at. Exceeding these limits is only possible on K-series CPUs, as it's the definition of overclocking.

TPU reviews are the only ones I can think of off hand that do per thread clock speed monitoring. I've actually referenced those before picking what to buy.
TPU's testing is potentially affected by power limits. Although they don't use a stress test, they still use a FP-heavy workload. What I'd do is use something like a simple Fibonacci loop, if you just want to see the upper limit on clock speeds vs. active cores.
 
I don't think proper turbo information is available anymore outside of the CPUs themselves. Intel does just do two and all core multipliers for stock behavior so once you know those they're the same for all of the same SKU. The only thing that I think changes is the V/F curve on a per CPU basis.

TPU reviews are the only ones I can think of off hand that do per thread clock speed monitoring. I've actually referenced those before picking what to buy.
No, they do a lot more different things all together.
What you are saying is only turbo 1 and 2, now there is also thermal and adaptive boost plus turbo 3 for the best two cores.
They can't release any table with fixed numbers anymore because all of this depends on what you are running how your system is configured how much cooling you have and so on.
ikk0lakbcpn61.jpg
 
They can't release any table with fixed numbers anymore because all of this depends on what you are running how your system is configured how much cooling you have and so on.
Cooling appears to affect only TVB. If you're not thermally-throttling and not hitting the power-limit, then is the table I posted applicable before or after TB 2.0 (for those CPUs which have it)?
 
Cooling appears to affect only TVB. If you're not thermally-throttling and not hitting the power-limit, then is the table I posted applicable before or after TB 2.0 (for those CPUs which have it)?
Cooling affects all of it, since it all goes up to 100, if you have worse cooling you are going to be at 100 a lot "sooner" ,as in while running less demanding stuff, while good cooling will let the CPU boost higher even when running more demanding stuff.

The n305 table appears to be turbo 2 but I could be wrong.
 
  • Like
Reactions: bit_user
No, they do a lot more different things all together.
What you are saying is only turbo 1 and 2, now there is also thermal and adaptive boost plus turbo 3 for the best two cores.
They can't release any table with fixed numbers anymore because all of this depends on what you are running how your system is configured how much cooling you have and so on.
ikk0lakbcpn61.jpg
TB 2 limits are hard limits for higher than 2 cores. Every other boosting technology is based on one or two cores. You'll find that no matter how good your cooling is once you go above 2 cores the system will not clock higher than whatever it's set at. As far as I'm aware it's been this way since ADL launch.
 
What I mean is this. On Linux, you can run the command: sudo turbostat --Summary 2>&1 | grep "MHz max turbo"
I'll verify with my 12700K when I'm not using a mobile device, but I'm pretty sure the clock table that command pulls is not accurate for real world behavior. It shows similar steps on my 12700K for P-cores but I've never seen anything but max turbo clocks or max all core clocks in usage. That of course doesn't mean it hasn't happened, but that I haven't noticed which is why I'll check when I can actually interact with that machine properly.
 
You'll find that no matter how good your cooling is once you go above 2 cores the system will not clock higher than whatever it's set at.
Yeah, if you have put "all core" to the max all core clock then it will not go above that....that also means that you have thermal and adaptive turbo disabled and have an all core overclock applied, forcing it to run the fixed all core clocks no matter what, until the CPU itself hits some limit and throttles down to safe itself.
 
The n305 table appears to be turbo 2 but I could be wrong.
I noticed, in the unfiltered output, that it says it's querying MSR_TURBO_RATIO_LIMIT. On a Gen 12 hybrid CPU, that tool also claims to query MSR_SECONDARY_TURBO_RATIO_LIMIT, showing the E-core limits. I found where those registers are listed in the Intel Software Developers guides (Volume 4 lists the MSRs), but it doesn't really say anything other than calling them the "maximum turbo ratio limit". When cross-referencing against the other volumes, that's a term I didn't find concretely defined.

Sure enough, that hybrid CPU supports TB 2.0 and I've never been able to coax any more than 5.1 GHz from it. So, it seems you're right that it's the TB 2.0 limit (or, at least one that TB 2.0 cannot exceed, since the N305 doesn't claim to support TB 2.0).
 
Last edited:
I'm pretty sure the clock table that command pulls is not accurate for real world behavior. It shows similar steps on my 12700K for P-cores but I've never seen anything but max turbo clocks or max all core clocks in usage.
Well, I'm on non-K CPUs, but my experience has been validated by that table. As I scale up threads running lightweight tasks, clocks do scale down according to the table, no matter how low my temps or how high my power limits.