Discussion Intel current and future Lakes & Rapids thread

[blckgrffn]

What are your preferred values for those settings and how much is the performance loss for ensuring systems stay stable and reliable over a long period of time?

Great question. I typically set PL1 to the old school advertised TDP (like 95W or 125W) and then PL2 based on the cooler and board. Disabling all core enhancements always. I don't use cheap parts but I'll use B365 or similar chipsets when the budget dictates it. Just doing an H370 build right now.

Real world performance sacrifices? That's a harder question to answer. I build a lot of gaming rigs (or did? gpus being what they are) and realistically reading here and other places, I didn't think it should impact them much at all. I haven't built HFR (200+ FPS) rigs that I know of, most are playing and just are happy with decent 1080P or 1440P performance when they set the slider to "high". This should almost always be a GPU limited scenario (given the mid range GPUs in question).

That said, I have used the Intel tools to investigate that the limits actually work, only extensively in one case but that was on my daughters 9700K build. It's a B365 board that doesn't have overbuilt power delivery and it has "only" Hyper 212 Evo cooler so I wanted it to work. I set PL1 to 88W (I believe - the equivalent of my AMD 3600 at the time) and PL2 120W. The Intel utility indicated that in all core workloads I was wattage limited and throttled down to about 3.4 ghz after exhausting PL2. That's something like 25% I suppose in some weird corner case scenario. Single thread boosting I am still consistently seeing (if the system can be trusted) ~4.6 ghz plus, which seems to be right where a 9700K should be.

I want to circle back though - the whole why of this is to prevent some mining malware or power virus type application (think Amazon New World and 3090's but maybe AVX or similar) to really use that default board configured headroom to escape the abilities of the board to deliver power or the CPU cooler to keep it from hard throttling (after a couple of years and some dust on everything). It is my understanding that in real world scenarios these limits should not noticeably impact the user.

Which gets me back to the power limits here (and the crazy Ampere power usages) where Intel seems bent on crazy power usage to get the last couple of percent of performance but where this costs board partners and everyone else real resources to build out to meet those demands - for reasons that don't benefit users.

It's like there is a benchmark beast mode in the firmware, enabled by default. I find it distasteful in the way I find the 5800x TDP the same way. It's not just an Intel thing.

Sorry for the long answer, I pondered it over lunch.

 

[igor_kavinski]

The Intel utility indicated that in all core workloads I was wattage limited and throttled down to about 3.4 ghz after exhausting PL2.

Intel XTU? Is it able to override whatever is set in the BIOS, meaning does it change the values in the firmware of the CPU, or does it run at Windows startup to set the values?

 

[blckgrffn]

Intel XTU? Is it able to override whatever is set in the BIOS, meaning does it change the values in the firmware of the CPU, or does it run at Windows startup to set the values?

Yes, that is what I used to evaluate the effect of PL1 and PL2 settings that I had manually configured via the UEFI interface, as hidden as they were. Call me old school or whatever but that's where I like my settings

I only used it to observe the activity of the CPU with its built in stress test. I didn't attempt to use it to manipulate any of the values (like Ryzen Master). It shows also if you are bouncing off of thermal limits, PL1/PL2 wattage limits and I think even other board imposed power limits when under load. It's been a bit, but I think I still have it setup.

The light was blinking on my wattage limits during its stress tests, so I felt confident that was the constraining factor and the CPU/motherboard were acting in accordance with my wishes.

I thought if anything would tell the truth, it would be XTU. 🤷‍♂️

 

[Ed1]

Intel XTU? Is it able to override whatever is set in the BIOS, meaning does it change the values in the firmware of the CPU, or does it run at Windows startup to set the values?

XTU does not modify the bios, it runs in windows and yes at startup if you want changes to stick but XTU doesn't have to run just the XTUservice.exe which uses extremely low resources.
I read somewhere it now supports Alder lake cores and can adjust each but have not seen it in action (I assume the bios will be able to do that too.)

 

[Zucker2k]

So anyone with 5+ Ghz Zen 3 to compare? Based on the posts of some pople it must be super efficient at this clock speed.

I don't know about 5GHz but how about 4.7GHz?



Granted, this is wall power but you can do some extrapolations between stock, PBO, and overclock. It's about 160watts for a 700mhz all core overclock, running a 5 minute loop of CB R20 from idle. Yikes!

AMD Ryzen 9 5950X Zen 3 CPU Review - KitGuru

AMD's new Zen 3 processors are finally upon us and they bring to the market a brand-new architecture

www.kitguru.net

 

[Asterox]

Hm ok, act quickly as ordered.

https://twitter.com/i/web/status/1453107202365984782

 

[Hulk]

Questions:

1. Do you think that Golden Cove will have higher IPC (throughput) than Zen 3?

2. At lower frequencies ~3500-4000MHz do you think Golden Cove will have similar efficiency? Meaning they will do the approximately the same amount of work using the same amount of Joules?

3. If you think #2 will be true do you think at a certain frequency beyond 4000MHz, Zen 3 will scale more efficiently with frequency than Golden Cove from an energy consumption point-of-view? Meaning Golden Cove energy consumption will begin to increase at a faster rate then Zen 3 as frequency scales beyond 4000+MHz?

4. If you believe #3 to be false and they will scale approximately the same with power as frequency increases, yet Zen 3 tops out at a lower frequency (power) than Golden Cove is it fair to say Golden Cove is less efficient than Zen 3 because it can reach higher frequencies at the expense of huge power draw?

 

[IntelUser2000]

Questions:

1. Do you think that Golden Cove will have higher IPC (throughput) than Zen 3?

Yes. It's clearly a wider architecture.

2. At lower frequencies ~3500-4000MHz do you think Golden Cove will have similar efficiency? Meaning they will do the approximately the same amount of work using the same amount of Joules?

3. If you think #2 will be true do you think at a certain frequency beyond 4000MHz, Zen 3 will scale more efficiently with frequency than Golden Cove from an energy consumption point-of-view? Meaning Golden Cove energy consumption will begin to increase at a faster rate then Zen 3 as frequency scales beyond 4000+MHz?

No.

If you believe #3 to be false and they will scale approximately the same with power as frequency increases, yet Zen 3 tops out at a lower frequency (power) than Golden Cove is it fair to say Golden Cove is less efficient than Zen 3 because it can reach higher frequencies at the expense of huge power draw?

I think just like Tigerlake, Zen will win out perf/watt wise against Golden Cove in all frequency ranges, but due to Intel optimizing for the higher clocks, the gap might shrink.

In the end Golden Cove is designed for higher frequencies, so comparing Zen 3 running at 5.3GHz makes no sense - since the latter chip shouldn't reach that high anyways. It'll blow past power since it's way out of it's design target, just like using LN2 cooling but needing 1.6V for achieving overclocking records. I bet you could get Gracemont core to 5GHz too. It just wouldn't make sense, since it's way out of it's design frequency range so not only it would require ridiculous voltage to reach that, it'd be less efficient as well.

Zen 3 doesn't need to reach 5.3GHz since they can fit 16 cores in a reasonable power envelope and die size, while Golden Cove can't. Simple as that.

 

[gdansk]

That's inline with my expectations as well. I suspect GC will be less efficient (work per joule) at all frequencies. But that is Gracemont's reason of existence...

 

[Hans Gruber]

Here is hoping that Alder Lake outperforms expectations. Zen 3 will win on power consumption regardless. I want AMD to be like AMD in the recent past, price/performance leader. I want a 5800X and my strike price is $250-$275. If Alder Lake is really good on performance, AMD will do what they do best. Cheap CPU's. I got my 3600 for $172 and thought that was a decent value. I just picked up a B450 simply for one last upgrade until DDR5. The cost savings (B450) go to a 5800x or the equivalent if AMD comes out with a Zen3 refresh CPU. I got the P31 1TB NVMe drive so I have no need for PCI-e 4.0

I will sit back for the next 2 or 3 years watching the DDR5 market.

 

[gdansk]

Here is hoping that Alder Lake outperforms expectations. Zen 3 will win on power consumption regardless. I want AMD to be like AMD in the recent past, price/performance leader. I want a 5800X and my strike price is $250-$275. If Alder Lake is really good on performance, AMD will do what they do best. Cheap CPU's. I got my 3600 for $172 and thought that was a decent value. I just picked up a B450 simply for one last upgrade until DDR5. The cost savings (B450) go to a 5800x or the equivalent if AMD comes out with a Zen3 refresh CPU. I got the P31 1TB NVMe drive so I have no need for PCI-e 4.0

I will sit back for the next 2 or 3 years watching the DDR5 market.

The expectations here seem very low, so I'd be surprised if Alder Lake does not exceed that low bar. The benchmarks coming out of China look rather promising. Power efficiency does not look promising for the desktop chips. But who cares for that in a desktop? Not me...

 

[Accord99]

2. At lower frequencies ~3500-4000MHz do you think Golden Cove will have similar efficiency? Meaning they will do the approximately the same amount of work using the same amount of Joules?

At those frequencies, core for core I'm expecting GC to be more power-efficient. Package power of 78.5W running AIDA Stress FPU at 6x4 GHz is pretty good (my undervolted 8-core Tiger Lake at 3.4GHz needs 56W for Stress FPU and 49W for CB R20) :

If the GC cores behave like WC, it would probably have a package power of ~70W to reach a CB 20 score that beats the 5600X.

 

[Zucker2k]

Yes. It's clearly a wider architecture.





No.



I think just like Tigerlake, Zen will win out perf/watt wise against Golden Cove in all frequency ranges, but due to Intel optimizing for the higher clocks, the gap might shrink.

In the end Golden Cove is designed for higher frequencies, so comparing Zen 3 running at 5.3GHz makes no sense - since the latter chip shouldn't reach that high anyways. It'll blow past power since it's way out of it's design target, just like using LN2 cooling but needing 1.6V for achieving overclocking records. I bet you could get Gracemont core to 5GHz too. It just wouldn't make sense, since it's way out of it's design frequency range so not only it would require ridiculous voltage to reach that, it'd be less efficient.....

I find your answer to be contradictory. In simple terms, what would prevent Zen3 from running all cores at 5GHz, from the 5600x to the 5950x? Short answer, POWER!
That means Zen3 will simply stop scaling with power at a certain point beyond 4GHz, while GC will blow past 5GHz with ease. So how can you claim efficiency for Zen 3 "at all frequencies."?
As you've aptly stated, Intel's willingness to tap into higher wattage for higher frequencies is deliberate. It simply is a strategy of fighting many 'slower' cores with less fewer cores. Operating at such high frequencies comes with a certain sacrifice of efficiency. If we understand this going forward, then we'll not be making blanket statements when we very well know Zen3 is not even capable of playing in that 5.3GHz without exotic cooling.
If we look at the power comparison between Zen3 and Tigerlake by Anandtech, I believe WC maintained linear efficiency scaling beyond 90w and showed no sign of waning off. Zen3, on the other hand, could only maintain linear scaling up to around 70-75w iirc, so that by around the 90w mark, it had gradually dipped to the same level as WC. Meaning, from that point forward, WC was going to demonstrate better efficiency over Zen3.
As @Accord99 has indicated, the GC chip to demonstrate the best efficiency capabilities of the GC arch is the i5 12400, with 6c/12t just like the 5600x. Preliminary results show it's beating the 5600x with same cores and threads clocked a tad lower, and using less energy to do so. So, I'm looking forward to all the reviews and praying Intel throws a surprise 12400 in the mix just to highlight this point. In any case, the 12600k is not aggressively clocked like the 12900k, so maybe we'll see some glimpses of GC efficiency with the less aggressively clocked chips.
Edit:
The power scaling analysis was done by techspot, not Anandtech. https://www.techspot.com/review/2262-intel-core-i7-11800h/

 

[ondma]

I find your answer to be contradictory. In simple terms, what would prevent Zen3 from running all cores at 5GHz, from the 5600x to the 5950x? Short answer, POWER!
That means Zen3 will simply stop scaling with power at a certain point beyond 4GHz, while GC will blow past 5GHz with ease. So how can you claim efficiency for Zen 3 "at all frequencies."?
As you've aptly stated, Intel's willingness to tap into higher wattage for higher frequencies is deliberate. It simply is a strategy of fighting many 'slower' cores with less fewer cores. Operating at such high frequencies comes with a certain sacrifice of efficiency. If we understand this going forward, then we'll not be making blanket statements when we very well know Zen3 is not even capable of playing in that 5.3GHz without exotic cooling.
If we look at the power comparison between Zen3 and Tigerlake by Anandtech, I believe WC maintained linear efficiency scaling beyond 90w and showed no sign of waning off. Zen3, on the other hand, could only maintain linear scaling up to around 70-75w iirc, so that by around the 90w mark, it had gradually dipped to the same level as WC. Meaning, from that point forward, WC was going to demonstrate better efficiency over Zen3.
As @Accord99 has indicated, the GC chip to demonstrate the best efficiency capabilities of the GC arch is the i5 12400, with 6c/12t just like the 5600x. Preliminary results show it's beating the 5600x with same cores and threads clocked a tad lower, and using less energy to do so. So, I'm looking forward to all the reviews and praying Intel throws a surprise 12400 in the mix just to highlight this point. In any case, the 12600k is not aggressively clocked like the 12900k, so maybe we'll see some glimpses of GC efficiency with the less aggressively clocked chips.

Yes, Intel has designed for higher frequencies, but I would assume they would rather have more cores able to perform the same work at lower frequencies. They simply cannot do this because of the size of the GC cores and the ring bus design. I would consider Comet Lake for sure and Rocket Lake in most cases, to be a loss to Zen 3, since they have equal or lower performance in both single and multi thread design, while using more power. If AL can take back the single thread lead while giving similar multi thread performance, it will at least offer some advantages for the higher power consumption.

 

[phillyman36]

Aren't the announcements(cpu, mobo etc) and pre orders supposed to be today 10/27? If so what time EST?

 

[Zucker2k]

ADL-S pricing leaks?

i9 12900K - $589
i9 12900KF - $564
i7 12700K - $409
i7 12700KF - $384
i5 12600K - $289
i5 12600KF - $264

Intel 12th Gen Core "Alder Lake-S" final specifications and pricing leak ahead of launch - VideoCardz.com

Intel Alder Lake will be more expensive than Rocket Lake We finally have the full specifications and pricing of the upcoming desktop series. Intel Core i9-12900K featuring 16 cores (8 Performance and 8 Efficient) and 24 threads will carry a boost clock up to 5.2 GHz, there is no Thermal...

videocardz.com

 

[Karnak]

That gaming perfomance looks quiet... interesting for Intel's own benchmarks. Will probably be even worse in third party reviews.

 

[diediealldie]

So, I'm looking forward to all the reviews and praying Intel throws a surprise 12400 in the mix just to highlight this point. In any case, the 12600k is not aggressively clocked like the 12900k, so maybe we'll see some glimpses of GC efficiency with the less aggressively clocked chips.
Edit:
The power scaling analysis was done by techspot, not Anandtech. https://www.techspot.com/review/2262-intel-core-i7-11800h/

This. Based on a rough extrapolation, 11800H (2nd best bin from Intel 10SF wafer) intersects with 5900HX (Very best bin from AMD Zen 3 wafer TSMC7) near 100W of package power. This means even Tiger Lake can achieve similar performance to power in 12 watts per core battle. This isn't a big power in desktop battle. Ryzen 5950X cores use more than 10 watts in 4.X Ghz region.

Intel 7(Former 10ESF) and new uarch(GC) will push Intel's curve further downwardsupwards, so Intel same big core count battle, Intel will win for sure. That is where i5 and i7 will be. But If we put PPA(performance per area) battle, then Intel will suffer due to the massive size of Golden Cove cores. i5 and i7 will not give intel a huge margin.

This is where i9 and Ryzen 5950 are fighting, and Intel plans to overcome this problem by adding tons of Atom cores, assuming that there are little workloads which needs only "Tons of big cores". This is quite a nice approach, knowing that even games are getting benefits from Hyperthreading. GM cores are still stronger than GC HT.

Intel might catch up uarch of the big core as well looooooong later. That remains to be seen, but at least I can show some respects to Intel because they tried to work together with MS to solve their fab + uarch conundrum.

 

[JoeRambo]

That gaming perfomance looks quiet... interesting for Intel's own benchmarks. Will probably be even worse in third party reviews.

Yeah, Anandtech and JEDEC timings on DDR5 will destroy them. Large L3 and L2 can carry only so much when main memory is slower than mobile CPUs from fruit vendors.

 

[exquisitechar]

ADL-S pricing leaks?

i9 12900K - $589
i9 12900KF - $564
i7 12700K - $409
i7 12700KF - $384
i5 12600K - $289
i5 12600KF - $264

Intel 12th Gen Core "Alder Lake-S" final specifications and pricing leak ahead of launch - VideoCardz.com

Intel Alder Lake will be more expensive than Rocket Lake We finally have the full specifications and pricing of the upcoming desktop series. Intel Core i9-12900K featuring 16 cores (8 Performance and 8 Efficient) and 24 threads will carry a boost clock up to 5.2 GHz, there is no Thermal...

videocardz.com

Looks good. People who said Intel would keep their pricing structure similar to before were right, I was expecting something else. Can't wait for in depth reviews.

 

[insertcarehere]

Interesting chinese thread with an insider who appears to have Alder Lake CPU on hand, here.

Rough translated comments:
-i9 12900k can peg a Noctua U14S at 100C, "real" TDP at 24xW (likely refers to max turbo power limit?).
-Large increase in performance w/ games & other software across the board, makes 11th gen looks like garbage. Performance (for 12900k) is "incomparable" assuming it runs at the above turbo power limit.
-Significantly lower "real" TDPs for the 12th gen i7 & i5, which would be fine with air cooling.
-Difference between 12th gen & 11th gen: For 12th gen, as long as supporting hardware & conditions permit (I. E Temps), the processor can run at "real" TDP (Turbo/PL2) indefinitely. Not necessary to manually OC.
-Early DDR5 is most likely not worth it, expensive and latency too high.

 

[Timorous]

ADL-S pricing leaks?

i9 12900K - $589
i9 12900KF - $564
i7 12700K - $409
i7 12700KF - $384
i5 12600K - $289
i5 12600KF - $264

Intel 12th Gen Core "Alder Lake-S" final specifications and pricing leak ahead of launch - VideoCardz.com

Intel Alder Lake will be more expensive than Rocket Lake We finally have the full specifications and pricing of the upcoming desktop series. Intel Core i9-12900K featuring 16 cores (8 Performance and 8 Efficient) and 24 threads will carry a boost clock up to 5.2 GHz, there is no Thermal...

videocardz.com

Those gaming scores are interesting.

Questions though.

DDR4 or DDR5? JEDEC spec or tighter timings/higher speeds?

W11? Before or after the L3 cache issue with zen3 was fixed?

Tdp and tau settings? Intel Spec or motherboard spec?

I foresee a lot of variance in the results come review day. Anandtech who use JEDEC specs and Intel tdp/tau specs may show considerably worse results than HUB who will likely run above spec ram and allow the motherboard to dictate power and tau limits. GN who will likely run above spec ram but stick to Intel Spec power and tau limits will be an interesting middle ground.

On top of that W10 vs W11 testing, will reviewers stick to a single OS for reviews or will they choose the best performing for each platform?

 

[phillyman36]

Newegg has Micron ddr5 modules up for preorder. 2x8@$116.00 and 2x16@ $212.00. Not what i really want but i grabbed a set just in case.

Are you a human?

www.newegg.com

Are you a human?

www.newegg.com

 

[mikk]

At those frequencies, core for core I'm expecting GC to be more power-efficient. Package power of 78.5W running AIDA Stress FPU at 6x4 GHz is pretty good (my undervolted 8-core Tiger Lake at 3.4GHz needs 56W for Stress FPU and 49W for CB R20) :

And it confirms what I said, AIDA64 FPU runs at a higher power than Cinebench MT on Intel. 78W AIDA64 FPU could translate into 70W Cinebench MT which is really good.

 

[mikk]

Those gaming scores are interesting.

Questions though.

DDR4 or DDR5? JEDEC spec or tighter timings/higher speeds?

W11? Before or after the L3 cache issue with zen3 was fixed?

Tdp and tau settings? Intel Spec or motherboard spec?

I foresee a lot of variance in the results come review day. Anandtech who use JEDEC specs and Intel tdp/tau specs may show considerably worse results than HUB who will likely run above spec ram and allow the motherboard to dictate power and tau limits. GN who will likely run above spec ram but stick to Intel Spec power and tau limits will be an interesting middle ground.

On top of that W10 vs W11 testing, will reviewers stick to a single OS for reviews or will they choose the best performing for each platform?

Also this:

https://twitter.com/i/web/status/1453195027128127490

E-cores disabled or enabled?