Intel Core i7-14700K has an 8P+12E Core Configuration

[JustBenching]

Well I don't have the chips you're looking for, so you can keep that dishonest tag to yourself

As for testing the 12900k if you can (re)enable AVX512 why don't you do that & test the max power consumption while setting the TDP at say 65W or whatever the BIOS allows, or is that way too honest for you?

I can test whatever you want, I have to downgrade the bios to enable avx though. But why do you keep insisting about the pl2 numbers? The cpu will never, under any circumstances, pull more than the pl2 it's set to. Ever. Pl2 is the max wattage the cpu will pull for the TAU duration, after which it will drop down to the pl1 number.

The TDP not matching power draw is an amd only thing, tdp is used to describe what cooler ryzen cpus need, not how much power they pull. The actual power draw can be seen from the PPT numbers. Intel doesnt have any tdp settings in the bios. It's just pl1, pl2 and TAU. That's it.

You don't have the cpus but that guy does, you can ask him politely. Why do you think I insist about the efficiency part? I have actually tested it.

1 zen 4 core is around 5-10% more efficient than an ald p core. A rpl p core is around 10% faster than a zen 4 core. So it goes without saying then when Intel cpus have e cores on top of that they Will be more efficient at same wattages.

 

[R0H1T]

Because you keep on insisting that it's a hard limit, AVX512 is the only one that should blow past it.

As for AMD TDP has never been equal to power consumption since Zen debuted, don't remember about Dozer & don't care either! From the link the other poster shared, these are power consumption numbers ~

Needless to say if you do go back to the old BIOS, that'd be at your own risk so take care about that.

 

[JustBenching]

Because you keep on insisting that it's a hard limit, AVX512 is the only one that should blow past it.

As for AMD TDP has never been equal to power consumption since Zen debuted, don't remember about Dozer & don't care either! From the link the other poster shared, these are power consumption numbers ~

View attachment 304487

That's what I said, tdp on amd isn't power consumption, power consumption is higher than tdp. Those numbers are from... GAMING. Thought we were talking about mt performance, no?

 

[R0H1T]

The numbers are lower coming from 5950x to 7950x, I'm assuming it would be for the same task but which one that isn't specified. 5950x has lower TDP than 7950x so there's that.

 

[JustBenching]

The numbers are lower coming from 5950x to 7950x, I'm assuming it would be for the same task but which one that isn't specified. 5950x has lower TDP than 7950x so there's that.

Yes, zen 4 is much more efficient in gaming than zen 3 was. Zen 3 was pretty bad. Intel on the other hand was pretty good on 12th gen and then they messed up on 13th.

 

[R0H1T]

I'd also like to add that written reviews are kind of a dying breed, it's a thankless job & you'll be paid maybe a 10th at best what Linus would be for 5 mins of entertainment on YT on a good day, he'd also not test this much in depth or probably do the testing himself! That's why I usually refrain from just saying the numbers are plain wrong. It takes 100's of hours to get there & even an inch here or there could mean catastrophe for the reviewer.

 

[chrcoluk]

How fast was gear 1 on ADL and then RPL? Did they even have gears? Are gears only for pairing with DDR4 or did they also have gears when used with DDR5? Was there a gear difference between different models of the same architecture? For example, i7 11700K is 3200MHz G1 while i9 11900K was 3600MHz G1.

I seem to have a wall at over 3200, I can only hit 3400 or higher stable with gear 2, some on reddit hit same wall, but then is also lots who can hit decent ram speeds. Seems to me RPL and ADL have a high variance in IMC quality. I couldnt try 3333 as not an option in my bios.

12 e-cores?

A genuine query.... W T F is intel doing? These now look more like server/workstation/encoding chips.... why not cut cost, cut thermals, cut power consumption and just drop another 8C 16T i7 variant without the E?

Or is Intel just lost in competing for the higher MT score card?

Reduced cache? Where X3D has excelled Intel is reducing memory.

I don't get it... unless 14th Gen earns great points with efficiency, power consumption and significantly superior ST performance uplift - doesn't excite me.

Marketing team wants good cinebench score, hence e-core boost is getting them what they want. Reviewers have such a huge influence now, its also the reason they gone the direction of unlimited power and low temps is wasted performance.

Then it's obvious you haven't tried them.

Here is a video review, just take a look

You HAVE to? What happens if you don't?

I like the from the wall test at the end, I think dabuer had similar results as well when analysing the actual power efficiency instead of just max power headlines.

 

[JustBenching]

I like the from the wall test at the end, I think dabuer had similar results as well when analysing the actual power efficiency instead of just max power headlines.

Yeah, reviewers test out of the box with 4096 power limits on the intel parts, as if anyone would ever run hour long multithreaded workloads at 350watts. That's absurd and it gives you no useful information. No professional is going to run 10 hour long blenders at 350 watts, so... It's pretty pointless to even test it. It's mostly click bait amd fans get drawn to such reviews so they can parrot how inefficient intel is

 

[A Computer Guy]

Yeah, reviewers test out of the box with 4096 power limits on the intel parts, as if anyone would ever run hour long multithreaded workloads at 350watts. That's absurd and it gives you no useful information. No professional is going to run 10 hour long blenders at 350 watts, so... It's pretty pointless to even test it. It's mostly click bait amd fans get drawn to such reviews so they can parrot how inefficient intel is

I run hour long multithreaded workload at least once a week when I zip a backup of my main virtual machine so it's not totally out of the realm of practicality to test for that time duration but it probably meaningless for the average PC user and in my case the energy usage doesn't matter in this use case unless I have a power outage and need to run off a battery.

 

[JustBenching]

I run hour long multithreaded workload at least once a week when I zip a backup of my main virtual machine so it's not totally out of the realm of practicality to test for that time duration but it probably meaningless for the average PC user and in my case the energy usage doesn't matter in this use case unless I have a power outage and need to run off a battery.

Do you care about efficiency? If you did, you wouldn't be running 10 hours at 350 watts. If you don't, then it doesn't matter, Efficiency graphs aren't for you

 

[A Computer Guy]

Do you care about efficiency? If you did, you wouldn't be running 10 hours at 350 watts. If you don't, then it doesn't matter, Efficiency graphs aren't for you

Generally yes but I was talking about running a mt task for 1 hour a week in my use case. A 10 hour task is a different ballpark where someone is probably actually doing some work if it actually takes that long and bean counters might get involved. You could make an argument by scaling that up for lets say 10 employees running hours of mt tasks per day and you have a case where looking at efficiency makes even more sense so someone can decide what to do about the costs. Take those same cores and put them on a farm and you have a whole economy of scale to deal with.

(follow up question)
So where are 12th and 13th (and 14th) gen cores used in server farms and how is that working out for Intel? In the efficiency conversations this element doesn't seem to come up much in discussion.

 

[R0H1T]

And you'd think this would be pretty clear looking at this chart, but here we are ~

TOP500 List - June 2023 | TOP500

www.top500.org

This would probably top the efficiency charts next year!

El Capitan Installation Begins: First APU-based Exascale System Shaping Up For 2024

www.anandtech.com

 

[JustBenching]

(follow up question)
So where are 12th and 13th (and 14th) gen cores used in server farms and how is that working out for Intel? In the efficiency conversations this element doesn't seem to come up much in discussion.

Server parts amd dominates in efficiency. Intel can't scale desktop cpus to servers, since the ring bus they use on the desktops doesn't work. For more cores they need a different approach, and amds interconnect is better.

Intel sapphire rapids are the latest intel server cpus and they use 12th gen cores. They dominate in AI workloads due to on chip accelerators but are kinda average or even flat out bad on other workloads.

Basically, the reason amd dominates in servers is the same reason their CPUs aren't that compelling in desktops. The ring bus intel is using is just better for home users, consuming much less power on idle and light loads which are the majority of what home users do with their pcs. My zen 3 chip consumes 30 to 40w just browsing the web while intel just needs 5. Ring bus superiority.

 

[R0H1T]

That is true only to a certain extent, AMD's actual cores are more efficient than Intel & you'd argue their memory subsystem as well.

AMD’s EPYC 7J13: Zen 3 Customized

Recently, curiosity drove me to do some microbenchmarking on Lambda Cloud’s A100 instance.

chipsandcheese.com

AMD’s Zen 4 Part 1: Frontend and Execution Engine

AMD’s Zen 4 architecture has been hotly anticipated by many in the tech sphere; as a result many rumors were floating around about its performance gains prior to its release.

chipsandcheese.com

AMD’s Zen 4, Part 2: Memory Subsystem and Conclusion

Please go through part 1 of our Zen 4 coverage, if you haven’t done so already.

chipsandcheese.com

AMD’s Zen 4, Part 3: System Level Stuff, and iGPU

We covered Zen 4’s core architecture in depth in two articles.

chipsandcheese.com

 

[JustBenching]

AMD's actual cores are more efficient than Intel

In general, they are, yes. You can see that in laptops. For desktops though, it doesn't matter, cause desktop amd cpus have the IF.. So no matter if their actual cores are more efficient (which is not true for all workloads, but let's go with it), in practice they fall behind in power usage due to the IF and the lower amount of cores on similar skus (R7 vs i7 etc.).

 

[P4-630]

Raptor Lake Refresh CPU core configurations

 

[AusWolf]

As ive asked repeatedly, we can test it. Same wattage, same mt benchmark. But nobody dares to be honest, they wanna keep claiming how efficient amd is, even though it isn't. So yeah, I'm wasting my time. Test it or stop arguing

Well I don't have the chips you're looking for, so you can keep that dishonest tag to yourself

As for testing the 12900k if you can (re)enable AVX512 why don't you do that & test the max power consumption while setting the TDP at say 65W or whatever the BIOS allows, or is that way too honest for you?

I don't know if you two are still arguing, but here's one: my 7800X3D scores roughly 18,300 in Cinebench R23 multi-core while consuming 84 W max (82-83 W average) - that is total package power. Limiting PL1 and PL2 on Intel, or PPT on AMD to 84 W (or TDP to 65 W which uses an 88 W PPT) should give you comparable results.

I also have a 7700X, but I can't test it now, as I'm about 1,500 miles away from home until the end of next week.

 

[Why_Me]

I see the AMD cult is out in full force today.

 

[A Computer Guy]

In general, they are, yes. You can see that in laptops. For desktops though, it doesn't matter, cause desktop amd cpus have the IF.. So no matter if their actual cores are more efficient (which is not true for all workloads, but let's go with it), in practice they fall behind in power usage due to the IF and the lower amount of cores on similar skus (R7 vs i7 etc.).

Ok I feel like we are making some progress finally. Correct me if I'm wrong or if what I'm about to say is colossally stupid but it seems to me R7 vs i7 are most efficient at different power levels (on desktop) and that might be why it's so difficult to get some consensus on efficiency in the conversation?

 

[AusWolf]

I see the AMD cult is out in full force today.

Nope. I just want the argument to be over, and I'm curious about the results, regardless of who wins.

I also have several other CPUs besides the two that I've mentioned, they're all Intel (various 1st, 4th, 7th and 11th gen Core).

Ok I feel like we are making some progress finally. Correct me if I'm wrong or if what I'm about to say is colossally stupid but it seems to me R7 vs i7 are most efficient at different power levels (on desktop) and that might be why it's so difficult to get some consensus on efficiency in the conversation?

I think that's exactly the case, although without having a 12th or 13th gen Core i7 at hand, I can't know for sure.

 

[JustBenching]

I don't know if you two are still arguing, but here's one: my 7800X3D scores roughly 18,300 in Cinebench R23 multi-core while consuming 84 W max (82-83 W average) - that is total package power. Limiting PL1 and PL2 on Intel, or PPT on AMD to 84 W (or TDP to 65 W which uses an 88 W PPT) should give you comparable results.

I also have a 7700X, but I can't test it now, as I'm about 1,500 miles away from home until the end of next week.

Good, thanks.

A 12900k with 2 P cores off scores 17988, with the full 8+8 configuration 20250.

13th gen should be 10% above that

but it seems to me R7 vs i7 are most efficient at different power levels (on desktop)

I haven't tested every possible power level, so I can't say yes, but in general -- I don't think it's' a close contest in any power level. For starters, with the i7 at 100w, it scores way way higher than what is possible for the R7 to achieve, even with ln2 cooling. So for anything above 90-100w, the i7 demolishes the R7. Below that range, I don't see the R7 doing much honestly


For example, my 12900k @ 125w scores 25k to 26k (that's with undervolting) and 24k at complete stock. No R7 on planet earth, no matter how much power you push through it can get that score.

 

[AusWolf]

Good, thanks.

A 12900k with 2 P cores off scores 17988, with the full 8+8 configuration 20250.

13th gen should be 10% above that


I haven't tested every possible power level, so I can't say yes, but in general -- I don't think it's' a close contest in any power level. For starters, with the i7 at 100w, it scores way way higher than what is possible for the R7 to achieve, even with ln2 cooling. So for anything above 90-100w, the i7 demolishes the R7. Below that range, I don't see the R7 doing much honestly


For example, my 12900k @ 125w scores 25k to 26k (that's with undervolting) and 24k at complete stock. No R7 on planet earth, no matter how much power you push through it can get that score.

Perhaps it is more of a case of the i7 scaling linearly with power consumption, whereas the R7 uses most of its higher power headroom for more voltage after a certain point, so it doesn't gain much - therefore it's more efficient at lower PPT, and therefore there's minimal difference between the performance of the 7700X and 7700.

I remember while I had the 7700X installed, I saw barely any difference at lower power limits.

 

[JustBenching]

Raptor Lake Refresh CPU core configurations

View attachment 304497

If those are true. WOW, an 8+8 i5 beating the i9 from 2 years ago. Thats basically like the 7600x beating the 5950x in multithreading.

Perhaps it is more of a case of the i7 scaling linearly with power consumption, whereas the R7 uses most of its higher power headroom for more voltage after a certain point, so it doesn't gain much - therefore it's more efficient at lower PPT, and therefore there's minimal difference between the performance of the 7700X and 7700.

I remember while I had the 7700X installed, I saw barely any difference at lower power limits.

Yeah, scaling on the 7700 stops way sooner because well, it has a lower core count than the i7. Basically it hits diminishing returns at I would have to guess 50-60w while the i7 probably does at 90+..

But still, like for example, 12900k @ 35w = 15200 cbr23 score. Don't think a 7700x can match that.. in fact I don't think any CPU besides 13th gen Intel can match that.

 

[A Computer Guy]

Good, thanks.

A 12900k with 2 P cores off scores 17988, with the full 8+8 configuration 20250.

13th gen should be 10% above that


I haven't tested every possible power level, so I can't say yes, but in general -- I don't think it's' a close contest in any power level. For starters, with the i7 at 100w, it scores way way higher than what is possible for the R7 to achieve, even with ln2 cooling. So for anything above 90-100w, the i7 demolishes the R7. Below that range, I don't see the R7 doing much honestly


For example, my 12900k @ 125w scores 25k to 26k (that's with undervolting) and 24k at complete stock. No R7 on planet earth, no matter how much power you push through it can get that score.

Isn't 2900k an i9? Wouldn't you compare that against R9 5950x?

 

[JustBenching]

Isn't 2900k an i9? Wouldn't you compare that against R9 5950x?

Yes, the 12900k is an i9, but it's basically the same configuration as the i7 13700k. 8p+8e cores. In fact the 13700k is both faster and more efficient.