Intel Announces Atom x6000E Series "Elkhart Lake", 11th Gen Core Tigerlake-UP3

[skeevy420]

A Zen 2 chip will boost as high as it can so long as:

1. The CPU silicon isn't too hot.
2. The motherboard VRM is willing to supply more power.

Now the maximum speed any 3950X core will go with factory settings is 4.7 GHz. In single threaded workloads, conditions #1 and #2 above are almost never violated. One maxed out core won't generate that much heat and the motherboard's VRM certainly won't have issues powering it. But now let's add more worker threads. The silicon will heat up and the VRM will be put under heavier load.

In my testing, I have gotten my 3950X to hit 4.3 GHz across all cores. This is with a 400 amp VRM and a cooler with six heat pipes and a fan that puts through 171 CFM.

So should AMD advertise the 3950X as having a max all-core boost of 4.3 GHz, then? What if I put active cooling on my VRM and switch to liquid cooling for the processor itself, thereby letting me reach an even higher all-core boost? Or perhaps more interestingly, what about someone who tries to do this with a 200 amp VRM and a cooler meant for the much weaker 3600 CPU?

How should max boost clocks be advertised?

Meant to reply to you a couple days ago...got distracted....

Preferably they should advertise the max clock as what the CPU can handle under a sustained load on all cores with their stock/factory air cooler. If it can only handle 4.3Ghz across all cores all the time with that setup, then that should be the advertised speed. If only two cores out of 8 are capable of going higher than 4.3Ghz and can only do that for a short time before being throttled, it's extremely disingenuous to call those "enhanced cores" the max speed of the CPU.

I know that they shouldn't be advertising "overclock before thermal throttle speed that only 1/4 of the CPU can even reach" as the max speed of the CPU. That's like advertising the 0-100 speed of a car and ignoring that it had Nitrous Oxide and was driving down Pikes Peak.

[CochainComplex]

If you are old enough to remember the Hi-Fi power rating "wars" of the 1970s, then I'll say no more about the fallacies of how "max boost clocks" are determined and why they are advertised.

Ultimately, if I remember correctly, it was a government regulatory agency in the USA that stepped in and forced standardized testing for various performance values in Hi-Fi products that are sold in the USA. Other countries around the world might have also joined together to establish and-or enforce product testing standards.

TL;DR Any test can be devised to prove any point you want to make. It's like lying with statistics; pick & choose the numbers you like.

Well I'm not old enough but I remember my friends in school (late 90ies/early00s) bragging about their soundsystem in Watt units.
After I had looked up what it means I have never really cared about it again.
(Sure this is depending on your speakers, but for mine everything 50-100W+ in 8Ohm is fine)
I don't care even if I have build the Amp myself.
Not so experienced people in this field always ask me first how much Watt? If I tell them I don't know or can't really remember 50 maybe 80 @8Ohm?
they are a bit baffled and reply with....but you have build them right?
And then I have to explain why it is not as important as people think it might be...

[jaxa]

Is nobody going to mention the Cortex-M7 core?

https://www.anandtech.com/show/16102...-now-available

There is now a new Programmable Services Engine to offload IoT workloads. This is a dedicated ARM processor, specifically an Arm Cortex M7, that supports real-time functionality, network synchronization, time sensitive networking, and low compute requirement workloads without needing to fire up the bigger cores. Some of the models support Time Coordinated Computing to enable worst-case execution time (WCET) and ultra-reliable low-latency communication (URLLC).

[MadeUpName]

I am wondering what the Xe graphics will brings. Core counts are nice but really don't tell us any thing useful.

[[email protected]]

In fairness, one of the early design constraints for Zen was the ability to scale efficiently enough to cover all of our current-at-the-time product ranges, so other than waiting to make sure we had met those design goals everything else was more-or-less planned.

One of the market realities was that in most segments the value of high single-thread performance never went away, even if that is the least efficient way to approach CPU design. Back in the Bulldozer days the conventional wisdom was that low end products should be designed for less power consumption and less peak IPC, but in fact what ended up mattering was high and efficient single-core performance plus aggressive power-gating for unused cores.

You can see the impact of this self-destructive market reality in ARM products as well, with a much larger focus on wide & deep microarchitectures to extract high peak IPC at the expense of power efficiency.

To be honest I do prefer the way thing are now, since you get a CPU that can be better utilised outside the targeted market. If AMD can sell a Zen based CPU for the price of a Atom, who cares if it is not a specialised architecture for embed systems? If it meets the power target, great.

I do have a Jaguar based CPU on AM1 socket, and was wondering what AMD sells now for that market. Those dual core Athlon 3000G can fly rings around my old quad core Jaguar, while consuming about the same power. And if I desire to upgrade for a more powerful CPU, there is the entire AM4 platform to choose from.

[discordian]

So, coming back to the topic at hand, does anybody know if those new Atoms are from a completely different family architecture, or are just bottom of the barrel Core chips rebranded as Atoms?

Own uarch, somewhat close to early Core chips AKAIK.

Elkhart Lake is supposed to be the first design that takles needs of Industrial/realtime from ground-up. You dont see this much in media, but their "Apollo Lake" for industrial use (x5-E3930) was announced to be available in 2016, and was finally available Q1-Q2 this year.

I ask because AMD choose to abandon having different architectures (like Jaguar's on the AM1socket), and just offer defective chips at lower prices, instead of having the trouble to make something knew and optimised for the low end.

Zen as architecture might be scalable, but using the exact same hardware for low and high-end is likely not a good idea. ARM (and now Intel) dont do big-/little cores on a chip for fun and giggles.
5 Year old ARMs still walk over current Intel Atoms.

[[email protected]]

If you are old enough to remember the Hi-Fi power rating "wars" of the 1970s, then I'll say no more about the fallacies of how "max boost clocks" are determined and why they are advertised.

Ultimately, if I remember correctly, it was a government regulatory agency in the USA that stepped in and forced standardized testing for various performance values in Hi-Fi products that are sold in the USA. Other countries around the world might have also joined together to establish and-or enforce product testing standards.

TL;DR Any test can be devised to prove any point you want to make. It's like lying with statistics; pick & choose the numbers you like.

Huehue, here Brazil, we got that bullshit till the nineties. I remember seen those huge PMPO numbers made to impress the buyers, and the small print of RMS numbers for the reality check.

[NotMine999]

maybe some average metric. Sure that opens a whole new can of worms. Fuel consumption measurements for cars have some standarized testing procedure but there is a lot of controversy about it too.

I got the feeling that the "up to" gets more and more unrealistic to achieve. E.g in the past. my FX 8350 hit the boost clockspeed longer than a few secs and under realistic loads.
As described above my laptop has one of this xeons. And it was really difficult to see 5ghz. You have to have some workload which is more then idle but not overloading it.
It would be more reasonable to say that this thing runs with 4.8ghz maybe 4.7ghz in this scenario.
Because this is what you mostly see when you push it with classic single threaded jobs ...eg. counting loop over an array with some arithmetics (no avx).

Bottom line it is difficult to get an appropriate metric. But it should not be driven to the point where we get only 6ghz for 250 ms after we have pushed the power button because then the die is still cold enough. ...none the less "up to 6Ghz".

If you are old enough to remember the Hi-Fi power rating "wars" of the 1970s, then I'll say no more about the fallacies of how "max boost clocks" are determined and why they are advertised.

Ultimately, if I remember correctly, it was a government regulatory agency in the USA that stepped in and forced standardized testing for various performance values in Hi-Fi products that are sold in the USA. Other countries around the world might have also joined together to establish and-or enforce product testing standards.

TL;DR Any test can be devised to prove any point you want to make. It's like lying with statistics; pick & choose the numbers you like.

[CochainComplex]

How should max boost clocks be advertised?

maybe some average metric. Sure that opens a whole new can of worms. Fuel consumption measurements for cars have some standarized testing procedure but there is a lot of controversy about it too.

I got the feeling that the "up to" gets more and more unrealistic to achieve. E.g in the past. my FX 8350 hit the boost clockspeed longer than a few secs and under realistic loads.
As described above my laptop has one of this xeons. And it was really difficult to see 5ghz. You have to have some workload which is more then idle but not overloading it.
It would be more reasonable to say that this thing runs with 4.8ghz maybe 4.7ghz in this scenario.
Because this is what you mostly see when you push it with classic single threaded jobs ...eg. counting loop over an array with some arithmetics (no avx).

Bottom line it is difficult to get an appropriate metric. But it should not be driven to the point where we get only 6ghz for 250 ms after we have pushed the power button because then the die is still cold enough. ...none the less "up to 6Ghz".

[Sonadow]

Some of these new Atoms are quite power hungry. 12W? Really?

That said, a lot of small Chinese laptop OEMs are churning out Apollo Lake and Gemini Lake laptops with backlit keyboards, upgradable M.2 SSDs and metal chassis (but soldered memory) at crazy low prices. I already own three such laptops and will love to see these chips find their way to these OEMs so that I can get a few more without breaking the piggybank.