Intel launches Rocket Lake; how much heat and power are okay?

Quizzical

Intel is launching its Rocket Lake processors today.  I've previously explained how it's a backport of 10 nm architectures to a 14 nm process node.

You can find reviews on a variety of sites.  The upshot is that it manages to roughly match AMD's Ryzen 5000 series in per-core CPU performance.  However, it tops out at 8 cores, while the Ryzen 5000 series goes up to 16, and even Intel's previous generation Comet Lake went up to 10.

But the big caveat is power consumption.  You might argue that you typically don't care about CPU power consumption, at least in a desktop.  And if one CPU tops out at 105 W, while another goes up to 125 W, it's reasonable to not care about that.  But when power and heat get far enough out of hand, maybe you should care.  What is out of hand?  Well, try this chart:



Ultimately, the only real reasons to buy Rocket Lake are if you aren't bothered by runaway power consumption and at least one of:
1)  you need AVX-512 support,
2)  you have a strong brand preference for Intel, or
3)  you can find Intel Rocket Lake CPUs in stock near MSRP, but not the competing AMD CPUs in the same price range.

Point (3) has been salient lately, and it's not clear what Rocket Lake inventory will look like.  Unfortunately for those who need AVX-512, that's precisely what causes power consumption to go nuts, sometimes pushing temperatures over 100 C or even crashing the machine.

The other problem on price is that while a Core i7-11700K is cheaper than a Ryzen 7 5800X and a Core i5-11500K is cheaper than a Ryzen 5 5600X (at least if you assume that all are at MSRP), once you get a better cooler to handle the higher power consumption, will the Intel CPU still be cheaper?  It might not.  The Ryzen CPUs will never go over 142 W for any thermally significant amount of time, and it's easy to handle that on air.  The Rocket Lake CPUs can stay over 200 W for extended amounts of time, and that takes either water or a high end air cooler.

olepi

From Arstechnica:

The Bad

• Lower multithreaded performance than 10th-gen
• Massively lower multithreaded performance than Ryzen
• Massively higher power consumption than Ryzen

The Ugly

• Rocket Lake-S seems more like gen-on-gen backsliding than improvement
• AMD processors still offer much higher performance and lower power draw, for roughly the same price

laserit

Well up here in the icy cold of Canada sounds like the 11700k might just be the ticket. I might just cancel my order for that new wood stove and use an 11700k instead.

And by the way Jean_Luc. I hug my trees every day

Quizzical

laserit said:

Well up here in the icy cold of Canada sounds like the 11700k might just be the ticket. I might just cancel my order for that new wood stove and use an 11700k instead.

And by the way Jean_Luc. I hug my trees every day

So you hug trees and then light them on fire?  Or is it the other way around?

Vrika

Quizzical said:

laserit said:

Well up here in the icy cold of Canada sounds like the 11700k might just be the ticket. I might just cancel my order for that new wood stove and use an 11700k instead.

And by the way Jean_Luc. I hug my trees every day

So you hug trees and then light them on fire?  Or is it the other way around?

He's bored at hugging trees and plans to buy equipment for setting them on fire?

olepi

I will say though that I am impressed in one way. Intel managed to get really high performance out of their 14nm process, if you don't mind infinite power and infinite heat.

If Intel was able to use this new architecture with a really good process, like 7nm or 5nm, it would be truly impressive.

olepi

Jean-Luc_Picard said:

olepi said:

I will say though that I am impressed in one way. Intel managed to get really high performance out of their 14nm process, if you don't mind infinite power and infinite heat.

If Intel was able to use this new architecture with a really good process, like 7nm or 5nm, it would be truly impressive.

I'm still not sure about those heat problems, at least on the 10 series.

My 10700k is on permanent turbo speed, and survives any stress test without overheating and passing 80°C.

Granted, I have watercooling, but not some elite pro stuff, just a basic closed circuit model.

Heat is really bad for chip reliability. As the metal gets hotter, it gets more pliable and tends to move around. It's called "electro-migration", and it can cause the chip to fail and be dead forever.

Based on Quizzical's first post, it looks like the top of the line Intel part is using almost exactly twice the power of the top Ryzen AMD part. This means more heat. It helps a little probably that 14nm wires are larger than 7nm or 5nm wires.

But aside from the packaging problems and fans/coolers problems, burning that much power is risky. It would be interesting to see Intel's reliability numbers for the new part.

Asm0deus

olepi said:

Jean-Luc_Picard said:

olepi said:

I will say though that I am impressed in one way. Intel managed to get really high performance out of their 14nm process, if you don't mind infinite power and infinite heat.

If Intel was able to use this new architecture with a really good process, like 7nm or 5nm, it would be truly impressive.

I'm still not sure about those heat problems, at least on the 10 series.

My 10700k is on permanent turbo speed, and survives any stress test without overheating and passing 80°C.

Granted, I have watercooling, but not some elite pro stuff, just a basic closed circuit model.

Heat is really bad for chip reliability. As the metal gets hotter, it gets more pliable and tends to move around. It's called "electro-migration", and it can cause the chip to fail and be dead forever.

Based on Quizzical's first post, it looks like the top of the line Intel part is using almost exactly twice the power of the top Ryzen AMD part. This means more heat. It helps a little probably that 14nm wires are larger than 7nm or 5nm wires.

But aside from the packaging problems and fans/coolers problems, burning that much power is risky. It would be interesting to see Intel's reliability numbers for the new part.

While you may have a point you got the term wrong, the movement of metal due to heat or cold is not "electro-migration".  This is thermal expansion.

Electromigration, which can cause voids and failures in a device, refers to the displacement of the atoms as a result of current flowing through a conductor. To suppress electromigration in the interconnect part of the equation, chipmakers typically use a capping or etch stop layer of material on a dual-damascene structure.

Quizzical

The Core i9-11900K hitting 296 W is only under an all-core AVX-512 load.  You could do that with an HPC workload, though if you think you want to do that, you're likely to be better off using a GPU instead.  But it would be a very strange thing to do for a consumer workload.

Under an all-core AVX2 workload, the same chip only hit about 241 W.  That, of course, is still quite a lot, but it's still atypical for consumer use.  Intel's nominal 250 W PL2 power, however, indicates that a 241 W workload might not be nearly so atypical as the 296 W one.

They were hardly the only site to find runaway power consumption, though.  Here's one from TechPowerUp:




That 419 W is for total system power, not just the CPU alone.  The Ryzen 7 5800X and higher AMD CPUs on that list are probably using just shy of 142 W each, but the rest of the system uses some power, too.  Even so, 419 W for the whole system is still way out of hand, and it comes when using Cinebench, which hasn't traditionally been regarded as a power virus.

Even if a power virus causes atypical power consumption, if that fries your hardware, is it really so comforting that the runaway power consumption didn't last for very long?  Processors are supposed to have something in place to prevent things from getting way out of hand like that, precisely to prevent a power virus from wrecking hardware.  Remember how the StarCraft 2 title screen fried some Nvidia GPUs?  A "power virus" can show up accidentally in weird ways.  It sure looks like the new Rocket Lake CPUs don't have robust protection against that.

This is the sort of thing that it seems likely that Intel will "fix" with a BIOS update or some other sort of update.  Of course, the way to reduce power consumption is to reduce clock speeds (and voltages, but that also requires reducing clock speeds), and hence performance.  This would hardly be the first time a hardware vendor tried to show power consumption and performance at wildly different settings while trying to falsely imply that you could have both at once.

laserit

Quizzical said:

laserit said:

Well up here in the icy cold of Canada sounds like the 11700k might just be the ticket. I might just cancel my order for that new wood stove and use an 11700k instead.

And by the way Jean_Luc. I hug my trees every day

So you hug trees and then light them on fire?  Or is it the other way around?

All in jest

I love my Cedars and other evergreens but you can't light up them Alders quick enough, they grow like weeds

I have 10 acres with about 3-4 acres of forest. I have 2 acres cleared around my home, then a nice trail through the forest and onto about a 4 acre pasture.

laserit

Jean-Luc_Picard said:

laserit said:

And by the way Jean_Luc. I hug my trees every day

Don't get me wrong. I love nature, forests, I'm a photographer too.

But I don't believe the cars or the computers are the main source of pollution. I'd rather look at the huge boats first, for instance.

I was just rubbing ya and yes I agree. Loss of and destruction habitat is the main ecological issue we face  imho.

laserit

Vrika said:

Quizzical said:

laserit said:

Well up here in the icy cold of Canada sounds like the 11700k might just be the ticket. I might just cancel my order for that new wood stove and use an 11700k instead.

And by the way Jean_Luc. I hug my trees every day

So you hug trees and then light them on fire?  Or is it the other way around?

He's bored at hugging trees and plans to buy equipment for setting them on fire?

I bet that 11700k will light them up

olepi

Asm0deus said:

olepi said:

Jean-Luc_Picard said:

olepi said:

I will say though that I am impressed in one way. Intel managed to get really high performance out of their 14nm process, if you don't mind infinite power and infinite heat.

If Intel was able to use this new architecture with a really good process, like 7nm or 5nm, it would be truly impressive.

I'm still not sure about those heat problems, at least on the 10 series.

My 10700k is on permanent turbo speed, and survives any stress test without overheating and passing 80°C.

Granted, I have watercooling, but not some elite pro stuff, just a basic closed circuit model.

Heat is really bad for chip reliability. As the metal gets hotter, it gets more pliable and tends to move around. It's called "electro-migration", and it can cause the chip to fail and be dead forever.

Based on Quizzical's first post, it looks like the top of the line Intel part is using almost exactly twice the power of the top Ryzen AMD part. This means more heat. It helps a little probably that 14nm wires are larger than 7nm or 5nm wires.

But aside from the packaging problems and fans/coolers problems, burning that much power is risky. It would be interesting to see Intel's reliability numbers for the new part.

While you may have a point you got the term wrong, the movement of metal due to heat or cold is not "electro-migration".  This is thermal expansion.

Electromigration, which can cause voids and failures in a device, refers to the displacement of the atoms as a result of current flowing through a conductor. To suppress electromigration in the interconnect part of the equation, chipmakers typically use a capping or etch stop layer of material on a dual-damascene structure.

What I was pointing out was the electro-migration (EM) of metal gets much worse if the metal is hot. It all goes together. Higher current gets you more performance, but also more heat. And heat is deadly for EM.

Sandmanjw

It is just so strange how time has shifted the chips being put out by intel and amd.

But even worse in a way....at least the old amd had people that liked to OC buying them.

The reversal of fortunes here is just mind boggling to me. 

Quizzical

Sandmanjw said:

It is just so strange how time has shifted the chips being put out by intel and amd.

But even worse in a way....at least the old amd had people that liked to OC buying them.

The reversal of fortunes here is just mind boggling to me. 

One way to think of it is that AMD's Ryzen 2000 series and later CPUs come with a really good factory overclock by default.  Intel CPUs are also clocked a lot closer to the chip's limits than they used to be.

If you go back far enough, all consumer CPUs were single core, and the clock speed was the only way that they had to distinguish different bins of the same architecture.  Today, no matter how far you overclock a six-core CPU, you can't turn it into an eight-core one.

Asm0deus

olepi said:

Asm0deus said:

olepi said:

Jean-Luc_Picard said:

olepi said:

I will say though that I am impressed in one way. Intel managed to get really high performance out of their 14nm process, if you don't mind infinite power and infinite heat.

If Intel was able to use this new architecture with a really good process, like 7nm or 5nm, it would be truly impressive.

I'm still not sure about those heat problems, at least on the 10 series.

My 10700k is on permanent turbo speed, and survives any stress test without overheating and passing 80°C.

Granted, I have watercooling, but not some elite pro stuff, just a basic closed circuit model.

Heat is really bad for chip reliability. As the metal gets hotter, it gets more pliable and tends to move around. It's called "electro-migration", and it can cause the chip to fail and be dead forever.

Based on Quizzical's first post, it looks like the top of the line Intel part is using almost exactly twice the power of the top Ryzen AMD part. This means more heat. It helps a little probably that 14nm wires are larger than 7nm or 5nm wires.

But aside from the packaging problems and fans/coolers problems, burning that much power is risky. It would be interesting to see Intel's reliability numbers for the new part.

While you may have a point you got the term wrong, the movement of metal due to heat or cold is not "electro-migration".  This is thermal expansion.

Electromigration, which can cause voids and failures in a device, refers to the displacement of the atoms as a result of current flowing through a conductor. To suppress electromigration in the interconnect part of the equation, chipmakers typically use a capping or etch stop layer of material on a dual-damascene structure.

What I was pointing out was the electro-migration (EM) of metal gets much worse if the metal is hot. It all goes together. Higher current gets you more performance, but also more heat. And heat is deadly for EM.

Yes thermal stress migration interacts/contributes with electromigration but they are not the same thing.

I was just pointing this out so people do not read this post and think they have learned something new but then find out it's not what they thought it was.

Quizzical

But what about the integrated GPU?  This was supposed to mean the first time bringing Intel's new Xe integrated GPU to a mainstream desktop platform where it can be evaluated fairly.  So how does it perform?

https://videocardz.com/newz/intel-awkwardly-forgot-to-release-graphics-drivers-for-rocket-lake

Apparently it doesn't.  Intel hasn't yet released a public driver for it, and that's apparently still a few weeks away.  And you thought MMORPGs had rough launch days.