Intel counters AMD’s big-cache PC chip with 5.5GHz 16-core rival The CPU horse race continues between Intel and AMD, this time with the impending availability of Intel's Core i9-12900KS, which is said to be able to hit 5.5GHz on two cores. Intel announced on Monday it will expand its 12th-generation line of PC-grade microprocessors with the "special edition" 16-core chip, which the US giant …
So it'll run at 5.5GHz, on only 1 or 2 cores at a time, until it gets too hot. I'd be more impressed if it was 16 cores all of the time.
Getting anything running at that speed, even for a short time, is fairly impressive, especially if the memory subsystem is actually up to keeping cores fed when they're running at that pace. But it does feel a bit like a headline grabber, rather than something that truly sustains performance. The proof will be in the pudding.
I just read the article over at ArsTechnica about this very CPU & came away with a "meh" impression.
It uses a hell of a lot of power to acomplish the feat, will probably need a fairly good liquid cooler system to keep it stable, and will almost certainly cause your power bills to go through the roof, all for the bragging rights of a 5.5GHz CPU.
Or you can use AMD's top CPU & run slower, cooler, & more efficiently on *all* cores instead of just two super speedy fast ones.
Intel at 14nm or AMD at 7nm, I wonder which I'd choose?
Yes, it was fine when the node measurement was just something that was used internally at the fabs, but tech bloggers with not quite enough knowledge have unfortunately turned it into a Holy Number. No surprise, as we’ve been here before several times with “numbers”, most recently with the confusion about high versus low TDP, cores versus threads, 64-bit versus 32-bit, and of course, the granddaddy of them all: Megahertz.
Alder Lake is built on what, until very recently, Intel called its “10 nm Enhanced Super-Fin” process. Despite having that “10 nm” in its name, this is actually able to pack more gates onto a die (100 Million transistors per square mm) than the slightly older TSMC “7nm” process (92 Million transistors per square millimetre) that AMD is using right now.
However, TSMC has a newer “5nm” process (albeit only for ARM SoCs) that packs in around 165 Million transistors per square millimetre, and Intel has no answer to that until its “Intel 4” process starts up, allegedly later this year, with an expected density of around 180-200 Million transistors per square millimetre. In other words, when it comes to wafer fabrication, whoever has the newest thing has the best thing.
But AMD is expected to jump to that TSMC 5mm process later this year, and so will take back the process lead; Intel’s first “Intel 4” CPUs, meanwhile, will be “Meteor Lake” in mid-2023.
I know it's 2022 and I know about Mooore's Law etc but this is just unimaginable:
"100 Million transistors per square mm"
"92 Million transistors per square millimetre"
"165 Million transistors per square millimetre"
So millions of transistors in a space no bigger than this > []
We've come a long way since I got my first 80486DX33 around winter 1990....which had a transistor count of: 1.2–1.6 million for the entire CPU
Yeah, it’s crazy isn’t it. The Motorola 68000, the first CPU I learned to program in assembly-language, was so-named because that was the number of transistors on the die!
…and the process node for the first 68000 chips was 3.5 μm, meaning that each of those transistors was around 500 times larger than those of a modern CPU.
Hmm. Announcing the all-new Motorola 68G. Never knew that was how the 68K got it's name. I do remember going to buy multi-packs of individually canned transistors from Tandys though*. And a friend who recreated a Z80 from discrete components. With modern densities, I guess a model would need real lanes to drive around it. If each transistor was in a 3-4mm can, it'd need a lot of land.
Also got me thinking about power & RF management, ie more power means more heat and physically larger transistors. But other than ending up with a CPU visible from space, I think I'll leave that project to someone else.
*Always remembering to pick up my free batteries.
The densities sound impressive and the technology that achieves it is impressive. But once you start thinking in the world of atoms then you realise that potentially there's a long way to go, which is why work on optical computers – bugger the size because switching is so much faster than using electrons – or molecular ones – a data centre the size of a sugar cube is ongoing, because we're reaching the physical limits of electronics, especially the ones like transistors dependent upon or susceptible to quantum effects.
My first hard drive was a Seagate ST125-1 20 Mb, in a 5.25" form-factor, using MFM as the transfer interface...
A year later I got a Seagate ST251 40 Mb HDD to partner the 16 Mb of RAM I had (at £100 per 4Mb SIMM) !!
Everything worked great under DOS...and then Windows 3.11 came along and sapped all the computing power and I needed an 80 Mb drive, so I switched to a SCSI interface with an Adaptec AHA2942 add-in card....and I could then install more HDDs !! (as the MFM interface was limited to 2 devices).
Nowadays, a 20 Mb file is nothing and I'm running a Gigabit Ethernet connection that can transfer 20 Mb in a blink of an eye !
My first hard drive was a Seagate ST125-1 20 Mb, in a 5.25" form-factor, using MFM as the transfer interface...
No it wasn't. ST-125 is a 3.5" drive. If it was 5.25" MFM it was ST-225. The same drive was sold as ST-238R to be used with RLL controller. Many people chanced it by using RLL controller with the ST-225 to get extra 50% storage.
MFM is actually the low level format (just like RLL) and the interface was ST-506.
I thought we had got passed the pure GHz numbers as a measure of CPU performance years ago? If it can't sustain running in 5.5GHz mode for a long period of time I can't see how that is good for gaming, as if it were to suddenly throttle mid game because the CPU starts to melt through the motherboard then the game will start to slow down as well.
If they can do it without have the power consumption of a small star and the heat output to match I'll be impressed.
The days of raw mips and GHz wars are well in the past. The frontier is getting enough performance to do the majority of everyday tasks with enough battery life for a day and without the risk of 3rd degree burns if the machine touches a piece of naked flesh
It was buried at the very end of the article because it's not complimentary to Intel, but the TDP of this part is 150 W. So despite half the cores being the all-but-useless "E" variety, it's going to dissipate 45 W more than a 5950X when flat out. TDP does not measure power dissipation at a given fixed performance level but it does provide a reasonable estimate of power dissipation when a processor is working flat out. To be 43% worse than your competitor despite crippling half your cores is shameful.
For an even more disgusting comparison, AMD's previous-generation 7452 (a midrange data centre processor) was released 3 years ago and has 32 (identical) cores with 146 MB of cache, along with 128 lanes of PCIe4, 8 memory channels, and all the logic that's in the external "chipset" on a desktop CPU, with a TDP of 155 W.
Another way to think about it is that the extra 45 W could power an additional R5700GE to provide those 8 "E" cores plus graphics and you'd still have 10 W left in your power budget. In other words, in the same power envelope, Intel are 8 performance cores and 10 W behind AMD, even assuming their cores could deliver the same performance (narrator: they can't). Ouch.
Despite all that, the reference to battery life seems misplaced here. The processor Intel announced is for desktop use, not mobile, as are the AMD 5800X3D and 5950X. While there are laptops made with desktop processors, they are rare and are assumed to be used at a power point most of the time. Even those invariably use nothing bigger than a 65 W APU. For comparison, processors intended for laptops typically have TDPs of 20-35 W, some even lower; the 45-50 W range are marketed as "desktop replacement" or "mobile workstation" parts, though of course they are still far less capable than even a bottom-end desktop CPU of the same microarchitecture and process. Suffice it to say, if you're in the market for anything being discussed here, it's not expected that it'll be running on a battery or touching your skin.
"Despite all that, the reference to battery life seems misplaced here. "
Valid criticism, however it feels like the days of large desktop is waning. The last two years have shown the need for mobility and flexibility, something that a desktop cannot provide, so instead we are looking for portable computers that sometimes act like desktops. There are niche applications that require the full power of a processor, but there are not enough of them to feed a monster like Intel
Of course the elephant in the room is Apple and their home grown M1 chip. Apple have shown that you can create a chip which has far less power requirement than Intel's best, while at the same time increasing performance. The idea that companies can roll their own, is probably giving Intel's CEO sleepless nights, because without Intel's virtual monopolly (or at last duopoly), Intel is virtually nothing
So we're essentially back at what intel did during the Pentium4 days, pushing up clock rate and power draw for no other benefits than getting the highest (frequency) numbers on a spec sheet.
It seems intel still has no path out of the quagmire they're in, and no real product to compete with Apple's M1.
So Intels latest chips beat AMDs last gen chips.
AMD still have AM4 after 6 years, Intel have had 3 different socket types for desktops in that time.
AMD still on DDR4, Intel probably using the fastest DDR5 that can be bought.
AMD still on PCIe4, Intel using PCIe 5.
AMD at EoL of Zen 3, Intel with latest Alder Lake architecture, not up to date could be sapphire rapids or raptor lake
Intel beats AMD, I would bloody well hope they would with all those advantages
Sockets lasting multiple generations is generally a good thing for consumers, because upgrade path.
DDR4/5 real world differences are negligible apart from cost.
PCIe4/5 hardly any cards can take advantage of the latter yet, so again, cost wise why upgrade?
Intel don’t hold socket designs stable, so having DDR5 & PCIe5 on board won’t save you a motherboard replace.
So your criteria for Intel beating AMD are largely my own criteria for AMD beating Intel in terms of consumer offering and especially value.
There’s no denying a 12900k is probably the best gaming CPU around right now. However, if you already have say, a 3800xt there is very, very little point upgrading. If you are into parallel workflows like video encoding etc I’d still take the 5950x over the 12900k too.
Yep, and probably in due course Intel will do something like the 3D cache thing.
It's great when the duopoly is working as it should - competition forces innovation. And quite often AMD seems to be Intel's "lab", innovating on architecture whereas Intel steams ahead with the fabs.
It was a shame when AMD went down an architectural blind alley on the "digger" architecture family, just as much as when Intel got lost in RSI-inducing iterations on 14nm (add +s to taste).
Here's hoping the two companies stay neck-and-neck for the foreseeable future.
PCI-E 4 Vs 5 isn't worth considering... there are only a couple of GPU's that can even saturate the full 16x bandwidth of PCI-E 3.0 and it's going to be a long time before GPU's fully saturate gen 4 bandwidth.
This is all about data transfer speeds for things like NVME drives... and given that my system can load in about 8 seconds from post to login on a sabrent rocket PCI-E4 NVME, whilst my mediaserver takes about 12 seconds on a WD Black PCI-E 3.0 NVME drive... the difference is negligible for the average user.
But higher numbers always mean better and faster when it comes to benchmarks, specifications and advertising blurb.
That most people will never actually use it is irrelevant and in the case of this chip I am really struggling to see what the point is.
So whoopie-doo, you can run 1 or 2 cores out of 16 at 5.5GHz and if yo are luck not fry everything. I feel the use case is somewhat limited.
Now from an HPC perspective, running 16 cores at 5.5GHz without melting everything because every cycle matters is a different thing. Even then there is a sweet spot between cost/cores/speed/power that these sort of bleeding edge chips are pretty much clear of.
Is Intel about to post a killer blow to AMD?
I doubt it.
I could very well be convinced they're releasing a chip which puts them in the ballpark, a little better at some, slightly slower at others, but competitive.
But Zen-4 and AM5 are coming, and when that happens, Intel are going to find out what it feels like to be Chris Rock.
The honest truth well demonstrated back in the AMD64 days is that when AMD releases a product that smashes the hell out of the Intel alternative Intel continues to lock AMD out of the market using their relationships with the large vendors until they could eventually produce something as competitive.
When they had built up a performance lead then their prices go up to recoup the lost profits, and R&D goes out of the window until AMD manages to either catch up or leapfrog the Intel offering, at which point the price of Intel gear may come down under the stratosphere again.
What interests me more, given the current climate and cost of energy... Is what is the fastest CPU, whilst also being the best value for money, whilst also being the most efficient.
We all know that 12th gen is a power hungry monster to get those speeds and performance, where AMD is doing very similar at a lower wattage.
Price is no longer AMD's biggest advantage... but intel allowing DDR4 & DDR5 allows for a better budget priced system... whilst AMD's AM5 platform is DDR5 only as far as I am aware at the moment.
So to peak my interest... I'd want to know which offers the best performance, when both running at identical speeds and identical power draw... if possible.
Mind you... not even looking to build a new system this year... 3800X already runs @ 4.4Ghz all core @~ just 1.25v, no need for faster ram because 32GB is enough and it's not worth going above 3800Mhz speeds due to splitting the IF from 1:1 down to 2:1 above that.
I'm still in the market for a GPU when prices drop to reasonable levels... maybe next year a 5900X upgrade to push the system out for a couple more years... Maybe a new build around 2025
Might even be an issue upgrading to a new GPU if they all switch to the new ATX3.0 connectors... and I only bought a new EVGA 850W PSU one a few months ago in preparation for a high end GPU
You mean 8 cores. 8 of them are so underpowered as to be useless in practice for a development box.
I need core count man. Full speed, equal processing power core counts. AMDs latest is no help to me because it has fewer cores than I do now; my next box will have more, hopefully double the count, if not better. It'll depend on the budget at the time...
"AMDs latest is no help to me because it has fewer cores than I do now"
AMD's latest CPUs, the recently-announced Milan-X, have 64 (identical) cores apiece, and there are many boards available that can put a single system image on two of them. What are you running today with more than 128 cores? An Ampere prototype? zSeries? An ancient CM-2, still perhaps the most badass-looking computer ever made, with 65,536 (1-bit) processors?
As you say, I suppose it depends on your budget. But if your budget is so small that you're talking about low-end stuff like this Alder Lake abortion or AMD's stacked-cache Vermeer, you can't be very serious about needing more cores even if you might want them. As the boys at the racetrack put it, performance costs money; how fast can you afford to go?
What horse race? This is a mediocre 8-core desktop part. It's hot, it's expensive, and it's slower than what AMD were selling 2 years ago. It's not an answer to gigantic stacked caches; it's barely an answer to Zen2. If this were a horse race, it would be a dull one indeed. Intel can't make a competitive processor and haven't been able to for 6 years now, but they can still issue press releases so they do so at every opportunity. They can't make a competitive process work either, so they renamed their obsolete processes using misleading numbers just like Cyrix did with clock rates many years ago. They can't compete with AMD on cache size or speed so they rebrand theirs "Smart Cache". None of these actions has made this a competitive situation. Everyone who knows or cares at all knows perfectly well that Intel's products aren't competitive with AMD's on any axis and are unlikely to be so any time soon no matter how successful the current regime's turnaround plan may someday be. If there's a horse race here, it's run every single day in data centres around the world, dollar for dollar, euro for euro, watt for watt, instruction for instruction. And every single day since Naples shipped, AMD's horse has crossed the finish line before Intel's left the gate. It's almost as embarrassing as it must be to keep writing about these processors as if they're genuinely competitive and spreading outright falsehoods like the notion that AMD somehow needs to catch up (to a horse that's lying down asleep in the stables). I guess it's a living. But that's the competitive landscape as it stands today.
Thing is, that's not a horse race, because a horse race ends. There's no way to predict when humans will stop buying and using computers but it seems likely to be a very long time from now. That's the only reason anyone cares about Intel at all (other than outlets like El Reg who are paid to promote their inferior products and do so with irritating shamelessness): it's just maybe barely possible that they can raise enough capital to someday catch up to the AMD-TSMC juggernaut. AMD's track record of squandering enormous technical leads makes this prospect a lot more plausible, and therefore a lot more interesting. But a horse race is over in minutes at most, and if Intel ever do catch up it will not be for many more years: their process technology is still 2 nodes behind TSMC's and the latter are not standing still. In that sense this is a lot more like a culture war: it's ebb and flow over a period of decades, with no likelihood there will ever be a "winner", only someone in the lead at the moment. And on that time horizon AMD and Intel both may well be irrelevant. For now, though, it's not competitive, this SKU changes nothing, and there's nothing much to see here.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
