Intel to Launch Core I9-9900KS Next Month: 5 GHz on All Cores | Hacker News Reader

Intel to Launch Core I9-9900KS Next Month: 5 GHz on All Cores

anandtech.com

44 pointsrbanffy79 comments

Does anybody else have a serious case of Deja-vu?

It feels like we're living in 2005 again. AMD has technical leadership, and Intel is bragging about clock rates.

back in 2005, Intel had Core 2 Due and Core 2 Quad already in its pipeline. I don't think Intel has anything comparable in its pipeline now.

I always saw Intel naming their product “Core” as a cynical ploy to confuse terminology by associating their product with multi-core computers. It is as if an automaker decided to brand their new motor with the name “Engine”.

Ha, I totally remember that confusion— my first laptop was the first gen white MacBook, with a Core Duo processor. But not a Core 2, or a Core 2 Duo, just a Core Duo. The other ones came along later.

I would be surprised if they haven't. i9-9900KS is still based on renamed Skylake core from 2016 scarred by Spectre mitigations at 14nm optimized to death.

My guess is that they were too optimistic with 10nm and no one started an alternative project on 14nm when the time was right.

They already claim +18% IPC on Ice Lake (because they basically dropped at least one generation and fixed security issues at least partially). When they resolve problems with the process (either by succeeding with 7nm or switching to TSMC) they will be very likely competetive again.

They claim that their 7nm process with EUV litography is on track to launch in 2021.

Oh boy I'd really like to see that, but considering the 10nm debacle I'm not too optimistic. If they actually manage to get EUV working well in 2021, we might never see widespread 10nm products at all.

Going by the roadmaps, that seems to be pretty much exactly their plan. Fabs that were slated for 10nm have been retrofitted for 14nm, or are rolling on to 7nm.

The 7nm and 10nm processes were developed by two independent teams. I'm pretty sure their 7nm team learned a lot of ways not to do things from their 10nm colleagues.

Whatever.

This is yet another pure marketing move to get their name on YouTube videos.

Almost nobody ( in their right mind ) would pay top dollar to have that mini nuclear reactor.

> Almost nobody ( in their right mind ) would pay top dollar to have that mini nuclear reactor.

Give it a few years after die-shrinks and efficiency improvements. A 5Ghz clock means a lot for solving problems that aren't parallelizable.

> Give it a few years after die-shrinks

There is no guarantee that after shrinking the transistor it will be possible to reach 5Ghz.

For example Intel has multiple problems with 10nm process and newer CPUs have lower clocks.

TSMC has already shrunken their die to 7nm. Intel can't keep up anymore. AMD wins this generation.

You can't (directly) compare semiconductor node sizes across manufacturers -- it no longer corresponds to any meaningful physical measurement and is just a made-up number from the marketing department. See https://en.wikipedia.org/wiki/10_nanometer:

> Since 2009, however, "node" has become a commercial name for marketing purposes that indicates new generations of process technologies, without any relation to gate length, metal pitch or gate pitch. For example, GlobalFoundries' 7 nm process is similar to Intel's 10 nm process, thus the conventional notion of a process node has become blurred. TSMC and Samsung's 10 nm processes are somewhere between Intel's 14 nm and 10 nm processes in transistor density.

Intel still beats on IPC. Their nodes are better-performing at the same process size, so they can be a little larger and still perform like a smaller AMD. That said, AMD is doing significantly better.

AMD is winning on IPC. Zen2 soundly passed up Intel on IPC.

Intel's higher clocks are its only remaining strength through which it can overcome the lower IPC just barely.

Hmm, my information is old. Thanks for the update. I was aware AMD had been better value for money for quite some time, but didn't know they had lost even IPC... big trouble for them. Makes me almost wonder if intel was purposely holding back 10nm until it really needed it. AMD beats on IPC, intel suddenly fixes their garbage?

> Makes me almost wonder if intel was purposely holding back 10nm until it really needed it.

Why would they do that? Intel didn't need to knock it out of the park, but they'd make more money doing yet another incremental improvement on 10nm just to get all the upgrade money. And to add in those laptop improvement features (wifi 6, integrated thunderbolt 3).

I'd understand them holding back large IPC improvements just to trickle it out, but holding back 10nm entirely doesn't make sense. Especially since it'd let them do even higher core count Xeon chips which make stupid huge profit.

> AMD beats on IPC, intel suddenly fixes their garbage?

Except AMD has beaten Intel on IPC and yet Intel has still not been able to fix their garbage. So much so they're doing yet another 14nm line of laptop SKUs (Comet Lake)

Why would intel do that they are going loose a lot market share in the server space the real money maker in the next couple of years large Intel buyers like Google have started looking at AMD already

Their advantage has largely been eaten away by their various performance harming security fixes.

We've gained more from IPC improvements (like 5-15% per generation) than clocks.

Intel's 10nm die shrink gives significantly lower clocks than their 14nm process though Ice Lake supposedly makes up for it with higher IPC.

I doubt Intel can pull off another die shrink after the 10nm disaster. It was supposed to come out three years ago IIRC and they have sunk incredibly amounts of money into the process node. The next node will take longer purely on Intel being more careful.

Processes aren't dependant on previous ones. For example AMD screwed up with 65nm, but 45nm were just fine (although under GloFo, but the work had started much earlier).

If the next process won't work their only option would be to switch to TSMC.

7nm process is already in preparation, with the launch scheduled to 2021, and Intel says they are on track, actually very happy with their progress. They worked on both processes in parallel.

Intel's 7nm is also smaller than TMSC's 7nm.

Die shrinks are hardly a thing anymore. The days of tick-tock are gone.

Winter is a comin'! Get the new Intel® Core™ i9 to warm up your whole house.

>In a bid to maintain an intrigue, Intel did not disclose TDP of its new product.

They basically released all the details but TDP. Very likely this sucker is a steaming hot mess. Wonder what the stock cooler is like, or if it's even possible to sustain these clock speeds on their stock cooler.

Intel has stopped delivering stock coolers with K-series CPUs for a while now. There is technically a sort of reference cooler but you'd have to buy that separately and nobody does that. Every single person in the market for a CPU like that was planning to use a "custom" cooling solution anyway. (Although custom in this case could just mean to put a big air cooler or AIO on it.)

Tbh the TDP doesn't really matter. It's a binned 9900K. We already know how those perform. It's literally just doing what siliconlottery does and getting some publicity for it.

> They basically released all the details but TDP. Very likely this sucker is a steaming hot mess.

Very likely it's just going to have the same "lie" for the TDP that they do for all their desktop chips. You can configure it to run at TDP specs, but ~nobody~ does. They instead run at PL2 specs (sustained power delivery rating for turbo), which for the 9900K for example is 210W despite the 95W TDP. Meaning per Intel's own specs the 9900K can run at 210W forever, as long as it has sufficient cooling.

The 9900KS isn't going to be a 5GHz base clock part. It's a 5GHz all-core turbo. The 9900K is, by comparison, 4.7GHz all-core turbo. So we're not talking much of a bump. If they leave the 9900KS at the same base frequency as the 9900K it'll have the same 95W TDP. Maybe they bump it by a 100mhz or something trivial and slap a 105W TDP rating on it. Doesn't really matter since it's a pretend number in made up fairy land anyway.

Great. Does it still come with an integrated hardware backdoor you cannot disable ?

I want open hardware that is not super expensive! Does it look like it is going to happen? To me it does not, but perhaps someone is more informed and up to date.

The single-socket POWER9 systems the Raptor folks sell aren't that much more expensive than a similarly performing Xeon box and are completely open.

Sadly I do not think that I will ever be able to afford it, just like I would not be able to afford a similarly performing Xeon box. :( My processor is from the Intel® Pentium® Processor G Series, a really cheap one.

I find them hard (sadly) to justify as developing on POWER brings me no palpable benefit over doing so on a low-end workstation-class Xeon box and their performance is way more than what I need. I find their work on the single-socket system commendable, but, if IBM wants people to have POWER9 desktops, they'll need to offer low-cost SKUs.

I really don't know why they don't do it - it's not like the CPUs are a significant part of the overall BOM of their p and i series machines and the profit margins are very thick. Their lowest of the low end box is already quite insane.

Nope, they completely removed it. Thank god that they had other similar systems laying around that regulated core functionality like power and booting! :D

Tin foil hat on? Edited: typo

If you duckduckgo 'intel processor backdoor' the first hit is:

https://www.techrepublic.com/article/is-the-intel-management...

So it doesn't seem like it is bullshit.

Another note: While yes, the foil of the hat is thin, the term is 'tin foil hat'.

Intel ME is not a conspiracy. It's a feature.

Is it tinfoil hat worthy when Intel has such a poor track record?

Any feature that sabotages the computer after 30 mins once disabled is not a feature. Intel me is a backdoor.

Just don’t look under the stage for the chiller

Tom's Hardware report that it maintained 5GHz on all cores on a standard 240mm AIO liquid cooler.

https://www.tomshardware.co.uk/intel-core-i9-9900ks-ships-oc...

It's like selling a car that can go to 140 MPH. It's nice but won't make your commute shorter.

Because my car can do 155mph? ;-)

> will feature a higher all-core Turbo frequency when it has sufficient cooling

Real world performance needs to be seen, but this feels like a stunt to grab headlines and direct some attention away from AMD hammering them relentlessly.

I am sure AMD appreciates a little breathing room to work on the issue that roughly half of Ryzen 3 CPU's are unable to reach advertised frequencies.

...by incredibly small amounts that have little to no measurable impact on performance. I'm not defending it, it's just such a hilariously overblown issue compared to how bad Intel's gaff is here..

I mean, it takes only a minimum of tinfoil-hatting to see the teacup-storm of the Ryzen frequency issues as a ploy by Intel. If true, this wouldn't be the first time they've resorted to journalistic skullduggey.

Whatever Intel's failings may be, their media/marketing department is (and has, at least as long as I've been building computers, been) far superior to AMD's.

"advertised frequencies" is a very interesting term these days, Intel obviously delivered its "advertised frequencies" but with all those security patches, the performance hit on Intel processors are up to 30% [1], which is believed to be 5x higher [2] than the impacts on AMD processors.

[1] https://www.techspot.com/news/77026-windows-update-reduce-sp... [2] https://hardware.slashdot.org/story/19/05/20/228218/intel-pe...

My Ryzen 3900X has absolutely storming performance despite this supposed problem. I can compile full featured qemu for one target arch in 19 seconds.

Meanwhile I'm retiring an Intel Xeon server that has slowed down so much it's become unusable (for my purposes) because of Spectre mitigations.

Whoa, sounds like it makes it possible to have a Gentoo install that's nice to use.

This was a bios bug and not an inherent limitation of the CPU.

The snark in the thread is just out of control, guys. This is supposed to be a technical forum.

In fact the part in question is shown at 8 cores, 5 GHz turbo, with a 95W TDP (at $488), which shows somewhat better power efficiency than the equivalent Ryzen 7 3800X (8/4.5GHz/105W/$399).

Now, that price differential may or may not be worth the moderate performance and power advantages to you, or you might be more enthused by the bigger Ryzen 9 and Threadripper parts instead which compete in different segments. But this certainly isn't an uncompetetive part from Intel.

Intel TDPs are the base clock, not the boost clock. AMDs are the boost clock.

A 9900K might specify a 95W TDP, but at 5GHz it will actually use almost 170W.

See https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9...

Cite for the AMD convention of TDP at a consistent boost? That's genuinely the first I've heard of it, and I'm almost certain that's wrong. TDPs have always been considered to be sustained values, which is exactly what "boost" is not.

The boost clocks generally are sustained values, at least if you have the cooling and power provision to make them so. Most good motherboards for Intel CPUs place no cap on boost duration although one is supported by the chipset. All of the benchmarks of high-end Intel parts like this are done with uncapped boost duration too. Yes, that does make Intel's TDP figures for this 5 GHz all-core boost part effectively meaningless.

AMD's current TDP figures are based the maximum thermal output in what they consider to be real-world applications with boost clocks enabled: https://www.techspot.com/article/1652-things-amd-needs-to-fi... (Their TDP used to be maximum thermal output full stop and the real-world figure used to be called ACP, but they changed it because comparing their TDP with Intel's made their CPUs look really hot.) They also have an actual guaranteed maximum which is higher than, but tied to, their quoted TDP figure. So whilst it's not actual TDP anymore it's still a lot more meaningful than Intel's numbers.

Hmm, you're right, I see anandtech data showing the 105W rated 3900x drawing a peak of ~120W. Still much closer to the rated TDP than Intel manages.

You can't really compare TDP and turbo clocks between AMD and Intel. [1]

The 9900k draws 168W at full load and is also specified for 95W TDP. This chip is likely going to use more. A 3900X draws 142W for comparison. (12 cores lower clocks 7nm)

[1] https://www.anandtech.com/show/14605/the-and-ryzen-3700x-390...

> In fact the part in question is shown at 8 cores, 5 GHz turbo, with a 95W TDP (at $488), which shows somewhat better power efficiency than the equivalent Ryzen 7 3800X (8/4.5GHz/105W/$399).

The 3900X 12C/24T uses less power under full load than the 9900K does: https://images.anandtech.com/graphs/graph14605/111362.png (142w vs. 168w)

Intel cannot touch AMD's power efficiency right now, at least not on the desktop market and up. AMD hasn't rolled out 7nm to the laptop space yet, so Intel keeps an advantage there.

The 9900K (and now 9900KS) are fast chips, but power-efficient they are very much not at all. Getting those clocks requires ramming power through the chip. Power efficiency is barely even a concern for this particular market area.

Read the article, first paragraph says "... but will feature a higher all-core Turbo frequency when it has sufficient cooling".

So at least they've added the caveat, though "sufficient" could mean liquid cooling only available to 5% of enthusiasts. Intel hasn't disclosed the TDP, so the skeptical reader would think it's not impressive enough to disclose in marketing material.

A good liquid cooler doesn't really buy you lower temperatures than a good air cooler, it just provides more thermal mass and a larger area for heat exchange to the air, so you can achieve the same sort of temps with less noise.

Pumps make noise. If you have a big case with a big fan, an air cooler should be quieter.

My apologies, I actually meant to say liquid nitrogen cooler.

judging by the number of 9900k and 9700k parts that easily overclock to 5 GHz all cores on air cooling, I really doubt a high end liquid cooling setup would be required.

> easily overclock to 5 GHz all cores on air cooling

Technically speaking, water coolers actually use air when cooling the big radiators.

yes, and heatpipes in air coolers also use liquid to transport heat to the radiator. language is messy.

all I'm trying to say is that a $60 mainstream "air cooler" that fits in a typical ATX case should be sufficient to cool this chip. it's a trivial cost for anyone considering what's likely to be a $500 cpu.

I understand people are still salty about the previous 5 GHz fiasco from last year, but this new part is literally just a binned 9900k. people are already running these at 5 GHz 24/7 without resorting to anything exotic.

For all of 3 seconds, after which the CPU catches fire unless you have an industrial chiller or clocks down to 2GHz to cool off.

To be fair, it's 5GHz turbo frequency - not an expert but isn't the entire point of the turbo mode to run in short bursts when necessary? Seems like for occasional 100% usage short workloads like compiling code it'd be great to go up to 5GHz even just for a few seconds and shave maybe a second or two off a 10s build. That's my understanding of what this could be useful for anyway.

Intel has been moving to advertising the turbo clock more prominently than the actual clock you get in typical use. Especially 10nm seems to be plagued.

Compared to a previous generation, it might boost higher for a few seconds but then clocks lower so your build takes the same time or even longer.

On the flipside, AMD's advertised frequency is the frequency that it'll have during typical use with a cooler rated for the advertised TDP. The boost clock is what it can boost on that cooler for some time, if you cooler is better you might get more.

I've got a regular ol' 9900k and it sits happily at its boost clock of 4.7 GHz the whole time on air cooling (though not 100% processor utilization for hours - it'll run through Cinebench happily without getting close to throttling), so hopefully this will be able to sit at 5.0 GHz steady assuming a relatively sane load.

News is welcomed by humans and software alike. But a bit more so by humans, since they enjoy a nice round number.

This marketing feels a lot like the days of the Pentium 4 and the NetBurst architecture.

nothing exciting here - Intel is still pushing 8 cores as its flagship consumer chip when AMD is releasing its 16 cores 3950x this month.

this is a very good reminder that for over a decade Intel tried everything possible to convince people to stay on quad cores so it can maintain unfairly high profit margin on its high core count Xeon processors.

Even if you hate AMD you have to be very thankful they are putting the screws to Intel. We need competition to drive innovation.

And I'm just sitting here waiting for the 3950X announcement so I can upgrade. Which, by the way, will drop right into my 2+ year old motherboard. Meanwhile, in Intel land, there's a new socket every 6 months.

Somewhat related: I'm looking to buy a new under-desk DL 4 GPU DL box. An AMD CPU seems like the obvious choice at this point. I can't find a motherboard that fits in a desktop case and has 4 GPU slots though. Does this exist for AMD?

You are going to need to look at a Threadripper CPU for this, the Ryzen lineup doesn't have enough PCIe lanes to support 4 GPUs. If you look for Threadripper CPUs you will certainly find EATX boards with 4 PCIe slots.

This is such a yawn inducing product. I'm not too sure this will interest the gaming segment much. Server farms will all be looking at the 4X return for purchasing EPYC over Intel and the power savings are icing on the cake.

Reply on news.ycombinator.com