News Intel Core i9-13900KS Review: The World's First 6 GHz 320W CPU

[Roland Of Gilead]

He explained it pretty well, bad things come in threes.
Bad 7xxx CPU sales, issues and returns of the 7xxx GPUs, and now we have to wait and see how the 7xxx 3dx will do.

You cynic, you!

PS, I think Intel are suffering from bad sales too!? I think it's been well reported in the last few days.

Edit: Would agree though the 7xxx X3D will be interesting to see how it stacks up. My guess prob a little faster than the KS, and perhaps more compatible with a wider range of games. As we know these will primarily be more gamer centric, but with higher core counts than last gen. Looking forward to see if AMD can get back at Intel this time for sure.

IMO as game engines progress, and demands creep higher and higher, the X3D could be the way to go with the extra L3 being more important for gaming, than not.

 

[TerryLaze]

You cynic, you!

Hey, he said it, not me.
At least I think that's what he meant.

PS, I think Intel are suffering from bad sales too!? I think it's been well reported in the last few days.

It's not bad sales.
Sales are just normalizing after years of very increased demand, intel had twice the net income for 4 years straight, we all knew that that amount of sales wouldn't last forever.
14bil quarterly revenue, which is what intel did this "bad" q4 of 2022 was pretty much an average quarter until 2019, maybe a tiny bit below but still within normal.

 

[Brian D Smith]

Doesn't seem to be helping their bottom line any if their profit/loss is accurate. But then, that's what they get producing CPU's that are SO power hungry that they need crazy cooling...people do not want watercooling maintenance and possible issues.

 

[TerryLaze]

Doesn't seem to be helping their bottom line any if their profit/loss is accurate. But then, that's what they get producing CPU's that are SO power hungry that they need crazy cooling...people do not want watercooling maintenance and possible issues.

Intel made CPUs that are extremely hard to kill, no matter how dumb you are, if you have an extremely overengineered system that is made to handle a lot of power (like all the reviewers use) then it will use a lot of power, if you have a crap system it will automatically still work as well as possible and use a lot less power.

If you want the least amount of maintenance and issues then intel is a great choice because it will just work no matter how bad the settings or the conditions you put it in. It will automatically adjust its settings to keep working.

View: https://www.youtube.com/watch?v=jv3uZ5Vlnng

 

[TheOtherOne]

Are you just going to call me a sheep, or making actual arguments to why I'm wrong here, especially also taking into account my second post? I'm really looking forward to your opinion.

It was rather a sarcastic reply about consumer behavior, not aimed at you but rather added onto your post. A pic is worth a thousand words!

 

[Roland Of Gilead]

Strike 1: October 28, 2022 Paul Alcorn reported that AMD issued a 'Performance Advisory' for Ryzen 7000 Processors. Paul said: "All is not as it seems." And which ultimately turned out to be the case and basically started to tainting the CPU at its very start! Then AMD topping the problem with pricing being completely out of touch with reality. The German retailer Mindfactory which provides daily processor sales data for both Intel and AMD reported in late December 2022 unpleasant reading for fans of the Team Red. In fact, it looking like Zen 4 chips were five times less popular than the previous-generation Zen 3. Yet the most popular Zen 4 chip the Ryzen 7 7700X, languished all the way down in 12th place and on the Geizhals shelves!

Strike 2: AMD as of last week had not enough Radeon RX 7900 XTX in stock for in a timely manner fulfilling RMA requests. A fundamental flaw within the thermal solution, throttling down to lower clock speeds and overheating and as confirmed by AMD itself. A sales person named Wally at the Tustin Micro Center last week told me that they do not get 7900XTXs anymore as AMD is short on GPUs, then is first honoring their own RMA’s and which are estimated to run over 55% . Why is the RMA number so high? Wally said that many of our MC customers after hearing the bad AMD news returned their purchases just prior to their 30-day return limitation whether or not their cards were working or not. Most wanted their money back and commented: “We want products to work as intended when they are sold off the shelf, and we don't want to jump through several hoops because the product never worked correctly before shipment.” I am sure myself that Best Buy in turn is going through the same drill!

Strike 3: The third time around...the jury is still out! The 7000X3D series coming soon! Not overclockable? From what I know AMD did not mention ‘direct overclocking’ support for Ryzen 7000X3D at any point thus far. I dearly hope that AMD will get their third chance or the new 7000X3D product introduction just right and that their marketing, pricing and actual performance will finally all dance together! I also wonder what the boys on the EVGA Kingpin MB overclock forum will have to say once the 7000X3D actually breaks loose!

That's clearer, thanks. I thought you were only referring to CPU's.

 

[bells622]

Not sure if the power consumption trade-off is worth it.

 

[PaulAlcorn]

Many thanks for your very considerate follow-up! I usually do not write longwinded commentaries in these forums and actually felt bad coming off as a cynic or even a AMD basher, which I am clearly not! After my Micro Center visit and an impromtu trip to sunny CA my thoughts kept racing back to the ‘off-the-cuff’ remarks made by the MC associate which compelled me to unload! Disappointing news! Then today the news about Intel having been forced in dropping their stock value practically to levels heretofore unseen. In this regard its a good thing that my home is in Stehekin, WA and away from the madding crowd. No roads lead to Stehekin and Amazon and all the others do not even deliver here. So for today and now I will keep on reserving my thoughts to the next snowfall and relish the quietness of my cabin in the sky!

What a view, nice!

 

[bit_user]

I found this result very interesting.

​

​

First, note that this is single-threaded. In the multi-threaded version, Intel is much higher than AMD, mostly due to AMD's lower power limits. That's not the point I wanted to make.

Rather, my reason for posting it is to highlight the discrepancy between the 7700X and its multi-chiplet siblings! This is a single-threaded test, so what's going on here? Could it really be that multi-die L3 cache snooping takes well north of 10 W? Because the turbo limits for these CPUs are:

• 7700X: 5.4 GHz
• 7900X: 5.6 GHz
• 7950X: 5.7 GHz

And weirdly, the 7900X took the top spot. So, that argues against clocks explaining most of the difference. If y-cruncher hits L3 especially hard, then it could be an excellent test showing the power overhead of Ryzen's multi-die distributed L3.

We could further speculate on why the single-threaded power is so much higher for AMD, in this case - AVX-512?

 

[bit_user]

FYI, this comment thread seems to have gotten de-linked from the review article. The new thread appears to be:

Review - Intel Core i9-13900KS Review: The World's First 6 GHz 320W CPU

We put Intel's Core i9-13900KS, the first 6 GHz CPU to market, through our extensive battery of tests to see how it fares. Intel Core i9-13900KS Review: The World's First 6 GHz 320W CPU : Read more

forums.tomshardware.com

Also, did the article get reformatted? I refreshed the page and it seems like the order of some things changed. One difference I noticed is that it was classified as News, but now is tagged as a Review. Maybe that explains it.

 

[bit_user]

Good grief, anyone choosing this over the K falls into the fool-and-their-money category.

You're making a value judgement. Not everyone is buying stuff on the basis of the perf/$ value. For someone with a lot more money, they might simply want the fastest gaming machine and don't care much about the cost. Heck, most who can afford a RTX 4090 wouldn't even bat an eye at a couple hundred extra $'s for the KS.

We see something similar with sports cars. They're not priced on the basis of horsepower/$. As you go up the power & speed range, price increases exponentially. And we could talk luxury goods, like gold chains and handbags, that serve almost zero function purpose at all. At least here, paying more money gets you better performance.

I'm not sure why there is a lot of hate for this chip ?

Poor power efficiency. Intel is building V12-powered muscle cars, in an era when a lot of people are looking to go electric.

It only uses those high watts when you're bench-marking, how often does a normal person do that, an hour every couple of months?

The benchmarks included real-world workloads, like software compilation, rendering, and video encoding.

If you want the least amount of maintenance and issues then intel is a great choice because it will just work no matter how bad the settings or the conditions you put it in. It will automatically adjust its settings to keep working.

But, then you're just wasting your money on a CPU you can't extract much of the performance from.

 

[bit_user]

Obviously all of the low-performance cores and the built-in graphics on this CPU are irrational. There is no reason for them to be on the chip.

The E-cores are the only way it can compete on multi-threaded benchmarks.

As for the iGPU, there are some people who buy these CPUs for non-gaming purposes. The iGPUs they have are perfectly fine, for daily desktop usage.

For this market segment, I think a Xeon CPU with 16 high-performance cores that can bost to 5.6GHz would be more logical.

Doubling the P-cores would throttle way faster, both thermally and via power-limits. When a single thread on a single P-core can use 48 W, what do you think will happen when you've got 32 threads on 16 cores? Performance won't scale like you want!

Concerning die area, I expect the 32 EU Xe iGPU is probably worth only about 2 P-cores, if that. So, you're really looking at a max of 14 P-cores, for the same price.

 

[bit_user]

Unfortunately, or fortunately for my wallet, both AMD and Intel have chosen to sell products which break their own efficiency curve in stock operation. While it's certainly possible to fix efficiency on them you are then paying for performance you're not getting and that fundamentally rubs me wrong.

This is weird reasoning. So, you're saying that if AMD had set lower power limits to make their CPUs stay in the efficiency sweet spot, you'd be fine with getting one? But, because they allow the CPU to boost higher by default, you consider it a "waste" if you have to change the settings to restrict it yourself?

I think a much more rational basis on which to decide is to look at how performance scales as a function of power limit. We see that most of the performance of the 7950X can be extracted by 125 W, whereas you have to keep juicing the i9-13900K to reach the knee in its perf/W curve.

​

Source: https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/2​

So, you're really not leaving much on the table, if you go with like a 7950X and restrict it to 125 W. Plus, you don't need to spend the money on a cooling solution capable of much more than that. If you opt for a model with fewer cores, the efficiency curve will bend sooner. So, the equivalent limit for the 7900X might be just 105 W.

 

[TerryLaze]

But, then you're just wasting your money on a CPU you can't extract much of the performance from.

That's what overclocking is...you either waste your money or your time, or both, trying to get the best possible results.
If you want to avoid all of that get a non-k variant on a non-z mobo that sticks to stock TDP, then you get what you pay for, which is going to be much less, and don't have to worry about anything.

Doubling the P-cores would throttle way faster, both thermally and via power-limits. When a single thread on a single P-core can use 48 W, what do you think will happen when you've got 32 threads on 16 cores? Performance won't scale like you want!

That's a weird reasoning considering that the 7950x uses 64W in that same scenario...

This is weird reasoning. So, you're saying that if AMD had set lower power limits to make their CPUs stay in the efficiency sweet spot, you'd be fine with getting one? But, because they allow the CPU to boost higher by default, you consider it a "waste" if you have to change the settings to restrict it yourself?

But, then you're just wasting your money on a CPU you can't extract much of the performance from.

Wait whut?!
Make up your mind maybe?!
How is restricting by changing settings better or even different from changing settings by just having different hardware?

 

[bit_user]

That's a weird reasoning considering that the 7950x uses 64W in that same scenario...

You've got it backwards. We don't actually know that the 7950X is optimal, as is, because AMD has no E-cores we can compare it to. If they did, then you're probably right that AMD would also have done better by building a hybrid CPU.

But, you're confused because I'm not making an Intel vs. AMD argument, here. @truerock was arguing that Raptor Lake shouldn't have been hybrid, and I'm trying to explain why it's better like this.

Wait whut?!
Make up your mind maybe?!

It only seems confusing because you took them out of context.
In case #1, I was arguing: don't buy a CPU without a roughly suitable cooling solution, or you might be wasting your money vs. a cheaper option.

In the second case, @thestryker wanted a new CPU but is lamenting the high power limits. In that case I was saying what difference does it make if you set the power limits a little lower or if the manufacturer did it? No matter what, it sounds like he doesn't want a 65 W version or fewer cores, so it doesn't make sense for him to just buy a non-K or non-X model. The point is he doesn't want all of the heat, even if he has to live with slightly lower performance.

Get it, now? Case #1 is about wanting not to waste money, while case #2 is about wanting not to waste power.

 

[thestryker]

This is weird reasoning. So, you're saying that if AMD had set lower power limits to make their CPUs stay in the efficiency sweet spot, you'd be fine with getting one? But, because they allow the CPU to boost higher by default, you consider it a "waste" if you have to change the settings to restrict it yourself?

I think a much more rational basis on which to decide is to look at how performance scales as a function of power limit. We see that most of the performance of the 7950X can be extracted by 125 W, whereas you have to keep juicing the i9-13900K to reach the knee in its perf/W curve.

​

Source: https://www.anandtech.com/show/17641/lighter-touch-cpu-power-scaling-13900k-7950x/2​

So, you're really not leaving much on the table, if you go with like a 7950X and restrict it to 125 W. Plus, you don't need to spend the money on a cooling solution capable of much more than that. If you opt for a model with fewer cores, the efficiency curve will bend sooner. So, the equivalent limit for the 7900X might be just 105 W.

At no point did I say anything was a waste simply that I wasn't interested in losing performance to get the efficiency back in line. It's not a huge difference, but I never said it was just that paying for performance I'm not getting is something I'm not interested in doing.

AMD also chose to make a very thick IHS which also hurts their efficiency it's not just power consumption in their case so I still wouldn't consider them. I'd certainly have had a less negative view on the 7xxx if it wasn't for that artificially holding things back as the Zen 4 architecture is pretty amazing with the way power scaling works.

 

[atomicWAR]

If you buy this CPU for anything else than actual high-end tasks, you are doing something fundamentally wrong, sorry. Speaking solely of Intel here. If all you do is browsing the web, you want a low-end CPU like an i3 or lower, not this thing. If you game, you want an i5 or i7, not this thing. The only use case for this CPU is high-end computing. And there, AMD is significantly more power efficient for similar results. In almost all scenarios, for that, not just Prime 95 or power viruses. Don't believe me, go back and take a real close look at the charts again.

No, just because you have a "mid-to-high end Z board, generously sized power supply and a AIO CPU cooler" in your build already, it does not mean this CPU "might be for you". As it is, the only reason for you to buy this CPU is if for whatever reason you want an Intel-CPU. Or maybe if you do a lot of high-end stuff that require single-core performance that isn't gaming. And before you call me an AMD-fanboy or some bs like that... 2/3 of my systems are Intel and the only one that isn't is a laptop. It's just, there is sensible, and then there is bonkers. This thing is bonkers.


If you aren't doing stuff that needs this performance why the heck do you buy a 12900K/KS or similar CPUs in the first place?!?
If you do, AMD provides better performance/watt.

This is exactly why I went with a 7950X. While I game, it wasn't the main focus of the build plus when I do game its not uncommon for me to be doing other compute intensive things in the back ground. And I do photo/video editing, light rendering and my rig serves as a NAS for the rest of the house at the same time. Yes that doesn't rule out a 13th gen i7/12th gen i9...BUT...

I'd much rather have the better performance per watt AMD offers (small office which heats easily/power costs) and plus the longevity AM5 platform will allow for upgrades for some time to come while the socket 1700 is nearing the end of its life. I am not some random AMD fanboy here. My primary rigs for the last 18 years have all been Intel. My last primary rig as AMD was a socket 939 platform that I upgraded from a FX-53->55->62 when AMD still ruled the roost in the Pentuim 4/D days. Though to be clear by the time I ran the FX-62 AMD no longer had the IPC lead as the first Core processors had dropped a few months prior but in socket upgrades are what AMD has always been great for allowing me to go from single to dual core over the platforms lifetime (god dualies were huge deal back then). I did have a AMD Phenom II X4 965 as a secondary gaming rig to my Intel Core 2 Quad Extreme X9650 (gave it to my nephew) and later a i7 920@ 4ghz when OCing was much more worth the effort . At which point it was retired. Point is I fanboy for no one. I buy what is best for my needs. Everyone should. Depend less on having the best or a certain brand and getting the best bang for your buck (how ever that might look depending on workload)

The Ryzen 7900X/7950X are very compelling CPUs for creators, those creators who game or those who game while doing a lot a other background processing who want to maintain a lower power/lower heat system than Intel can provide ATM.

 

[TerryLaze]

You've got it backwards. We don't actually know that the 7950X is optimal, as is, because AMD has no E-cores we can compare it to. If they did, then you're probably right that AMD would also have done better by building a hybrid CPU.

But, you're confused because I'm not making an Intel vs. AMD argument, here. @truerock was arguing that Raptor Lake shouldn't have been hybrid, and I'm trying to explain why it's better like this.

How exactly are you explaining that hybrid is better in this scenario?!
48W is the measurement of one P-core alone and not the sum of p and e-core together, so it doesn't show off the benefits of hybrid.
If 16 cores that use 64W each can not throttle then how could one that uses 48W do worse?!
What you said:
"Doubling the P-cores would throttle way faster, both thermally and via power-limits. "
Just as a funny aside, 16x64W = 1024W...
(8x48W=384W 16x48=768W )

It only seems confusing because you took them out of context.
In case #1, I was arguing: don't buy a CPU without a roughly suitable cooling solution, or you might be wasting your money vs. a cheaper option.

In the second case, @thestryker wanted a new CPU but is lamenting the high power limits. In that case I was saying what difference does it make if you set the power limits a little lower or if the manufacturer did it? No matter what, it sounds like he doesn't want a 65 W version or fewer cores, so it doesn't make sense for him to just buy a non-K or non-X model. The point is he doesn't want all of the heat, even if he has to live with slightly lower performance.

Get it, now? Case #1 is about wanting not to waste money, while case #2 is about wanting not to waste power.

But in case #2 you still pay more money for a CPU that you will not use all of its performance....

 

[bit_user]

AMD also chose to make a very thick IHS which also hurts their efficiency

It hurts thermal efficiency, but I doubt it has much impact on perf/W. The main effect is that, to avoid thermal throttling, you need a beefier cooling solution than you otherwise might (for a given power limit).

 

[bit_user]

How exactly are you explaining that hybrid is better in this scenario?!
48W is the measurement of one P-core alone and not the sum of p and e-core together, so it doesn't show off the benefits of hybrid.

Yes, it's incomplete information, but the point is that you wouldn't magically get a linear speedup. If 8W of that is uncore and 40W is core, then 8 * 40W + 8W = 328W, which is already above what the KS' limit. So, adding another 6 or 8 P-cores would mean they'd all have to throttle way back.

Eh, we'll have some idea about P-core scaling once the Xeon W 2400-series launches. That should prove the benefits of hybrid, for a desktop power profile.

But in case #2 you still pay more money for a CPU that you will not use all of its performance....

In case #2, you pay for the X or K version, because you still want more performance than the base version. You're simply willing to sacrifice a little of the top-end, because it comes at disproportionate power expenditure. That's a reasonable thing to do, especially if you live in an expensive electricity market and it's during the summertime where you have the double-whammy of having to run the A/C to get rid of the excess heat.

 

[TerryLaze]

So, adding another 6 or 8 P-cores would mean they'd all have to throttle way back.

Ryzen throttles back from over 1000W down to 230 and still does well...
If cinebench relies on wide cores far more than on clocks than it will also run better on 16 cores of intel even if they run very slowly.
It would probably still be less efficient than ryzen but it wouldn't be as terrible as everybody seems to think.

In case #2, you pay for the X or K version, because you still want more performance than the base version. You're simply willing to sacrifice a little of the top-end, because it comes at disproportionate power expenditure. That's a reasonable thing to do, especially if you live in an expensive electricity market and it's during the summertime where you have the double-whammy of having to run the A/C to get rid of the excess heat.

But both cases do the same thing, if you have everything on auto then the amount of cooling the cooler can do will decide how high the CPU will boost, and this one is the same for both intel and AMD.
They both auto boost to the limit of your cooler, so choosing the cooler is the same thing as choosing the TDP in the bios.

 

[Avro Arrow]

 

[Roland Of Gilead]

Many thanks for your very considerate follow-up! I usually do not write longwinded commentaries in these forums and actually felt bad coming off as a cynic or even a AMD basher, which I am clearly not! After my Micro Center visit and an impromtu trip to sunny CA my thoughts kept racing back to the ‘off-the-cuff’ remarks made by the MC associate which compelled me to unload! Disappointing news! Then today the news about Intel having been forced in dropping their stock value practically to levels heretofore unseen. In this regard its a good thing that my home is in Stehekin, WA and away from the madding crowd. No roads lead to Stehekin and Amazon and all the others do not even deliver here. So for today and now I will keep on reserving my thoughts to the next snowfall and relish the quietness of my cabin in the sky!

No, I wasn't suggesting either being a cynic or a basher

In fact, I'd say that I enjoy reading your posts. You often use old school English, which I find a throwback to gentler times , and lovely to read!

As @PaulAlcorn mentioned - Wow, what a view!!! Absolutely stunning.

 

[Roland Of Gilead]

We could further speculate on why the single-threaded power is so much higher for AMD, in this case - AVX-512?

Good point!

 

[bit_user]

Ryzen throttles back from over 1000W down to 230 and still does well...
If cinebench relies on wide cores far more than on clocks than it will also run better on 16 cores of intel even if they run very slowly.

Cinebench is one thing, but what about non-vectorized workloads? Intel's Gracemont cores can deliver something like 60% the integer performance of one thread running on a P-core at like 20% of the power. That's why having 16 of them is so potent.

And even for vector workloads, the E-cores aren't really that much worse.

It would probably still be less efficient than ryzen but it wouldn't be as terrible as everybody seems to think.

A P-core only Intel CPU wouldn't be winning hardly any multi-threaded benchmarks, and that's worth a lot.