Review: Intel Core i9-14900KS processor

$100 for an extra 200mhz,

 

It's a bargain. Back in 1994, I had to pay $200 more for a 10MHz uplift.

Spoiler

 

This is more of a special /collectors edition than anything else because the extra 200mhz are meaningless anyway.

 

just for fanboys
plus while the regular I-9 14900k is a power pig, this porker is a full blown hog
you can run a complete gaming rig at this level

 

You got GPU score of only 34,600 on 3DMark: Time Spy Standard with new fastest CPU Intel!? We blame Windows 11 because this OS is slower than Win 10 for CPU.

I got GPU score of 37,900 running W10 with CPU i7-12700KF@OC 5.2GHz, RAM 2x16GB DDR4@3600MHz and RTX 4090 non-OC version@stock speeds:
https://www.3dmark.com/spy/43275018 on 11/22/2023.

 

I picked up a KS, already sent it back because the binning on these are no where near what the 13900KS was. Its more just a like a normal 14900k with more voltages applied. There are more low SP level 14900ks, there are mid to high range SP KS. Hope that the KS i get next is better. Really want to run 6ghz all core. Also these cpus need to be direct die cooled.

 

I was surprised to see that the review was done in a 16x4 RAM configuration rather than 16x2. That would have lowered the gaming scores a bit I believe.

 

Lowered by a negligible level, as is the KS's improvement is over the K (or even some Ryzen models). I hardly see the point in looking at the gaming graphs when even a mobile chip offers enough FPS for competitive gamers.
The only people looking to buy this CPU are those who want to break world records. If high-performance gaming with the latest tech is your priority, an overclocked 14600K is a much more sane product. Or, a Ryzen X3D model.

 

That configuration likely hurts more than helps. The kit he used is likely Samsung based(old gen 1 DDR5) based on the timings of CL36. Hynix 2x16 kits(the only relevant memory for stability and performance reasons) at the same 6000MT/s can have timings of CL 28-32 which will perform better overall.

I think it's time to change the system memory setup for reviews.

4dimm setups are neither realistic nor recommended for end users. Intel/AMD and the board partners recommend against it too.

6000MT/s is also very slow for Intel platforms.
A user buying an i9/i7 isn't pairing it with a 6000MTs 64GB kit, provided they know any better.

2x16/ 2x24 @ 7200MTs will be achievable by the majority of setups. It's a much more realistic setup as it's neither expensive or complicated to run.

At the end of the day that memory configuration in the review doesn't make any sense.

Not negligible at all. On my setup I've benched up to 8600MT/s (I daily 8400) and it is significantly faster in memory bound titles.
That's the whole point of X3D chips. Extra cache on CPUs helps in those titles which makes ram speed less impactful. But on non large cache CPU additional memory speed scales well.

 

Come on, 8400-8600 is a unicorn setup, needing super highend 2 dimm motherboards, good silicon memory controller, memory fan and maybe 2 computers, one for games only and one for serious work related tasks when stability is needed.

 

I don't have such problems on windows 11 at all.
https://www.3dmark.com/spy/42768225

 

The context here is gaming benchmarks. I don't doubt that 8600MT/s would yield a noticeable improvement in memory-intensive workloads but in games, the improvement wouldn't matter - either the bump is insignificant (like going from 99FPS to 100) or imperceptible (going from 400FPS to 450).
I suppose if for some reason you're gaming from the iGPU or a RX 6500 then I'm sure games would see a noteworthy boost in performance at such RAM speeds haha.

When it comes to the X3D chips, that's more to compensate for inter-core communication. Keep all the cores boosted and lock a game to a select few cores and the V-cache isn't going to be quite as useful. Still better than not having one, but not a lot.

 

It's not a unicorn setup. 100% stable under stringent loads like y-cruncher.
There are guys run 8800-9000 on 2dimm boards. I'd consider that a unicorn setup.

Regarding IMC, many 14th/13th gen CPUs can do 8000+ on a board that supports it. Good 4dimm boards can also do 8000-200 ~ MT/s.

I am specifically talking about gaming benchmarks not other workloads.

No offense implied, but I can tell you lack the knowledge and first hand experience when talking about inter-core communication and saying V-Cache won't help a lot.
Words like "I doubt" and "won't matter" are followed by opinions. If it didn't really matter then there would be no market for products like that.

So to iterate. AMD implemented V-cache specifically to accelerate gaming performance. That is exactly what they say it is for in their first sentence of their page of info. The additional benefit is it also helps other memory sensitive workloads.
Inter-core communication issue is a direct result of the CCD design of Zen chips.
There is huge latency increase from inter-core communication on chips that have multiple CCDs, like the 7950x, which comes with a performance penalty. This why there are situations where the 7950x can perform worse than a 7800X.

Chips like the 7700X/7800X, a single CCD design, have no such issue. But due to the design of the Infinity Fabric in Zen chips, AMD is far behind Intel in maximum memory performance.
Roughly 6000-6200 MT/s whereas Intel can do >8000.

That's another reason AMD had to implement V-Cache to accelerate performance.
It can help a lot, depending on the game. And it can do nothing at all in games or workloads where they aren't memory bound.

I would hardly consider a 50% boost of a 7800X3D(a single 8 core CCD) over the non X3D single 8 core part insignificant or imperceptible.

Regarding Intel, memory speed is much more important. Benchmarks that use 6000 for both AMD V-Cache chips vs Intel i9 like the review here provide irrelevant data for people who are interested in realistic comparisons.

Let's create a gaming scenario that is memory bound where the 4090 is only 70% GPU usage( it happens in many titles). Let's also assume that the memory performance is the only holding thing holding performance back and the core design differences or clock speeds are irrelevant.
Memory speed is 6000MT/s for all CPUs. Assume V-Cache results in maxing out GPU usage of 4090, meaning it's used to its full potential.

7800X = 80 FPS
7800X3D = 104FPS
14900KS = 80FPS

30% is a significant increase and is noticeable in this case. And remember, there are real titles where it's 50% faster vs non 3D cache parts.
For the Intel, a reasonable kit that the majority of users will be able to use without issue is ~7200MT/s for not much more than a 6000MT kit.
7200 is a 20% increase but that doesn't directly translate to a 1:1 gain so it's hard to extrapolate an exact amount, but let's say it scales at 1:2.

In this case, memory performance would scale like this for the i9.

6000MT/s = 80 FPS.
6600MT/s = 84 FPS
7200MT/s = 88 FPS
8000MT/s = 94.4 FPS
8400MT/s = 96 FPS
8800MT/s = 98.8 FPS

In this case, the gap of performance between it and the X3D chip is much less with faster memory.
Are speeds over 8000 realistic for most? Of course not. But they are achievable for people that would likely be buying such a chip anyways because they will pair it with a 2dimm board

TLDR.
When memory speed is lacking, extra cache helps a lot. Significantly faster memory reduces the gains that additional cache has.

So if a 7800X3D with using a 14000 MT memory kit existed than the cache really wouldn't make any difference. But since it doesn't, AMD needed to employ V-Cache to increase performance.

 

Thank you. But please you should back to all stock speeds for your GPU. Do your GPU have any OC *version* as stock speeds?

 

It's still pretty cold where I live and my mom who I moved into my house loves being warm. I could get her one of these.

 

I'm testing some low voltage gaming now with the new 14900ks.
5700mhz all core on p cores and 4500mhz on e-cores.
Result was pretty good using watercooling with 20-21c water.
PS: Cpu isn't delidded

Running 8600c38 ddr5 max tuned ofc (edit: memory voltage is for 8600c36, forgot to turn it down for c38. Watercooled dimms, so doesn't matter anyway)
Battlefield 2042 128player gaming after ~30 min.

 

I trust Hilbert and Buildzoid, and both of them have trouble with 8000 memory speed stability on a 4 dimm mainboard (Hilbert: MSI Z790 ACE MAX) that alone cost more then my 7800X3d and Asrock b650e steel legend.
32GB sticks are also hard to find for those speeds, so users are locked to 48GB or 32GB with 2 sticks.
XMP8000 32GB kits are more expensive then a 6000 64GB kit.
The price difference between a 14900ks and mainboard with 8000 memory setup VS 7800X3D with 6000 memory and pretty good mainboard is enough to buy a 7900XTX.

I have no idea how you are getting notable performance improvements with memory speed, when this review shows that a blindtest with a 14600k and 14900ks would be very hard to pass at 1440p and above.

 

Key phrase being 1440p and above. Memory gaming benchmarks are typically done at much lower resolutions, ie 720p or so to get the game as CPU bound as possible. Higher resolutions make most games GPU bound, so yes, it would erase any of the performance increases from faster RAM.

This doesn't mean that faster RAM doesn't matter for high resolution gaming however. It does help, by making frametimes more consistent with less spikes, but of course the impact on framerate would be negligible.