One thing we have been requested plenty of occasions throughout HUB Q&A collection is, “Why would not Intel develop their very own type of 3D V-Cache?” It is a thought-provoking query, particularly given how useful 3D V-Cache has been for AMD’s Ryzen chips, regardless of some drawbacks.
These drawbacks embody decreased core clock speeds, increased energy utilization, and better working temperatures. Within the case of Ryzen CPUs, the clock velocity discount is not important since these components aren’t clocked very excessive to start with, and the efficiency uplift achieved with the bigger L3 cache greater than offsets the lower in clock velocity. Energy utilization is not a serious concern both, and whereas thermals might be a difficulty with Ryzen processors, throttling is simple to keep away from.
Within the case of Intel’s CPUs, a minimum of their present thirteenth and 14th-gen Core collection, the facility finances is fairly effectively maxed out. This has been the topic of a lot dialogue over the previous few years and just lately got here to a head.
We additionally know that Intel depends closely on clock frequency to get essentially the most out of their components. So that you is perhaps questioning if extra L3 cache might clear up their points by decreasing clock speeds and due to this fact energy consumption, whereas additionally boosting gaming efficiency past what we’re at the moment seeing.
To get an concept of whether or not extra L3 cache could be useful, we have rehashed some testing performed with the Tenth-gen Core collection, however utilizing the newer 14th-gen components (aka Raptor Lake). Basically, this can be a cores vs. cache benchmark.
We took the Core i9-14900K, Core i7-14700K, and Core i5-14600K, disabled the E-cores fully, locked the P-cores at 5 GHz with the Ring Bus at 3 GHz, after which examined three configurations: 8-cores with the 14900K and 14700K, 6-cores with all three processors, and 4-cores with all three processors.
This testing methodology supplied us with nice perception into how the Tenth-gen collection in contrast and what was the reason for many of the efficiency uplift in video games again then (the reply in that case was L3 cache). Nevertheless, the Tenth-gen collection had significantly much less L3 cache, simply 20 MB for the Core i9’s, 16 MB for the i7’s, and simply 12 MB for the i5’s. Quick ahead to immediately, and the 14600K has extra L3 cache than the 10900K at 24 MB, whereas the 14700K has 33 MB, and the 14900K has 36 MB.
The cores themselves are clocked increased, although the configuration right here has modified loads, particularly with the E-Cores disabled. Lastly, we’re additionally utilizing DDR5-7200 reminiscence and the GeForce RTX 4090, so let’s have a look at what the outcomes of the benchmarks appear like.
Benchmarks
Beginning with the Murderer’s Creed Mirage outcomes, this information seems nothing just like the Tenth gen testing we did. The outcomes are very sudden as they seem CPU-limited, they usually certainly are. But, including extra cores, from 6 to eight, or rising the L3 cache capability would not enhance efficiency. Somewhat, the bottleneck right here for the 6 and 8-core configurations seems to be the clock frequency of 5 GHz, suggesting that the first limitation of the 14th gen structure is frequency, not L3 cache capability, a minimum of on this instance.
Dropping down to only 4-cores does see efficiency fall away, and now cache capability performs a really small function. Nonetheless, it is outstanding to see the 14900K drop lower than 20% on this take a look at with simply half the P-cores lively.
Helldivers 2 does see some efficiency degradation when dropping from 8 to six lively cores, although we’re solely speaking a couple of 7% discount within the common body fee with a a lot smaller lower within the 1% lows. Dropping to only 4-cores does slash efficiency, significantly within the 1% lows, that are decreased by a bit over 40%, resulting in a lower than best expertise.
Ratchet & Clank, like Murderer’s Creed Mirage, sees nearly no distinction in efficiency between the 6 and 8-core configurations, once more suggesting that core clock frequency is the first bottleneck right here. It is not till we scale back the lively P-core depend to only 4 that we see some efficiency degradation, and the drop-off right here is far the identical no matter L3 cache capability.
The Spider-Man Remastered outcomes are attention-grabbing for a couple of causes. Firstly, with 8 lively cores, the 14900K and 14700K delivered just about similar outcomes at 5 GHz, so once more in that instance, the clock frequency was the first bottleneck. However when dropping all the way down to 6-cores, it appears cache turns into extra essential. Although the margins aren’t precisely big, the 14700K was simply 3% slower than the 14900K, for instance.
Then fairly oddly, when dropping down to only 4-cores, the outcomes develop into frequency restricted once more, regardless of efficiency solely dropping by 22% when in comparison with the 8-core outcomes.
Cyberpunk 2077: Phantom Liberty sees little or no distinction between the 8-core configurations. The 14700K is once more simply 3% slower than the 14900K, and we’re solely seeing a 7% discount in efficiency with 6-cores lively after which a 22% discount from 8-cores to only 4-cores. No matter what number of cores are lively, it appears at 5 GHz cache capability performs a really small function.
Hogwarts Legacy sees no distinction in efficiency with the 8-core configurations, with a minor variation seen with 6-cores. The distinction between the 14600K and 14900K right here is simply 5%, and it is even much less with simply 4-cores lively. The efficiency distinction between 8 and 6 lively cores can be very small, with 4-cores decreasing body charges by simply shy of 20%.
Horizon Forbidden West is one other sport that seems primarily frequency restricted as cache capability makes no distinction right here, whatever the core configuration. Dropping from 8 to six cores solely decreased the efficiency of the 14900K by 5%, after which from 8 to 4 cores noticed a 22% drop, which is what we have usually come to anticipate.
Dragon’s Dogma II is thought to be a really CPU-limited sport, however like many of the titles examined, we’re not seeing a lot of a efficiency enchancment when rising L3 cache capability. The 14900K, for instance, was simply 4% sooner than the 14700K when all 8 P-cores had been lively.
With 6-cores, the 14700K was 9% sooner than the 14600K, whereas the 14900K was once more 4% sooner than the 14700K. Then with simply 4-cores lively, core depend turns into the first bottleneck.
The Baldur’s Gate 3 efficiency was once more a lot the identical utilizing both the 14700K or 14900K, and we solely noticed as much as a 5% lower in efficiency when dropping all the way down to 6-cores. Then with simply 4 lively cores, the 14900K was simply 14% slower when in comparison with the 8-core configuration.
The Final of Us Half I has some actually attention-grabbing outcomes for us. On this instance, we’re a 5% efficiency drop from the 14900K to the 14700K operating with 8-cores enabled. However with simply 6-cores, that margin is decreased to only 2%, after which nothing with 4-cores. Somewhat, it is the core depend that makes essentially the most distinction on this instance.
For instance, the 14900K noticed a ten% discount in common body fee when going from 8 to six cores and a 36% drop for the 1% lows. Then from 6 to 4 cores, the common body fee dropped by an additional 24%, with a 32% discount for the 1% lows. So going from 8 lively cores to only 4 noticed the 1% lows greater than halve on this instance.
Starfield is one other sport that scales effectively with cores, and yet one more instance the place the cache capability has little to no affect on this testing. The 14900K noticed an 11% efficiency discount when going from 8 to six cores, after which an additional 24% discount from 6 to 4.
Overclocking at 5.7 GHz
Earlier than wrapping up this testing, we went again and re-ran a couple of of those benchmarks with the cores clocked 14% increased at 5.7 GHz, together with a 33% enhance to the Ring Bus at 4 GHz.
We did this as a result of we anticipated that some extraordinarily well-informed folks on the market would name the testing ineffective as a result of 14th gen processors can clock increased, and 5 GHz is unrealistic. Regardless of already studying that the bottleneck is certainly clock frequency and, in some cases, core depend, not cache capability. So sure, clocking these components increased will enhance efficiency, however past that, it will not educate us something new.
Anyway, having examined a couple of video games, we discovered related traits, so the limitation right here for 14th gen actually is clock velocity. Extra cache would not seem to assist.
What We Realized
We have got to say these outcomes had been fairly stunning, as they are much completely different from the Tenth-gen outcomes we recorded three years in the past. Again then, we discovered that, for essentially the most half, when locked on the identical frequency, there wasn’t a lot distinction between the Core i5, i7, and i9 processors in most video games. And when there was a distinction, it might largely be attributed to the L3 cache.
However with the 14th gen, the cache capability would not seem to matter a lot. In most examples, 24 MB was sufficient. Oddly, an elevated cache capability usually helped essentially the most when fewer cores had been lively, although that wasn’t all the time the case.
What made essentially the most distinction was the core depend; the improve from 4 to six cores was usually very important, although none of those components include simply 4 cores. There have been additionally a couple of examples the place going from 6 to eight cores made fairly a distinction.
We will thus conclude from this information that including 3D V-Cache to those Intel 14th-gen Core CPUs would doubtless be detrimental, solely serving to cut back gaming efficiency (a minimum of in immediately’s video games) which is stunning.
We additionally think about that going to 10 and even 12 P-cores in immediately’s video games would solely lead to a really minor efficiency acquire over 8 cores. So, for Intel, the one method ahead with their present structure is clock velocity, which explains why issues have gone the best way they’ve, a minimum of up to now.