Let us start with the large, pink, elephant in the room. The 14th Generation is best described as a refinement of the 13th Generation. Yes. One could cut and paste the 13th generation’s data… and just change the frequencies the CPUs run at to create the ‘new’ 14th Generation data sets. Yes. We can already see the next salvo in the ongoing meme war being “10nm++++” and “the Great Lakes Recycling” being all the rage with AMD fans. We do not consider this a deal breaker but certainly will understand and sympathize with those frustrated with the current state of the consumer desktop CPU marketplace.
The reason for this recycling is varied but boils down to Intel’s inherent conservativism. Yes. The 14th generation was coyly hinted at being both Intel’s debut of Compute Tile technology (their take on ‘chiplets’) and the ‘sunsetting’ of their monolithic architectural design… and the ‘end’ of the “Core-I” brand (with the ‘-I’ being dropped). Since Intel refused to purposely release a half-baked product (e.g. Zen 1 and its latency storm issues) this did give Intel’s design team a chance to let their Core-I monolithic design go all out in one last blast. So, it is indeed true that this is (technically) the fourth generation to use Intel 7 / “10nm” fab process (12,13,14th gen + blink and you missed it Tiger Lake). It is also indeed true this is the 13th Generation with a speed bump and a new name. It is even true that this will harm the Intel brand in the short term – as only enthusiasts and historians will get what Intel is doing.
The upside to all this controversy is buyers get access to a 6GHz twenty-four core (8 p-core, 16 e-core) monster for about what AMD wants for their flagship sixteen core model. Yes, the 13900KS was also (somewhat) clocked the same as the 14900K but thanks to a smidgen more node process refinement the 14900K does not require as much voltage to offer 6GHz and is a tiny bit cooler running than the last generation’s 5.8GHz 13900K… and noticeably cooler running than the 13900KS.
The downside is obvious. There is a grand total of zero architectural changes in the i5 and i9 14th Generation Core-I processors (the i7 is a different matter, but as we do not have one on hand, we will not comment on it). This means up to eight ‘p-’ / Performance cores in two blocks of (up to) 4 cores configuration (for the i9 with the i5 once again having 2 of the p-cores disabled/laser cut). Also, with the 13th generation these ‘big’ p-cores make use of 80KB of L1 cache per core, 2MB per core of L2, and total L3 being ‘up to’ 36MB (including the L3 for the e-cores… and once again the i5 with two cut p-cores and two blocks of e-cores ‘only’ has 24MB L3 in total).
For the most part, this is fine for one more generation as AMD is still behind the curve on L1 and L2 cache. For example, the (flagship) Ryzen 7950 makes use of 64KB of L1 cache per core and 1MB per core of L2… and Intel’s L1 and L2 cache is still more sophisticated than AMD’s. Intel simply has more resources to research advancements in highly esoteric L1/L2 cache algorithm regiments compared to AMD. To be precise the 14th gen uses a 12(data)/8(instruction)-way configuration for the L1 (vs. AMD’s 8/8) and a 16-way for the L2 (vs. 8 way)… and the AI-Leaning based “L2P” L2 cache algorithms have also been slightly improved. All of this adds up to better efficiency on top of simply more of it compared to AMD. So while disappointing, it is tolerable. For one more generation.
With that said Intel is behind the 8-ball when it comes to L3. Yes, their “Dynamic INI” algorithms (a dynamic inclusive/non-inclusive mechanism that monitors and adjusts in real-time what should be cached based on real-time active workload modeling) introduced in the last generation do make their L3 still cutting edge… but L3 is where AMD has placed more of their limited resources and it has been paying dividends. Even excluding exotic ‘3D’ stacking options, the bog standard R9 7950X offers a whopping 64MB of L3 cache.
Yes, L3 is slower than L2 (which in turn is slower than L1)… but it is eons faster than going off-chip to the RAM. Put bluntly, one can have the absolute best L3 algorithms ever dreamed up but “quantity has a quality all its own”. AMD cores can simply make more guesses and put more data in their L3 than Intel. Which in turn closes the performance gap – and sometimes even exceeds it. Once again things are not dire (or even desperate) but Intel will have to release a new generation before AMD releases their next-gen CPUs lest Intel lose all the IPC gains that they clawed back with the 12 and 13th generations.
Shifting over to the Atom… err… ‘e-cores’. Once again nothing has changed beyond frequency boosts. This is fine as Intel already fixed the 12th gen e-core cache issue in the previous 13th generation. This means each e-core has 96KB total (64KB instruction + 32KB data) L1 cache per core and each block of four e-core clusters has a shared L2 cache of 4MB. This means that even when all four e-cores are active they are less likely to have to go to L3 to get a cache hit… which they also share with the p-cores (and thus further increases the chance of a cache hit well beyond what one would expect from a big.Little architecture).
So, while it is indeed true we want to see each e-core’s L1 increased (in size and complexity), and have the L2 doubled again (i.e. 2MB per core / 8MB per block) in the next generation… the amount of cache they do have access to results in a fairly consistent user experience. After all, and to put those numbers in perspective, the e-cores in the 14th generation i9 have the same amount of L2 cache as the last Intel 16 core HEDT ever made (Core i9 7960X, which ironically had a Turbo Boost of 4.4GHz… or the same as what the e-cores come clocked at in the i9-14900K); and the e-cores in the 14th generation i5 have the same amount of L2 cache as the Intel 8 core HEDT 7820X processor (which also ironically had a Turbo Boost of 4.0GHz or also the same as what the e-cores come clocked at in the i5-14600K). Needless to say, they are nowhere near as cache-starved as they were two years ago when Intel ‘created’ the idea of desktop e-cores. Nor are they as ‘low performance’ as certain circles of fans have intonated… as these e-cores use a similar(ish) architecture as the HEDT era used.
The mesh interconnect (aka ‘Ring Bus’) also has not changed in any notable way from the previous generation. That is to say, it still is extremely fast (mid 4 to 5GHz depending on core load/heat load /etc.), is fairly efficient, and gets the job done in such a way that is not dependent on the Integrated Memory Controller’s frequency. The downside is while the improvements implemented in the 13th generation did indeed negate most of the ‘e-core saturation storm’ the e-cores can still become a smidgen starved under extreme loads… and why some PC gaming enthusiasts still feel the need to (wrongly) disable the e-cores. Instead of taking such extreme measures, for most, a touch of overclocking to the core will eliminate, or at least obfuscate, any bottlenecking in games and allow those uber-hardcore gamers to keep their e-cores active. Which can make a difference as they can handle a lot more than one thinks.
Sadly… the IMC has also not changed. While rumors, which in retrospect are “pure copium”, said that Intel would be improving things from the decent DDR5-5600 to DDR5-6400 such things were not meant to be. Instead, the ‘14th’ Generation IMC is once again limited just like the ‘13th’ Generation. That is to say that when using two sticks of single rank memory (AKA ‘1DPC 1R’… aka “One DIMM Populated per Chanel using single rank RAM”) the IMC still supports speeds of only up to DDR5-5600. With two sticks of dual rank DDR5 used (aka ‘1DPC 2R’) the IMC tops out at DDR5-5200, and when one wants to use a Fully Monty 128GB (or 192GB) configuration of 4 bigboi sticks the IMC is going to downclock to DDR5-4400.
Of course, this is all ‘factory certified’ performance levels, and just as with the 13th Generation the IMC is capable of more. How much more will vary, but with our samples we easily hit DDR5-5000 with 4 sticks of 32GB DDR5. The same is true with two sticks of 32GB RAM and hitting DDR5-5600 was a snap. Just as it was with the 13th generation. All that will be necessary is to increase the voltages, loosen timings, and play with things… as XMP is still not 100 percent reliable when dealing with 2DPC 1 and 2R configurations. This is the same quirk as the last gen… as this is the same IMC as the last gen.
This is, in our opinion, the largest complaint many will have with the 14th gen being a refined 13th gen. While it is true oddball 48GB sticks are becoming commonplace (allowing for 96GB in 1DPC 2R at DDR5-5200 without IMC overclocking) and those dual kits will satisfy some, 128GB of RAM is and about the only way to keep 24 cores from being RAM starved for HEDT / “workstation on a budget” user scenarios. We sincerely hope that Intel’s next-gen IMC is much more robust… as much like L3 cache levels DDR5 IMC performance is an area AMD is quickly pulling ahead of Intel.
Thankfully, the upside to the IMC not changing is the fact that Intel once again is allowing value-orientated buyers the luxury of using DDR4 sticks. This benefit really cannot be overstated or overhyped. While there is a bit of performance impact… being able to ignore DDR5 pricing for another round of system upgrades is going to be a major selling feature for a lot of buyers. Especially since AMD AM5 is DDR5 only and this is an ‘Intel exclusive’ feature.
This brings us to the actual major changes that Intel has introduced with the 14th generation. Namely, frequencies, which we have already gone over, and power level profiles. With AMD moving the Overton Window into formerly ‘here hold my beer’ territory, Intel has felt comfortable enough with offering even higher power profiles. Yes. The 13th generation already turned the dial to ‘10’ with PL2 settings of 253 watts for the i9 and 181 watts for the i5. With the 14th generation, these defaults have stayed the same… but Intel has added another power level tier. Aptly named the ‘Extreme Power Delivery Profile’ which allows the amperage to peak at 400A for the i9 and 200A for the i5. Furthermore, with amps now specified there is no longer a PL1 (aka “Processor Base Power”) vs. PL2 (aka “Maximum Turbo Power”) distinction, nor is ‘Tau’ (aka the length in seconds the CPU is allowed to stay in PL2 mode) a setting one need worry about. Instead, it is now all about power draw. Basically, there is no longer a timer on the PL2 power peaks and if one so chooses, PL1 and PL2 are now basically the same as far as the internal power algorithms are concerned. So instead of X seconds at 181/253 watts, and then it drops down to a lower tier…. 14th Gens can run in PL2 as long as they do not A) try and slurp down more than 2/400A of power and B) the temperatures stay below the thermal limitations set out by Intel and your motherboard’s BIOS (e.g. below 73c for Thermal Velocity Boost, below 100C for true thermal limitation). Needless to say, this is not a good idea if you do not use an ultra-high-end cooling solution. We recommend a 360mm AIO as a good starting point if such things interest you.
Last but certainly not least is due to the extremely short interval between the 13th and 14th generation Intel has made the extremely wise decision of breaking with “tradition” and allowing a 3rd generation of processors to be used with the same socket. They have gone as far as not releasing a new chipset to accompany the 14th Gen Core-I processors! As such, instead of a ‘Z890’ motherboard is required to use the 14th generation CPUs… existing 13th and 12th Gen owners need only upgrade their Z690 or Z790’s BIOS and swap in the new CPU. That is a game changer for value-orientated buyers who want/need/desire more power but really cannot afford an entirely new system.
Of course, motherboard manufacturers are not exactly enthused with this breaking of the status queue ante and have released improved Z790 motherboards. For example, MSI calls their Z790 gen 2 motherboards Z790 “Max” motherboards. In a future review, we will go over what changes the gen 2-based Z790s bring to the table but the biggest is the inclusion of WiFi 7. It remains to be seen if such tweaks warrant a new generation of models, but if one is looking for a new motherboard to go along with their new Intel 14th Gen CPU MSI Z790 ‘Max’ options are certainly worth considering over their non-Max predecessors.
