The Intel Core i7-12700K and Core i5-12600K Review: High Performance For the Mid-Rangeby Gavin Bonshor on March 29, 2022 8:00 AM EST
CPU Benchmark Performance: Power, Office, and Science
Our previous set of ‘office’ benchmarks have often been a mix of science and synthetics, so this time we wanted to keep our office section purely on real-world performance.
For the remainder of the testing in this review of the Core i7-12700K and i5-12600K, we are using DDR5 memory at the following settings:
- DDR5-4800(B) CL40
Comparing power draw to other competing CPUs, both the Core i7-12700K and Core i5-12600K are noticeably more power-efficient than previous generations including both Intel's 11th Gen and 10th Gen Core. Though at full-load with no overclocking, AMD's Ryzen 5000 and 3000 series processors remain much more power-efficient.
In our office benchmarks, Intel's 12th Gen Core reigns supreme in Agisoft Photoscan due to its higher core frequency and IPC performance.
In the majority of our science-based benchmarks, both the Core i7 and Core i5 did well. The only benchmarks that didn't favor the 12th Gen Core series processors were in 3DPM 2.1, but more specifically in the AVX test.
Post Your CommentPlease log in or sign up to comment.
View All Comments
Gondalf - Tuesday, March 29, 2022 - linkBeh! in short words Intel have the best 7nm (or 7nm equivalent) desktop cpu, this is pretty evident.
Same applies to Mobile parts obviously. AMD approach look like rudimentary and silicon hungry versus Intel big-little silicon.
We'll see on finer nodes next year or so. Still i have the idea that Intel will never be anymore behind AMD like in past years.
eek2121 - Tuesday, March 29, 2022 - linkKeep in mind that Zen 4 is coming later this year and will make Alder Lake 2nd again. The flip/flop continues.
From a technology standpoint AMD has the best setup. They can just add chiplets or cores to increase multicore performance.
They have 90-95% of the performance for less than half, or in some cases, 1/3rd the power.
Stop trolling. It makes you look immature and childish.
DannyH246 - Tuesday, March 29, 2022 - linkYup. Completely agree, Intel are finally competitive with what? A 2yr old AMD product, but at triple the power. Yet some of the posters on here are acting like Intel are saviors who have finally returned to save us all from our PC ills. They forget that Intel literally screwed us all for a decade with high prices, feature lock outs, core counts limited to 4, new motherboards required every year, anti competitive behavior, and anti consumer practices. Why anyone would want Intel to return to the dominant position that they used to occupy is beyond me.
Khanan - Tuesday, March 29, 2022 - linkFully agree.
lmcd - Tuesday, March 29, 2022 - linkPretty delusional here -- AMD's whole plan with Bulldozer was to create vendor lock-in where you used an AMD GPU to accelerate floating point operations on an AMD CPU. Consumer-friendly to claim 4 cores are really 8? Thuban outperformed Piledriver until its instruction set finally was too limited years later. AMD just wanted a marketing point.
Also, how are you playing both narratives at once? Intel both wasn't willing to deliver more than 4 cores and has too high of power consumption in its parts with more than 4 cores? Which is it? Because it quite literally cannot be both.
Samus - Wednesday, March 30, 2022 - linkOMG people are still going on about the shared FPU BS.
It HAD to be that way. The entire modular architecture was built around sharing L2 and FPU's. There was nothing inherently wrong with this approach, except nobody optimized for it because it had never been done, so people like you freaked out. If you wanted your precious FPU's for each integer unit, buy something else...
Otritus - Wednesday, March 30, 2022 - linkIt did NOT have to be that way. AMD intentionally chose to share the fpu and cache creating a cpu that was abysmal in performance and efficiency, and deceptive in marketing. Thuban was a better design, so updating the instruction set would have made a better product than bulldozer.
I along with most people chose to buy something else (Sandybridge to Sky Lake). The lack of a competitive offering from AMD created the pseudo monopoly from Intel, and many people rightfully complained about the lack of innovation from Intel.
Also there is no optimizing for hardware that isn’t physically present. 4 fpus means 4 cpu cores when doing floating point tasks, which is highly consequential for gaming and web browsing.
There is no arguing for AMD’s construction equipment family. It sucked, just like Intel’s monopolistic dominance and Rocket Lake.
AshlayW - Wednesday, March 30, 2022 - linkFYI, most instructions in day to day workloads are integer, not floating point. Your argument that Floating point is "highly consequential" for gaming and web browsing is false.
mode_13h - Thursday, March 31, 2022 - link> most instructions in day to day workloads are integer, not floating point.
> Your argument that Floating point is "highly consequential" for gaming
LOL, wut? Gaming is *so* dominated by floating point that GPUs have long relegated integer arithmetic performance to little more than their fp64 throughput!
Geometry, physics, AI, ...all floating point. Even a lot of sound processing has transitioned over to floating-point! I'm struggling to think of much heavy-lifting typical games do that's *not* floating point! I mean, games that are CPU-bound in the first place - not like retro games, where the CPU is basically a non-issue.
Dolda2000 - Saturday, April 2, 2022 - linkWhile games certainly do use floating-point to an extent where it matters, as a (small-time) gamedev myself, I'd certainly argue that games are primarily integer-dominated. The parts that don't run on the GPU are more concerned with managing discrete states, calculating branch conditions, managing GPU memory resources, allocating and initializing objects, just managing general data structures, and so on and so forth.
Again, I'm not denying that there are parts of games that are more FP-heavy, but looking at my own decompiled code, even in things like collision checking and whatnot, even in the FP-heavy leaf functions, the FP instructions are intermingled with a comparable amount of integer instructions just for, you know, load/store of FP data, address generation, array indexing, looping, &c. And the more you zoom out of those "math-heavy" leaf functions, the lesser the FP mix. FP performance definitely matters, but it's hardly all that matters, and I'd be highly surprised if the integer side isn't the critical path most of the time.