Gaming performance and efficiency compared with the i9-12900KS
I don’t want to and can’t provide a big review with the various other SKUs in comparison – Igor does that far more professionally and of higher quality anyway. Instead, I just grabbed the new absolute high-end CPU i9-12900KS and practically compared how much more performance you really get with an i9 relative to the i5, and how much efficiency falls by the wayside in the search for the last extra FPS. The CPU- and RAM-demanding game Shadow of the Tomb Raider is used for this, in 1080p and with the “Highest” preset. An Nvidia GeForce RTX 3090 Founders Edition with maximized temperature and performance limits acts as the pixel accelerator. The data is collected with the tool Nvidia Frameview 1.2, based on the open source software Presentmon.
As RAM I also threw the DDR5 JEDEC-Config with 4800c40 into the ring and additionally a manually optimized 6000c32 setting for each CPU, as it should also be stable on most 4-DIMM Z690 boards by now. Of course, you could also run the i5-12490F with an even cheaper DDR4 board without any problems, but there is no such board with an external clock generator and thus BCLK-OC capability as of today. In addition, the SA voltage is also locked on non-K chips and thus fixed at 0.9 V, which could become a limiting factor especially for high 1:1 clock rates in DDR4.
The i9-12900KS is of course the fastest in gaming, and another 9.2% faster when it is also reliably fed by the faster RAM. The i5-12490F increases almost exactly in the same ratio when comparing JEDEC to the manually set RAM config. With all cores set to 5 GHz, the imported Chinese even manages to clearly pass the 12900KS with standard boost and RAM settings, which is probably how it will be operated in most systems in the end, unfortunately. While the difference in the average FPS is only 5% compared to the i9, it is already a 9% deficit in the 1% low FPS.
The picture of the frame time variances coincides with those of the FPS. The tuned i5 with fast RAM is virtually on par with the i9 with slow RAM. By the way, overclocking the i9-12900K manually is of course possible, but it requires much better cooling than an AIO and only brings marginal gains.
And if we now look at the efficiency, in the form of the power consumption per frame, we don’t even want to turn the clock screw on the i9. The i9 with its 8+8 cores has to pay a lot for its advantage in gaming at the power supply. Even the overclocked i5 is almost twice as efficient as the i9 with the same RAM configuration. The efficiency advantage even grows to almost three times when the i5 is also allowed to automatically boost as energy-efficiently as Intel actually intends and when the RAM runs at 4800c40.
While we’re on the subject of efficiency, I still have to say a word about AVX512. Intel has deactivated the instruction set for all chips from 2022 and also retroactively for all 12900KS on the silicon level. This is not important for gaming in most cases, but it is for some compute tasks and benchmarks. So here the i5 with AVX512 can save an additional 7-9% in power consumption in Prime95, while the i9 with only AVX2 starts to throttle the clock after a few minutes due to the temperature and still consumes over 240 W on average.
Of course, the i9-12900KS with its 2 extra P-cores, 8 extra E-cores and approx. 1 GHz higher clock rate is also significantly faster – the additional power consumption doesn’t go to nowhere – but this diagram does make you think and fantasize: How efficient would an Alder Lake CPU without E-cores, with 8 or even 10 P-cores and AVX512 be? Of course, I don’t want to withhold the test system with the used hardware from you:
Test systems |
|
---|---|
Hardware: |
|
Cooling: |
|
Housing: |
|
Periphery: |
|
Measuring devices: |
|
8 Antworten
Kommentar
Lade neue Kommentare
Urgestein
Moderator
Urgestein
Veteran
Mitglied
Urgestein
Urgestein
Veteran
Alle Kommentare lesen unter igor´sLAB Community →