Monday, January 10th 2022
Intel Core i5-12490F Beats Core i5-12400F By 15% in Early Performance Benchmarks
A few days ago, we reported a strange Intel Core i5-12490F processor that appeared in the Chinese marketplace. The processor uses the C0 silicon that Intel sits on a pile of and repurposes it to make these odd chips for Asian markets. As we found out, this C0 silicon is a heavily cut-down version, with only six high-performance P-cores present. Compared to the regular Core i5-12400F, it has a bigger L3 cache arriving at 20 MB and slightly higher clock speeds where the base stands at 3.0 G and a boost frequency that manages to ramp up to 4.6 GHz. As a reference, the regular Core i5-12400F has 18 MB of L3 cache, a base frequency of 2.5 GHz, and a boost speed of 4.4 GHz.
Thanks to the early benchmarks, we have the performance numbers in two cases where Intel's Core i5-12490F model is compared to the regular Core i5-12400F. According to the GeekBench data, the first case proves that higher clock speeds of the strange processor, coupled with higher L3 cache, prove to be of help as the single-threaded performance grows by 10%. In comparison, the multi-threaded results show an even more considerable improvement at 15%. The second test shows smaller margins compared to the Core i5-12500, where the Core i5-12490F processor now only leads by 2.5% and 5% in single-threaded and multi-threaded workloads, respectively. This indicates that we have to wait for more benchmarks to see how this design stands in the Alder Lake family and see just how big of an improvement comes from higher frequencies and bigger L3 cache.
Source:
via Tom's Hardware
Thanks to the early benchmarks, we have the performance numbers in two cases where Intel's Core i5-12490F model is compared to the regular Core i5-12400F. According to the GeekBench data, the first case proves that higher clock speeds of the strange processor, coupled with higher L3 cache, prove to be of help as the single-threaded performance grows by 10%. In comparison, the multi-threaded results show an even more considerable improvement at 15%. The second test shows smaller margins compared to the Core i5-12500, where the Core i5-12490F processor now only leads by 2.5% and 5% in single-threaded and multi-threaded workloads, respectively. This indicates that we have to wait for more benchmarks to see how this design stands in the Alder Lake family and see just how big of an improvement comes from higher frequencies and bigger L3 cache.
18 Comments on Intel Core i5-12490F Beats Core i5-12400F By 15% in Early Performance Benchmarks
(4.6-4.4)/4.4 = 0.045
1.11*1.045=1.15995
So ~ 15%
Even though different dies these should be with identical architecture in terms of cores. Cache alone should not play that large a part.
I would be very happy.
Its not like the single thread gap is of any interest anymore for most applications, including gaming. Oh... how the tables have turned the last five years. Intel's hammering a performance metric that is yesterday's news for MSDT.
Zen2 was good because it had a much big cache than Zen1
Zen3 was good because it unified the big cache, making it faster.
Zen3 3D should be good because it has even more cache
Alder Lake is good because it has a bigger cache than prior generations
This 12490F is better because it has a bigger cache that the 12400F
If you haven't spotted the pattern yet, you're on your own!
Pretty much all the games I play are more dependant on IPC/single thread perfomance than core count and I do no work or any kind of CPU intensive tasks on my PC just casual gaming+media.
Sure if we strictly talking about high performance PC in 2022 then yes I agree that 8/16 sounds about right but in most cases/users the 12400 is plenty enough imo.
For a all-round performant gaming/photo editing/coding rig, a fast 6 core is plenty. Unless you can find a higher core count on discount, most of even enthusiasts don't really benefit much for 8+ cores.I've been saying for a long time that gaming will get to a point of diminishing returns, until games get more demanding for CPUs, which is not currently happening at a high rate.
Higher IPC helps for more responsive applications in general though.The fact police is here;
Zen 2 - Major improvements in CPU front-end, greatly improved FPU engine. L1I cache is actually smaller, while L3 can be larger (depending on config).
Zen 3 - Major improvements in CPU front-end, including branch prediction, prefetching and µop cache. Improved load/store throughput, faster FPU operations, and various improvements to smaller queues and caches which matters much more than the unified L3, which is in fact slower.
Zen 3 3d - We don't know yet.
12490F - Going from 18 MB to 20 MB L3 is not going to make a big difference. The clock speed is causing the difference here.
Caches have increased and decreased many times over the last 30 years, and their bandwidth have increased tremendously. Don't get too fixated on caches and miss the big picture. The largest improvements to come are not caches, caches kind of come secondary to other design considerations.
So you're not wrong, because the higher turbo speeds did more than nothing, but they made only a tiny fraction of the difference; Clock-normalised tests (performed at 4.0GHz) had Zen3 still 15% faster so that was just unifying the cache and other architectural optimisations.BRB, off to microwave my testes.Whoa there, you and your facts! That's not we want to see in these conjecture and rumour threads ;)
My post was very tongue in cheek as it's obviously not just cache. I referred back to this particular AMD slide quite a few times in the numerous discussions just before and after the Zen3 launch - maybe 30% of the 19% IPC uplift came from caches.
I'm genuinely curious about the 5800X3D since AMD themselves are not claiming any significant architectural changes; It's simply regular 5800X with 3x the cache. The fact it's significantly faster whilst running at lower clocks is impressive, but so far we only have AMD's claims and not any independent, real-world testing.
- Cache prefetching - L2 (controlled by front-end)
- Execution engine - (back-end)
- Branch predictor - (part of front-end)
- Micro-op cache - (part of front-end)
- Front end - (I assume they mean misc. front-end improvements)
- Load/Store - (part of the back-end)
While L3 is a spillover ("victim cache") for L2, so it's a fairly minor contributor to performance gains. There are probably some edge cases where the unified L3 matters a bit, but in general it's fairly minor. :)Well, the fact is that we have fairly little information about the real world difference. ;)
There are probably some workloads where the difference is substantial though.