That's very specific to scalable workloads (i.e. >= 32 threads), though. It's not as if its efficiency on lightly-threaded tasks improved anywhere near as much, yet that what most people spend most of their time running.
This is what I experienced, side-grading from a 12900k to a 13900 and I run it at 65W, I see better performance than my 12900k (Passmark CPU went from ≈11k to ≈12k, if I let it eat up power I can get to 15-16k. but in its new 6L case... not worth it)
Same efficiency for 12th an 13th gen on their 13900k review...That's very specific to scalable workloads (i.e. >= 32 threads), though. It's not as if its efficiency on lightly-threaded tasks improved anywhere near as much, yet that what most people spend most of their time running.
I'm also suspicious of exactly what they mean by 65 W, in that context. You're treating it as if they mean PL2=65 W, but that figure appears to be referring to PL1. That doesn't mean you're wrong (it wouldn't be the first marketing slide to have such an inconsistency), but it just shows that we should really want to see controlled experiments and not just a marketing slide.
When we dig into some actual data, efficiency on Cinebench nT, which is definitely a scalable workload, only improved 17.8%. Hardly insignificant, but far short of their 41% and also leaving it just 75.4% as efficient as 7950X.
On single-threaded Cinebench, it actually regressed, being just 89.3% as efficient as the i9-12900K.
It still beats many Ryzen X-series CPUs, but such a big drop isn't what I'd expect to see after hearing someone speak so effusively about its efficiency.
Don't get me wrong: I'm not anti-Raptor Lake. I was just confused by the incongruity of how you were talking vs. all the benchies I've been seeing.
Inconsistency noted. I'd expect the newer data from the 7950X3D review is more relevant (i.e. more current OS patches and software revisions), if not outright better quality, but we shouldn't read into it without more information.Same efficiency for 12th an 13th gen on their 13900k review...
Difference of 54 to 70 on the 5800x non-3d
They either screwed up or something changed on how they measure.
What I recall from ChipsNCheese' analysis suggests Raptor Lake didn't address several of Alder Lake's significant deficiencies. It would be great to see some optimizations of the cache subsystem, for instance.
That would depend on the settings they use, they call it stock but you can see the multi core power draw is above the 253W limit so it's not stock.
www.computerbase.de
Exactly. All I am saying, is I personally saw an 8% uplift at 1/3 the power draw (Both single and multi core). Computer is cooler and it encodes faster, enabled me to move from 20L to 6L case without having to bother with undervolting. Wasn't expecting this much of an uplift going from a 12900k to a 13900.
The main thing that shows is why E-cores were such a good idea (which I've been saying since before Alder Lake even launched). Simply add more cores & clock them a little bit lower and you get way better efficiency. I'm sure if you did the same comparison between i7-7700K and i7-8700, you'd find the same result. In fact, it's the basic idea behind GPUs - make simple, efficient cores, clock them lower, and have lots of them.
Facebook
Twitter
Instagram