• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Intel Core i9-12900K allegedly scores 30K points in Cinebench R23

EsaT said:
Bazillion tiny cores won't make one thread execute any faster.
While some things multithread and can scale to really high core counts, some things don't benefit any no matter how many threads you make.
Basically anything in which next operation depends on previous operation won't do anything with higher number of smaller cores.

And scheduling is going to be nightmare for realizing full performance:
You don't want threads going to heavy weight core eating unnecessary amount of power/thermal budget when not needed.
But if such serial execution of commands needing thread/workload goes to lite core even momentarily that's going to hit performance.
(moving workload between cores has delay)
Also it's really hard to make smaller core without cutting out some instructions/their hardware execution unit.
And emulating such instructions would be really slow compared to hardware, while moving thread back to full core again adds its penalty.

Basically arbitrarily multihreading stuff without any inter thread dependancies, like Cinebench, is the best case.
(also for hiding horrible memory latency)
Click to expand...

Thread director will have a lot to do, or not get too much wrong. Clearly Intels short term strategy is to use a software fix to overcome a hardware limitation.

It worked for Apple, though they didn’t face the same level of challenge Intel do.
 
EsaT said:
...
Also it's really hard to make smaller core without cutting out some instructions/their hardware execution unit.
And emulating such instructions would be really slow compared to hardware, while moving thread back to full core again adds its penalty.

Basically arbitrarily multihreading stuff without any inter thread dependancies, like Cinebench, is the best case.
(also for hiding horrible memory latency)
Click to expand...
This. We saw a similar behaviour with Hyperthreading in some games and software. This notion can also lead to some previous misguided beliefs, like 6 cores 12 threads is much better than 8 cores 8 threads, when the opposite is mainly true.

This is also why Cinebench Single thread score only gives a partial story concerning games and is not the best metric to use to assess the full range of gaming performance.
 
How will overclocking work with this big little setup? Clock the big cores individually or whole cpu offset?
Possible to turn off big cores and see how far small ones will go? 6ghz on little cores?
Interesting times ahead.
 
Dave2150 said:
Source: https://videocardz.com/newz/intel-core-i9-12900k-allegedly-scores-30k-points-in-cinebench-r23

Intel back on top, again with an inferior 10nm (Intel "7") process. Will be fun once Intel's fab's eventually match or exceed TSMC, which won't be that long away most likely.
Click to expand...

30K is not that impressive.... its good, Alderlake looks properly good but you shouting about a 30K score as if that puts AMD back in their gimp box, you're out of your mind.....

5950X. 32,393, the average for the 5950X is also around 30K, and i'm pretty sure Zen 3D will score a lot higher again.

https://forums.overclockers.co.uk/posts/34402096/

If you want to see something truly impressive look at this.

3990X. 83,146.

https://forums.overclockers.co.uk/posts/34225165/
 
It seems to me that this is just a a bit of a gimmick for the fact that at the moment Intel can't manufacture a low cost 16 core chip like the 5950X. I expect Intel's current situation is the reason why AMD went down the chiplet route, multiple smaller dies linked together rather than one huge complex die. I mean AMD even have the I/O as a separate die so if there's anything wrong with it in manufacturing they don't have to throw out or sell for peanuts a perfectly good CPU die.

Outside of applications like Cinebench I'm struggling to see what benefit the small cores will provide? offloading background services perhaps but is that really going to provide a boost to games? in the vast majority of situations outside of things like Cinebench CPU's have many cycles to spare so the smaller cores seem a bit unnecessary.
 
Last edited:
MartinPrince said:
This is also why Cinebench Single thread score only gives a partial story concerning games and is not the best metric to use to assess the full range of gaming performance.
Click to expand...
Single core performance is actually the best indicator for max gaming performance.
Simply because many things in core code of the games don't multithread.
But Cinebench is bad measure for that in high memory latency platforms, because of running only in caches of CPUs.

So my prediction is that Alder Lake's performance is contradictionary:
Very fast in some things, but little or no improvements in other. (maybe even slight regression in the worst cases)
Low memory access latency has been the strong point of Intel for single threaded and gaming performance and now that flies out of the window faster than APFSDS.
And with single monolithic die can't see Intel even affording big caches similar to AMD.
 
mmj_uk said:
It seems to me that this is just a a bit of a gimmick for the fact that at the moment Intel can't manufacture a low cost 16 core chip like the 5950X. I expect Intel's current situation is the reason why AMD went down the chiplet route, multiple smaller dies linked together rather than one huge complex die. I mean AMD even have the I/O as a separate die so if there's anything wrong with it in manufacturing they don't have to throw out or sell for peanuts a perfectly good CPU die.

Outside of applications like Cinebench I'm struggling to see what benefit the small cores will provide? offloading background services perhaps but is that really going to provide a boost to games? in the vast majority of situations outside of things like Cinebench CPU's have many cycles to spare so the smaller cores seem a bit unnecessary.
Click to expand...

Yeah it feels like a bandaid solution in response to not being able to keep up on core counts and running into problems with power consumptions.
 
humbug said:
30K is not that impressive.... its good, Alderlake looks properly good but you shouting about a 30K score as if that puts AMD back in their gimp box, you're out of your mind.....

5950X. 32,393, the average for the 5950X is also around 30K, and i'm pretty sure Zen 3D will score a lot higher again.

https://forums.overclockers.co.uk/posts/34402096/

If you want to see something truly impressive look at this.

3990X. 83,146.

https://forums.overclockers.co.uk/posts/34225165/
Click to expand...
It is quite impressive when you consider this is with an 8 thread deficit to a 5950X while also being 90% faster in multithreaded over Intels previous flagship.

What this will translate to though is strong performance lower down the stack were Intel will have a 4 thread advantage over both the 5600X and 5800X with their equivalently priced CPUs and these are the CPUs most people buy.
 
Joxeon said:
It is quite impressive when you consider this is with an 8 thread deficit to a 5950X while also being 90% faster in multithreaded over Intels previous flagship.

What this will translate to though is strong performance lower down the stack were Intel will have a 4 thread advantage over both the 5600X and 5800X with their equivalently priced CPUs and these are the CPUs most people buy.
Click to expand...

Yeah... :)
 
I am not buying into all these leaks.
To be taken half seriously they at least need to be using W11 and that wasn’t.
My expectations are low and I am in no hurry, so will wait for the NDA to expire.
I do expect the benchmarks to be all over the place with it being such an oddball architecture.
 
Yeah I expect ADL to beat Zen 3 but not crush it, the leaked pricing sort of confirms that.

If 12900K crushes 5950X in everything they will charge $999 for it, why not if its so good?

Since you have 12700K for around $600-700 then going down from there.

All these leaked synthetics so far are not sensitive to latencies and we know how that fared for RKL, with many of these top end boards being DDR5 only, will be interesting if in some cases DDR4 beats them in gaming.
 
Toasty said:
Yeah I expect ADL to beat Zen 3 but not crush it, the leaked pricing sort of confirms that.

If 12900K crushes 5950X in everything they will charge $999 for it, why not if its so good?

Since you have 12700K for around $600-700 then going down from there.

All these leaked synthetics so far are not sensitive to latencies and we know how that fared for RKL, with many of these top end boards being DDR5 only, will be interesting if in some cases DDR4 beats them in gaming.
Click to expand...

5000 series with infinity fabric running 1-1 with DDR4 4000 cas15 T1 will take some beating. The 3D V-cache Zen3 will be very strong across the board, particularly in games.
 
You must log in or register to reply here.
Back
Top Bottom