A new way to address CPUs (Cores/Threads) on HWBot

[yosarianilives]

It's relevant for fx because we're making categories for cpus with different core strengths in different core configs. if fx shares part of its instruction pipeline between two cores, not just cache, then creating a one core per module category is no different from creating a 12900k category for disabling all the "shit cores" (they're Skylake speed, stop acting like they're super slow guys). 

 

I still think that just adding up all the cores is the best, if Intel says it's 16 core then it's 16 core. This approach was used when people wanted fx to be 4 core with ht historically, the decision was if amd says its 8 cores then it's 8 cores. Of course historical decisions don't have to be future decisions as undeniably some historical decisions are probably not the way to move forward. 

[Leeghoofd]

We agree to disagree,  I see AL as 2 different CPUS/cores on the same die... FW was a genuine idea but AMD just messed up their module config with Bulldozer and Vishera, on top of that there was the lack of a proper OS to support the idea/goals of the design team.

With option 4 we can highlight the strength of the big cores and the weaknesses of the entire core bundle as the Skylake IPC will not be able to match latest CPU gen releases. No clue what FX can/could show besides raw mhz

[Splave]

I think the first "non-homogeneous" core CPUs (aka no-homo CPUs) that will compete at the top there will be some growing pains. I think people just want to hear if there is a mass exodus hwbot is willing to possibly change

Edited August 23, 2021 by Splave

[3urner]

Maybe it also needs to be mentiond that quad threading cpus will become a thing in the future. So 1 core with 4 threads also needs to be handelt by the system. 

Edited August 25, 2021 by 3urner

[TaPaKaH]

Now that mainstream CPUs from both sides get 16 cores, wouldn't it be the right time to ask ourselves if we actually need to split CPU globals per number of cores?

Back when the split was introduced, there were maximum three relevant classes (1,2,4) with a HUGE difference in between. But these days you have at least fourteen (1/2/3/4/5/6/8/10/12/14/16/18/24/32) and it doesn't look like it's going to get any simpler with all the 'exciting' ways that manufacturers can customise their CPU layouts.

It's starting to get very similar to situation with GPU rankings:

• CPUs have 2/3-digit core count, GPUs have 2/3-digit ROP count
• Given core count can be achieved by 1 or 2 cpus (or even core types), given ROP count can be achieved by 1 or 2 cards
• It's very rare that nVidia and AMD release graphic cards with the same ROP count (or even at the same price point) to avoid direct comparisons, Intel/AMD will likely eventually get there too

...and I don't see anyone discussing splitting GPU ranks per ROP count "to give guys with GTX3060 a chance and make it more affordable"

[der8auer]

I guess eventually it might make more sense to split globals in few categories such as Mobile, Desktop, High End Desktop and Server similar to what Rauf already suggested. I'm very open to such solutions but as I already pointed out we will need quite a lot of time to make this happen.

Regarding AMD FX I have no problem changing this to 4 Cores if this makes the community happy. Probably won't change anything tho?

 

 

[Noxinite]

9 hours ago, der8auer said:

I guess eventually it might make more sense to split globals in few categories such as Mobile, Desktop, High End Desktop and Server similar to what Rauf already suggested. I'm very open to such solutions but as I already pointed out we will need quite a lot of time to make this happen.

Regarding AMD FX I have no problem changing this to 4 Cores if this makes the community happy. Probably won't change anything tho?

There was a lawsuit that decided that there is no official definition for a core, so AMD could market it as 8 cores even if it shares resources in pairs. IMO just list it as AMD marketed it.

[yosarianilives]

18 hours ago, Noxinite said:

There was a lawsuit that decided that there is no official definition for a core, so AMD could market it as 8 cores even if it shares resources in pairs. IMO just list it as AMD marketed it.

Same for Intel

[Matsglobetrotter]

to me the changes proposed sounds like a Sponsor wanting to change the scoring to make them look better behind the scenes.  core is core thread is thread. and if multicore is not competitive compared to old generation but it beat the crap out of a single core score then that is what it is. Very sad moment when we try change a fundamental basis of he scoring. Indeed no processors for with equal threads to cores has been competitive for years when they should have been. We want performance on both single core and multi core and multi thread. tough luck if the new processors cant do it right. I think that would reflect the reality of the trickery.

[der8auer]

5 minutes ago, Matsglobetrotter said:

...core is core thread is thread....

Welcome to 2021 where core is not core anymore or why do you think I started this discussion? Thanks for your useless input

 

[Matsglobetrotter]

3 minutes ago, der8auer said:

Welcome to 2021 where core is not core anymore or why do you think I started this discussion? Thanks for your useless input

 

Great the user community give useless input. Thanks you very much... appreciate the comment.

[Matsglobetrotter]

I suggest  admin accept comments as they are without blasting a user. it normally create better open discussion among the community. there are like/dislike buttons. Meanwhile I'm sure it was just a heated moment while trying to resolve something for the bigger community. Keep pushing forward i know the intent is good.

Cant wait to finally get all my gear through customs to start benching... in a month...

[Rauf]

Always conspiracy theories here... I'm sure hwbot points are extremely important to intel...

Big.little makes so much sense from a practical point of view. First generation with something new is not always super, we'll just have to wait and see. but I'm sure this is the future. So we can either stick our heads in the sand and compare performance in an irrelevant way, or adapt to something that makes sense. For now I think the best way would be to create all new categories, like 8+8, 6+6 or something. Then see how everything goes and make a bigger overhaul when there are more facts and there is time for development.

Edited August 26, 2021 by Rauf

[Leeghoofd]

yep fully correct summary Tobias, it is something completely new. We will have to see how it develops (ranking/performance wise) and will need to learn/adapt in the process....

I don't see the point in considering it simply as a 12 or 16 core... And these conspiracy theories do my head in. Again Team Red will go the same path so we better learn and adapt now ( if needed by making corect or incorrect decisions/mistakes right now)

[Splave]

should we consider that AMD 5950x is 8+8 even though they are matching cores, and 8+8+8+8+8+8+8+8 etc for thread ripper? Where do we draw the line. 

[Leeghoofd]

Matching cores old style, mixed cores like with AL/new AMD, split up in standard ranking and one for the performance cores... thats a clear line...

 

 

 

must resist to feed the troll, must resist...must resist to feed the troll,must resist to feed the troll,must resist to feed the troll

[3urner]

14 hours ago, Rauf said:

For now I think the best way would be to create all new categories, like 8+8, 6+6 or something. 

In my opinion thats the best and only correct solution. The old categories will still remain but most likely get more irrelevant over the years as the new cpus are X+Y and this category gets filled more and more. 

[aperacer]

Maybe the new CPU´s should stay in their own ranking, like 8+8 or 6+4 and should not compete in the old rankings by disabling the E-Cores.

17 hours ago, der8auer said:

Welcome to 2021 where core is not core anymore

 

Intel and AMD are going this way, because this is the future for them and for us.

[Splave]

I actually wasn't trolling and my point couldn't be more relevant to this discussion. AMD is doing this already but the cores just match. Makes it even more relevant when you clock the cores separately like we will be with alder lake. None of the top 5950x scores using matching clocks, I know because I have them. 

Close the thread if the decision is already made. It seems to be giving the false sense of value to other opinions than the already chosen decision.  The people that run the show should decide anyways as there is always a spilt in what should be done.

 

❤️

[I.nfraR.ed]

Let me stir it a bit more.

Big.LITTLE doesn't mean just two complexes of equal cores anymore.

Just one example (yes, first 4 cores are still A78, but one is clocked higher and has twice the L2 cache). They probably can't be overclocked by the user. Would a similar desktop CPU able to compete in the single-core global category if it is possible to disable the rest of the cores?

MediaTek MT6893 Dimensity 1200 5G (6 nm)

1X Ultra Core: Arm Cortex-A78 up to 3GHz with 2X L2 cache 3X Super Cores: Arm Cortex-A78 up to 2.6GHz 4X Efficiency Cores: Arm Cortex-A55 up to 2GHz

Mediatek calls it 8-core. There are also other models with 2 high performance cores + 6 efficient cores = still a octa-core CPU.

Edited August 27, 2021 by I.nfraR.ed

[Niuulh]

On 8/25/2021 at 4:16 PM, der8auer said:

I guess eventually it might make more sense to split globals in few categories such as Mobile, Desktop, High End Desktop and Server similar to what Rauf already suggested.

Sound like the more convenient way imo.

Less Global categories could be a good thing considering it will become a mess to clearly define CPUS. Would be easier and i think less prompt for drama...

Instead of having dozens of global categorize, maybe rework/balance or encourage users to modern hardware points, as currently sub are rare for CPUs/GPUs other than the biggest of said generation.

[Noxinite]

I kinda like how 1x, 3x and 5x are "legacy" global so to speak, but that's a bit off topic...

[No1Spank]

Would it not make sense to keep global score for total cores the same and then add global points for overall thread count next?  I suppose people would re-bench everything with HT switched off to get some more points.

Then if that works out could you not add global points for the type of cores that are active for these new mixed CPU's.

People could get points twice on the mixed chips as they would get points for both types of core.

[CENS]

If 12900k has 16 cores, treat it as a 16-core cpu. Period. If Intel decides to make 8 of them stronger and 8 of them weaker,  have HT on one and not on the other, thats then Intel's decision. From what you hear it even performs like a "true" 16-core 5950x. So what's the big deal.

Also 12900k is monolithic. For example the 5950x isn't even monolithic and has chiplets. Can I disable chiplets and make it join the 8- or 4-core category? No I can't. 

So if you have a 12900k you can join two categories and essentially gain double the points in some way, that's just weird and complicated.

Allowing disabling part of the chip so it can joins a different category so it looks stronger is contradicting previously made statement regarding disabling cores for frequency validation. ppl can disable all they want it is still an 16-core chip, and therefore it should get treated that way eventhough some cores are stronger than others, some are more efficient than others. It's the overall performance that counts. Regardless of the amount of threads right?

When Intel comes out with with quad HT, what do we do then? It's still X amount of cores but 4-times the amount of threads. The way I look at it is the individual cores just got more powerful/efficient. But plz don't make it more complicated with dividing CPUs into multiple categories.

Please correct me if I'm wrong but we do we differ between a 6-core with HT and one without? 

Edited September 3, 2021 by CENS

[Strunkenbold]

If we choose the option where the user needs to disabling cores, I would only let the performance cores result count. It would be unfair for a CPU to have the possibility to obtain 2 times global or hardware points. Also we need to consider that if we change our decision at some later point and decide that those processors can make use of their full native core configuration, all core disabled results are getting pretty useless. So we might want make some conservative decision now like the 16 cores are 16 cores approach and better implement something smart when there will be time for.

An, probably, over-complicated approach would be to create benchmark categories for the individual core configuration. I just took a look at the Snapdragon SoCs and there would be four possible combinations for 8 cores:

8P
4P + 4E
2P + 6E
1UP+1P+6E

So that means there would be an additional 4 + 4, 2 + 6, 1 + 1 + 6 Cinebench ranking for instance. Every user could compare apples to apples then. However, I would not give global points for each config as would be an insane inflation of points because of dozens of new categories. Instead we should group those configurations in an unified ranking. The easiest way would be to just count all existing cores again. So in the above example all those configurations go into 8 cores ranking. IMO that would be fair. No one tells that you have to win with a CPU featuring 50 % efficiency cores over a CPU with 100 % performance cores.

However if we want to change something, another approach would be to divide the amount of efficiency cores by two.

4 + (4/2) = 6 cores ranking (for Intel this would also mean 12 threads vs 12 threads, doesn't matter for ARM however)
2 + (6/2) = 5 cores ranking
1+1+(6/2) = 5 cores ranking

Wouldn't be to hard to configure those new CPUs with some cores less in the db. So really a quick and dirty solution. Biggest concern is when efficiency cores are getting better than Rocket Lake and Zen3 because of architectural improvements or changes like added HT or Quad HT. 

To pick up the idea with splitting categories into low, mid, high and high end we might could come up with grouping existing categories in an unified global points ranking. Like

1 core
2 cores
4 cores (consisting of 3 core ranking and hybrid variations)
6 cores (consisting of 5 core ranking and hybrid variations)
8 cores (also including possible 7 cores and hybrid variations)
12 cores (anything from 9 to 12 and hybrid variations)
16 cores (anything from 13 to 16 and hybrid variations)
24 cores (anything from 17 to 24 and hybrid variations)
32 cores (anything from 25 to 32 and hybrid variations)
64 cores (anything from 33 to 64 and hybrid variations)

Low core count categories are kept for historical reasons. 3 and 5 cores are getting killed as they are only Phenom and doesn't qualify for globals anymore IMO. Anything from 64 is server stuff and hasn't much to do with oc. It shouldn't be awarded just because someone works in a data center.