Brain-storming: How to show CPU performance in CPU Database

[W1zzard]

I'd like to add some sort of performance indicator to our CPU database, but not sure how/how to present.

As we all know, single-threaded performance is vastly different to multi-threaded, and gaming performance is even different to those 2. Also GPU-limited matters

Any ideas how to present something that's easy to understand for the average Joe coming from a Google search? and who just wants to know "what's faster?" or "what should I buy"?

 

[Ferrum Master]

It should have CS bench at FHD and as baseline with some very popular GPU. Because of the popularity and userbase.

 

[Vayra86]

Perhaps combining results on these three metrics (single threaded perf, multi threaded perf, and gaming) in a single ranking is worthwhile? Give people a sorting option so they can get specifics as well; if I want the fastest gaming CPU I can click to sort on gaming and poof its on top, etc

And then you can cross check that value with the average price/ MSRP to get a bang/buck rating for every purpose of every CPU.

 

[HD64G]

Night thought @W1zzard! In your shoes, I would choose a single app that is open code and doesn't get affected by corporation pressure to its devs. Maybe y-crancher-Pi or something like that? Anything that fully utilises many cores, high clocks and cache in order to show anything better that a cpu has that increase performance when compared to others. And ofc I would put that as a discaimer with bold: That performance ratio is based on testing the cpus on the X app only and it is clearly just an indication of their performance. Other apps might utilise cpus differently and offer other ranking between those cpus. Just an idea. Oh, and single core performance ranking in 2023 is totally useless imho. As for gaming potential (I wouldn't care much personally), I would just use the timespy score or any game that stopped updating and isn't CPU bottlenecked @1080P.

 

[Konyalee]

Honestly I am not qualified enough to tell you what to do but Paradox Interactive games that I play (Europa Universalis 4, Hearts of Iron 4, Crusader Kings 3 and Victoria 2) are single threaded games that aren't GPU limited even at 4K.

Maybe I am ranting about lack of testing games I play and that's selfish but when I was choosing my 7950X3D, I couldn't find a single benchmark that shows how 3D V-Cache affects performance so I couldn't decide between 7800X3D vs 7950X3D vs whatever intel has to offer (don't really know how intel software works so I avoided when it was possible). My only idea is maybe make a seperate testing for "Single Threaded Games FPS".

 

[Assimilator]

Average Joe isn't going to care about anything except gaming, so I'd say just take a spread of the of the most popular multiplayer games (CS:GO, Fortnite, Rocket League, whatever else the kids play today), plus some of the most demanding ones (Cyberpunk at least), benching all these at 1080p, 2K, and 4K, then normalizing the results into a single aggregated score. Keep it simple, and importantly easy for you to maintain.

don't really know how intel software works so I avoided when it was possible

What does this have to do with anything?

 

[W1zzard]

just take a spread of the of the most popular multiplayer games (CS:GO, Fortnite, Rocket League, whatever else the kids play today), plus some of the most demanding ones (Cyberpunk at least), benching all these at 1080p, 2K, and 4K, then normalizing the results into a single aggregated score.

For all the 3197 CPUs that we currently have in the database?

 

[TumbleGeorge]

Average Joe isn't going to care about anything except gaming

What? I am average Joe and interested of time of rendering of something.

 

[Vayra86]

What? I am average Joe and interested of time of rendering of something.

No you are tumble George

I get your point though. If its going to be gaming only make it a gaming thing, if not the focus should be on more than just gaming perf

 

[Assimilator]

For all the 3197 CPUs that we currently have in the database?

Wait, are you looking for an algorithm on how to compare CPUs on specs i.e. without raw benchmark data for all of them, similarly to what you do for GPUs?

 

[W1zzard]

Wait, are you looking for an algorithm on how to compare CPUs on specs i.e. without raw benchmark data for all of them, similarly to what you do for GPUs?

Yeah, no other way I can imagine this to be feasible

 

[freeagent]

Maybe with GFlops?

But with 3K CPUs that would be a terrible chore.

 

[QuietBob]

IMO the indicators for ST and MT performance should be simple numbers that anybody can quickly verify and relate to. Rather than showing averages from a number of commercial apps, I'd suggest using two classic benchmarks: CPU-Z for single and Cinebench R23 for multi. However, I'm not sure whether presenting the results from proprietary software would be viable from a business perspective.

As for "gaming performance", I believe that approximating it with a number is quite impossible. There are way too many variables at play here. Instead, I'd suggest grouping the CPUs into performance brackets or tiers. Why not start with a modern 4c8t SKU as the baseline and then extrapolate from here? This ranking should take into account ST performance, cache size and -- to a degree -- the number of cores.

 

[XL-R8R]

My suggestion is to put this mental and physical effort into something more constructive; forum changes, social media presence... or just take some time off @W1zzard?


I see this being a difficult task due to the amount of variables in CPU's spanning multiple generations - regardless of how you sort/organise them all - and it'll result in a 'messy' chart with various confusing subcategories at the very least.

Unless you stick to a scientific regime of accuracy by keeping all test conditions as fair/closely matched as possible for the 3197 CPU's in question - which would no doubt require a complete retest of all CPU's mentioned - I don't see this as feasible in a practical way.

 

[Solaris17]

separate chronologically by arch.

Then list by by base clock and max turbo freq?

 

[Squared]

The specs are pretty meaningless, in my mind. The 5800X3D is way better at gaming than the 5800X, but I don't think even AMD expected that until they tried it. The 13900K has way less cache than the 7950K but it games better. I think it'd be better to just include a link to the processor review. Or use an aggregate of all your benchmarks. Turn the gaming relative performance on each processor into a point score and use that, and do it in such a way that the score only changes a little as you swap out games in the benchmark one at a time, so any two processors less than 2 years apart will have a mostly-accurate relative score. Aggregate all the other benchmarks as well into a web score and a productivity score.

 

[Eskimonster]

Defo should be Open source program used, when thats settled. The perf should combine singlethreaded workload and multithreaded workloads, lastly a powerusage either over time or pur clockcycle.

 

[Vayra86]

The specs are pretty meaningless, in my mind. The 5800X3D is way better at gaming than the 5800X, but I don't think even AMD expected that until they tried it. The 13900K has way less cache than the 7950K but it games better. I think it'd be better to just include a link to the processor review. Or use an aggregate of all your benchmarks. Turn the gaming relative performance on each processor into a point score and use that, and do it in such a way that the score only changes a little as you swap out games in the benchmark one at a time, so any two processors less than 2 years apart will have a mostly-accurate relative score. Aggregate all the other benchmarks as well into a web score and a productivity score.

Yeah there is also the core count conundrum wrt games. Some generations really do need to be beyond 4 cores, newer ones not so much.

 

[R0H1T]

In some ways this is easier than GPU's though it would still need a lot of time & effort, you can use public databases like hwbot for older CPU's & most of them would scale almost perfectly wrt cores & certainly frequency. Can put them as an estimate & the rest, which are tested, can be shown having real world data like for CB23, AIDA64, Geekbench et al. But would need lots of time & effort for even one set of tests like Cinebench.

 

[Fouquin]

Maybe y-crancher-Pi or something like that?

Y-Cruncher scores are heavily influenced by DRAM. The same chips with DRAM at different clocks produce vastly different results.

It's a tough problem to solve. Aggregate benchmark results would be one solution as it would weed out the strengths/weaknesses of architectures, but how do you get those scores without months of manual input? Approximations could be made for many classic architectures like K10, where Core\HT\NB clocks*cores=perf, but once we start getting into the modern chips that have varying core configurations with entirely different architectures inside, multiple different busses, mixed instructions... It's a herculean task at best.

 

[evernessince]

I'd like to add some sort of performance indicator to our CPU database, but not sure how/how to present.

As we all know, single-threaded performance is vastly different to multi-threaded, and gaming performance is even different to those 2. Also GPU-limited matters

Any ideas how to present something that's easy to understand for the average Joe coming from a Google search? and who just wants to know "what's faster?" or "what should I buy"?

A radar graph would be a perfect fit I think:

It's perfect for visualizing data including the differences. between products In the graphic above for example, the red line could represent CPU A and the blue CPU B. You can adjust the number of points based on the number of metrics you want to measure, although this does work best with 5 or more metrics. Just off the top of my head: Single thread, multi-thread, Application performance (mixed workloads), game performance, energy efficiency. Definitely others you can add as well, price or performance per dollar for example.

 

[johnspack]

I think you'll need to have to include both single and multi-threaded performance numbers. For those who want gaming performance, they probably want to look at single threaded
performance. For those looking for workstation type performance, they would want the multi numbers. My old xeon 2699 for instance... horrible single threaded performance.. but
multi isn't too darn bad! I personally use a variety of benchmarks to compare performance... but not really sure how you could make it simple....

 

[evernessince]

I think you'll need to have to include both single and multi-threaded performance numbers. For those who want gaming performance, they probably want to look at single threaded
performance. For those looking for workstation type performance, they would want the multi numbers. My old xeon 2699 for instance... horrible single threaded performance.. but
multi isn't too darn bad! I personally use a variety of benchmarks to compare performance... but not really sure how you could make it simple....

The problem with this is that gaming performance isn't represented accurately by single thread or multi-thread. For example, the 7800X3D is worse than the 13900K in both those categories but pulls ahead in gaming performance. Gaming performance needs to be it's own separate thing.

 

[OneMoar]

3 runs of cinebench (both single and multi core) to heat load
averge the 6 numbers put it in a table one for single one for multithreaded
done
gaming performance is irrelavent at anything over 1440p

 

[pavle]

How about a "potential" index for all 3197 CPU's cores x threads x frequency with adjustments for instructions higher than SSE 4.1 and "3D" cache?
It can be added - the CPU can achieve said potential when all its cores and features are properly used.
A quick idea for avoiding re-testing 3k CPU's.