Discussion UserBenchmark: 4 cores are enough

[Kocicak]

UserBenchmark.com apparently removed 64 thread comparison from their CPU benchmark, maximum is now 8. I guess nobody needs more than 4 cores, right???

 

[amrnuke]

There is definitely no use in having more than 4 cores / 8 threads. We know this is true because Intel have traditionally capped almost all of their mainstream processors at 8 threads.

 

[Kocicak]

Yes, Intel usually knows well what to do. The higher core count processors which they make now are a horrible mistake (a one-off), they should stop making them as soon as possible! Not only nobody really needs such large monsters, they also clog Intel production lines badly.

4C/8T are a sweet spot for 99.9% consumers.

 

[Staples]

I am not sure if these responses are serious But I’d say that not many people use software that takes advantage of more than 4 cores. they Should tone down the test to 32 or 16 threads since I think that is more realistic. I don’t even knnow if there is such a CPU with 64 threads.

 

[Markfw]

I am not sure if these responses are serious But I’d say that not many people use software that takes advantage of more than 4 cores. they Should tone down the test to 32 or 16 threads since I think that is more realistic. I don’t even knnow if there is such a CPU with 64 threads.

I have 2 of them, the 2990wx has 64 threads and 32 cores. I just got a dual EPYC systems that has 128 threads.

For them to do this reeks of somebody buying them off. Either that or actual stupidity.

 

[DeathReborn]

In other news 640KB of memory is all you need...

I have seen 16+ threads being used by consumer (non-profesional) software quite a lot so this does sound like a rather dodgy restriction.

 

[Hitman928]

8T max is ridiculous in 2019. I can understand not wanting to go up to 32T or 64T (even though there is now a consumer 32T CPU), but 12 - 16 thread performance is definitely relevant for many people.

If they want to create categories like light office work, gamer, and prosumer rankings and then limit the light office work to 8T (or even 4T), then fine, but to just put a hard limit on every result at 8T is absurd.

 

[Kocicak]

... put a hard limit on every result at 8T is absurd.

It may seem absurd to somebody wanting to see fair comparison between processors for various use scenarios.

They did it, so for them is must make sense somehow. They are perhaps concerned now just with the "real world performance". Does "real world average Joe" need 64 threads? No. If you think about it, does he even need hyperthreading? Nah.

Imagine the beautifull simplicity of the benchmark, if they used just 1,2 and 4 threads. You would be comparing only three numbers. Something to think about...

 

[amrnuke]

Yes, Intel usually knows well what to do. The higher core count processors which they make now are a horrible mistake (a one-off), they should stop making them as soon as possible! Not only nobody really needs such large monsters, they also clog Intel production lines badly.

4C/8T are a sweet spot for 99.9% consumers.

I was being tongue-in-cheek but generally I don't have a logical problem with them doing what they're doing. I think the sweet spot for most gamers is 8-12 threads, however you want to divvy that up. 8C/8T, 6C/12T, whatever.

In any case, I don't think hampering 8+ thread results is all that interesting, unexpected, or even, technically, all that wrong on UserBenchmark's part, when I think of it logically and consider their target audience.

The reason being that when you are determining the best chip for the "everyman's purpose", that is, for someone searching "3600 vs 9600K" - you have to realize that a person going to UserBenchmark's site as a "source" is someone who is not going to be as tech savvy. They are likely to be just gaming and browsing. and they are likely to be very happy with a 9600K. And if UB's goal is to provide the best service for that audience, then there you go. I found that UB appropriately stratifies chips fairly well for gaming. The 9400F is tiered below the 3600 and 3600X, and the 9600K above them. Similarly, at the higher price point, it appropriately places the 9700K higher than the 3700X, which is consistent with results you see in reviews comparing the two for gaming.

For productivity, what that means for someone using Userbenchmark as the go-to site to compare, which is to say, not a technically savvy person, that probably means browsing, office, and watching videos/movies. Someone using the chip for rendering, game development, etc is probably not going to be relying solely on UB for their decision, as they probably know a little more. And in browsing and office duties, the 9700K is indeed better than the 3700X.

Now, for the group of us who are debating at a high enough level why their stratification is incomplete or misleading, the general idea I have is that... that's the point. Someone who knows enough about the CPUs to make an in-depth decision on the merits of single core vs multi-threading workloads, memory bandwidth, platform differences, whether their workload benefits from SMT or not, and so on -- simply put, we ain't relying on UB in the first place, even BEFORE they started changing their results.

 

[Kocicak]

I was being tongue-in-cheek

Me too, sorry if you wrote your long reply because you believed I was serious.

 

[fleshconsumed]

LOL I upgraded my HTPC to 4C8T 3400G from 2C4T i3 because i3 was getting slow at waking up from sleep and browsing web pages. This is my HTPC machine for chrissake... 4C8T is the new entry level. Maxing out benchmarks at 8T which is what I consider pretty much bare minimum for any real work these days is downright dishonest.

 

[maddie]

8T max is ridiculous in 2019. I can understand not wanting to go up to 32T or 64T (even though there is now a consumer 32T CPU), but 12 - 16 thread performance is definitely relevant for many people.

If they want to create categories like light office work, gamer, and prosumer rankings and then limit the light office work to 8T (or even 4T), then fine, but to just put a hard limit on every result at 8T is absurd.

That guy (Shrout ?) recently from PCperspective, now at Intel, has been agitating for sticking to "real world" usage benchmarks, as defined by him and his team, in comparing CPUs. Here's a win for them.

 

[Topweasel]

Honestly this is a better response than their previous attempts hide their Intel bias by completely devaluing multi core performance till a 9980xe looked worse than an i3. But what they are relying on is that they have a userbase of people looking for CPU's that will check their site that doesn't know what they have done (for whatever "completely legit" reason they can come up with). It now even to Intel fans becomes a bad site to look at CPU performance unless you were specifically trying to talk a person away from a 3600 to an 9400 or 9100. That's the problem its still a tech site and as a tech site most of its viewership is directly pushed by users like us. They are really taking a chance that their traffic doesn't nose dive because of its general irrelevance.

 

[Kocicak]

Remember, that their site comes as a first result when you google CPU vs. CPU. They are an "official Google CPU comparing tool".

When comparing products and being in such position, can they actually afford to present results that somebody could interpret as misleading? Would not they face some legal consequences?

 

[Topweasel]

Remember, that their site comes as a first result when you google CPU vs. CPU. They are an "official Google CPU comparing tool".

When comparing products and being in such position, can they actually afford to present results that somebody could interpret as misleading? Would not they face lome legal consequences?

Probably not. They aren't a marketing company and lies on the internet are not treated like lies on the TV or in Adds. On top of that as long as their designed benchmarks back it up they can do it.

What this does mean is that Intel can now make slides and advertisements using their numbers sell their products.

 

[Kocicak]

... lies on the internet are not treated like lies on the TV or in Adds.

Really? I did not know that.

 

[Topweasel]

Really? I did not know that.

You just need to look into US politics. Again the big thing is that Userbenchmark in itself isn't a marketing site, doesn't offer their site and their results to sell a product of theirs. They are just a Tech review site (with a tech benchmark) with affiliate links. That means that it pretty much is the wildwest on what they can do and the fact that they have a benchmark even a home grown one that they can slide whatever bias around they want backs up their numbers makes them bulletproof there. Doesn't mean they can't get bad plubicity on tech forums, reddit and maybe more from actual tech journalists and have to deal with the weight of the complaints, or take a hit on traffic (which hurts their bottom line).

 

[krumme]

Colgate back in the days used to sell their products simply by saying people should brush their teeth more.

Intels strategy here seems to be: use your PC for less. I mean then you are really on the defensive and thinking short term goals.

 

[Kocicak]

Intels strategy here...

We are not sure yet that Intel owns (or subsidies in some way) UserBenchmark. It just seems like it.

 

[Kocicak]

BTW UserBenchmark put a link to this video:

under the "Incompetent smearers" text on their web site. Calling names on their official website... Interesting.

 

[maddie]

Colgate back in the days used to sell their products simply by saying people should brush their teeth more.

Intels strategy here seems to be: use your PC for less. I mean then you are really on the defensive and thinking short term goals.

I can see it now.

User:
What sort of PC do I need for X, Y, Z.

Intel:
Just as we have moved from mainframes with hundreds of user, then minis with tens, then PCs with one or a couple, now we have the latest and greatest solution.
No more sharing your PC between tasks. Get a separate 8T CPU for each major task needed.

 

[amrnuke]

It now even to Intel fans becomes a bad site to look at CPU performance unless you were specifically trying to talk a person away from a 3600 to an 9400 or 9100.

If you compare the 3600 to the 9400 the 3600 comes out on top on UB -- in gaming, workstation, and desktop ratings. Same with 3600 vs 9100. The only reason you might consider a 9100 over a 3600 is cost -- the 9400F under $100 is close enough to the 3600 in gaming and general desktop use to make it compelling. UB is NOT pushing anyone to a 9400 or 9100 in comparison to 3600.

IMO, the rankings for the broad population of computer users seem generally correct when put up against benchmark results from Anandtech, Techpowerup, Tomshardware, etc. The only caveat is that UB becomes more inaccurate the more rendering, compiling, encoding, encrypting, and compute-heavy work you look at. That's where Ryzen pulls ahead from Intel. It's also a use-case that the majority of "general public" users are agnostic about.

That's the problem its still a tech site and as a tech site most of its viewership is directly pushed by users like us. They are really taking a chance that their traffic doesn't nose dive because of its general irrelevance.

Their traffic will depend on a lot of factors, not the least of which would be relevancy. As long as they correctly stratify the chips for the majority of users, they will continue to be used by those types of users.

So while I personally would NOT recommend an Intel-based system for my purposes (multiple medical record systems open at once, Plex server, JMP/SPSS work, photo processing, MS Office, video/audio editing, light gaming), I do not have any issue with UB's results that place the 9600K above the 3600, or the 3600 above the 9400, or the 9700K slightly above the 3700X -- for the purposes of the majority of people. Again, my theory being that if you're smart enough to be doing the tasks Ryzen is better at, then you're probably smart enough to know not to trust a single source for your data. And even in that case, if all you're doing is gaming, Office, web -- most of the review websites support buying a 9600K over a 3600X if all you care about is "performance now, I don't care about anything else", which you will discover when you do your research.

If you're not the kind of person that even thinks about researching before making several hundred dollar decisions (applied to this situation, that would be at least doing a meta-analysis of several solid reviews) then I highly doubt you're the type of person using multiple medical records, compiling code, doing heavy rendering work, etc. where Ryzen is going to excel.

But we know that a LOT of people don't even research their auto decision, which could be a life-or-death, multi-thousand-dollar decision, or are more concerned about brand and image than actual results. Which is why you see Jeep Wranglers used to commute on inner-city highways at 80mph. People care... but not about the same things as most of us who are more conscientious about things.

All of that being said, UB behaves like a petulant child and they are, at best, a decent chip stratifying website for the lay public.

 

[Kocicak]

Their traffic will depend on a lot of factors

Primarilly on the amount of money they pay Google to be displayed as the first search result?

 

[amrnuke]

Primarilly on the amount of money they pay Google to be displayed as the first search result?

I know this is a half-joke, but having worked in SEO for small businesses, and still having connections with those who do, I just want to clarify that the idea that you can pay Google for top organic search results is, to an absurdly high degree, untrue.

 

[chrisjames61]

I am not sure if these responses are serious But I’d say that not many people use software that takes advantage of more than 4 cores. they Should tone down the test to 32 or 16 threads since I think that is more realistic. I don’t even knnow if there is such a CPU with 64 threads.

Have you been living under a rock?