[kllrnohj]

The specific issue is how userbenchmark weights CPUs seems to be very coupled to Intel's specific ideas on how many CPU cores there should be, or at least very firmly stuck in the 5 years ago. So last year the weights were 40% for single-core performance, 58% for quad-core, and 2% for "multi-core".

The idea was this was supposed to be what games care about, but it isn't. Modern games have issues with even 6c/6t CPUs, such as the horrible 1% lows on the 9600K in Far Cry 5: https://www.gamersnexus.net/hwreviews/3407-intel-i5-9600k-cp...

It looks like Userbench has since adjusted to weight up to 8 threads of performance? Which is maaaaybe less trash if all you care about is gaming. But the Core i5 series still tops charts on userbench despite reviewers no longer recommending the i5's due to performance inconsistency.

They even make ludicrous claims like that the 9100F is perfectly fine for gaming, and is even 10% better than a 2700X. They seem to be basing this decision entirely on older games or games specifically built for as broad a userbase as possible (eg, CSGO, Fortnite & Overwatch). Meanwhile actual reviews say things like "The quad-core Core i3-9100F was hopeless in Battlefield V, pretty bad in Assassin’s Creed: Odyssey, fairly useless in The Division 2, and weak in Shadow of the Tomb Raider." https://www.techspot.com/review/1983-intel-vs-amd-budget-cpu...

So even if you're an Intel fan, userbench is still a terrible way to pick a CPU.

[paulmd]

HWUB is not known for being particularly even-handed in their editorial positions. They tend to 'beg the question' by picking game suites that produce the outcome they want to discuss, and tend to over-reach on the conclusions.

"fairly useless" here is over 60 fps average in the heavy titles and 90-110 fps in the multiplayer titles, with a similar ratio of minimums as the 1600 AF (so no more or less prone to stutter). And that's with them loading the dice by picking the absolute most thread-heavy games they could find, most games the 9100F does comparatively much better than that.

And the reality is that Zen1 and Zen+ actually are pretty weak in gaming. Zen2 made a ~30% improvement over Zen1 in gaming performance (much better than the "average" gains for other workloads), and it's still 10-15% behind the fastest Intel processors. Zen1 especially was hot garbage in gaming, those thread-heavy titles aren't representative of its average performance. About all you can say is that it aged better than the 4Cs that Intel had on the consumer platform at the time (or the 8100/9100F/etc that followed), an OC'd 8700K lays a smackdown on it and an OC'd 5820K remains extremely viable even today.

I'm not going to defend userbenchmark's composite scores, but gaming performance does heavily depend on per-core performance even today. Having 8 faster cores is still more desirable for gaming than 16 slower cores. And single-core performance is a good analogue of "per-core performance" so this number remains very relevant.

[kllrnohj]

> "fairly useless" here is over 60 fps average in the heavy titles and 90-110 fps in the multiplayer titles

You missed the point. The point was UB claimed the 9100F was 10% faster than a 2700X. In reality the 2700X absolutely massacres the 9100F in gaming performance. Higher average FPS, higher min FPS, etc...

Even the 1600AF trivially beats the 9100F.

> with a similar ratio of minimums as the 1600 AF (so no more or less prone to stutter).

1600 AF in Battlefield V: 126 average, 91 1% lows

9100F in Battlefield V: 116 average, 49 1% lows

That's not a similar ratio at all.

> Zen2 made a ~30% improvement over Zen1 in gaming performance

No it didn't. You're massively misrepresenting (or mis-remembering) Zen1's gaming performance.

https://tpucdn.com/review/amd-ryzen-7-3700x/images/relative-...

3.6ghz/4.4ghz boost 3700X is ~11% faster than the 3.6ghz/4ghz boost 1800X in 1080p gaming.

Even at 720p it's a 15% gap between those two, not 30% https://tpucdn.com/review/amd-ryzen-7-3700x/images/relative-...

> Zen1 especially was hot garbage in gaming, those thread-heavy titles aren't representative of its average performance.

No it wasn't. It lost to the equivalent Intel CPU, but it was far from bad. You could easily pair a Zen1 CPU with just about any GPU and never see a significant bottleneck. The exception being the absolute top-end. And, critically, if you had an older Intel quad core, like a 7600K, the Zen1/Zen+ CPUs were still an upgrade in gaming performance.

See for example at 1440p the gap between Zen1 & Zen2 & Intel being almost nonexistent even with a 2080 Ti: https://tpucdn.com/review/amd-ryzen-7-3700x/images/relative-...

> Having 8 faster cores is still more desirable for gaming than 16 slower cores.

Of course, but you still need enough cores to avoid stuttering. Which means...

> And single-core performance is a good analogue of "per-core performance" so this number remains very relevant.

Is not correct at all. Single-core performance isn't an analogue of anything these days. You need a minimum number of cores and good single-core performance.

And it's not just HWUB with these conclusions that an i5 is no longer sufficient. Gamersnexus has the same recommendations: "In more games each year, we’re noticing the cut-down Core i5 exhibiting high frametime variability that counteracts its fleeting performance superiority with unreliable, stuttery behavior. The AMD R5 3600 is more reliable and consistent in its performance across all games we’ve tested, making it the better gaming option." ( https://www.gamersnexus.net/guides/3533-best-cpus-of-2019-ro... )

[Dylan16807]

> So last year the weights were 40% for single-core performance, 58% for quad-core, and 2% for "multi-core".

And for reference, they used to have it at 30% single core, 60% quad core, 10% multi core. But that didn't advantage Intel enough, or something.

[ambicapter]

> Modern games have issues with even 6c/6t CPUs

Any idea why?

[paulmd]

It's really specific to Far Cry 5, and it produces some quite strange results that people are putting way too much weight on.

0.1% lows really tank on FC5 on processors without SMT, for example a 5.2 GHz 9600K has less than half the 0.1% FPS as a stock Pentium G5600 2C4T processor. In other words it's stuttering on the 9600K but running ok on the G5600.

https://www.gamersnexus.net/images/media/2018/cpus/2600k/int...

4C4T processors (R3 1200) do OK, but that one is AMD, so it isn't clear whether it's specifically something the engine is doing wrong around Intel processors, or if there's some hardcoded assumption that if there are 6+ cores then SMT must be available, or what.

But I mean, this specific game is not evidence that "6C6T is no longer sufficient for gaming", it's just a badly programmed game that has something going wrong under the hood on 6C6T processors.