Hacker News Reader

Comment by kllrnohj | Hacker News Reader

Well, two things there.

1: Userbenchmark.com is terrible and nobody should use it for anything. At least their CPU side of things is hopelessly bought & paid for by Intel (and even within the Intel lineup they give terrible & wrong advice), maybe the GPU side is better but I wouldn't count on it.

2: The real question there isn't "why is the 1050 Ti not faster?" it's "how did you run a 1050 Ti on MacOS in the first place, since Nvidia doesn't make MacOS drivers anymore and hasn't for a long time?"

View on HN

> Userbenchmark.com is terrible and nobody should use it for anything. At least their CPU side of things is hopelessly bought & paid for by Intel (and even within the Intel lineup they give terrible & wrong advice), maybe the GPU side is better but I wouldn't count on it.

To provide some elaboration on this: Their overall CPU score used to be 30% single, 60% quad, 10% multicore. Last year around the launch of zen 2 they gave it an update. Which makes sense; the increasing ability of programs to actually scale beyond four cores means that multicore should get more importance. And so they changed the influence of multicore numbers from 10% to... 2%. Not only was it a blatant and ridiculous move to hurt the scores of AMD chips, you got results like this, an i3 beating an i9 https://cdn.mos.cms.futurecdn.net/jDJP8prZywSyLPesLtrak4-970...

And there was some suspicious dropping of zen 3 scores a week ago, too, it looks like.

I don’t see that as evidence of blatant bias for Intel. The site is just aimed at helping the average consumer pick out a part, and I think the weighting makes sense.

Most applications can only make use of a few CPU-heavy threads at a time, and these systems with with 18 cores will not make any difference for the average user. In fact, the 18 core behemoth might actually feel slower for regular desktop usage since it’s clocked lower.

If you are a pro with a CPU-heavy workflow that scales well with more threads, then you probably don’t need some consumer benchmark website to tell you that you need a CPU with more cores.

But lots of things do use more than 4 cores, with games especially growing in core use over time. Even more so if you want to stream to your friends or have browsers and such open in the background. To suddenly set that to almost zero weight, when it was already a pretty low fraction, right when zen 2 came out, is clear bias.

> In fact, the 18 core behemoth might actually feel slower for regular desktop usage since it’s clocked lower.

It has a similar turbo, it won't.

The amount of processes running on a windows OS reached 'ludicrous speed' many years ago. Most of these are invisible to the user, doing things like telemetry, hardware interaction, and low level and mid level OS services.

A quick inspection of the details tab in my task manager shows around 200 processes, only half of which are browser.

And starting a web browser with one page results in around half a dozen processes

Every user is now a multi-core user.

The one that really made it screaming obvious is the description of the Ryzen 5 5600x still somehow recommends the slower 10600k https://cpu.userbenchmark.com/AMD-Ryzen-5-5600X/Rating/4084

And they added some "subjective" metric so even when an AMD CPU wins at every single test, the Intel one can still be ranked higher.

There's a reason they've been banned from most major subreddits. Including /r/Intel.

Why should I care about a subreddit? They are all probably moderated by the same poeple. It could be one person got offended and happens to be a mod

Re #2, the Nvidia web drivers work great if you're on High Sierra

Regarding 2. I think that none of those benchmarks were run on MacOS. Their benchmark tool seems to be Windows-only https://www.userbenchmark.com/ (click on "free download" and the MacOS save dialogue will inform you that the file you are about to download is a Windows executable).

The gfxbench link in the OP of the M1 vs. GTX 1050 Ti days that the 1050 was tested on MacOS. That's what I was referring to.

1. Today I learned something new. Still, can’t let great be the enemy of good. It may be imperfect but it’s the source I used. Do you have a better source I can replace it with?

2. That’s a good question and I don’t have an answer for that.

Sure, great is the enemy good etc , but the allegation here and in other threads is that these benchmarks are bad. Or, worse, inherently and deliberately biased.

As for a better source, I don't know with the M1 being so new, but that's no reason to accept bad data, if this benchmark actually is as bad as others here are saying.

Reply on news.ycombinator.com