Passmark releases PerformanceTest v10 benchmark, adds AVX512 - Receives Critique for favoring Intel

Bah, skip unsupported test, calculate total score by average tests score.

Or does AVX512 provides super high score that boosts average score?

 

Ah Intel, back to your old shenanigans hey?

 

Please don't judge a very complex situation from a rant tweet, it's not fair.
The v10 beta was available since last October... lots of people complaining how much was important for their business but actually nobody took care to support the development.
That's what you get when you don't care.
For sure there's margin of improvement and probably some mistakes but just blaming a small team trying to do the best it's childish.

I do suspect the one to blame is Microsoft with their VS2019 compilers; they are probably not optimized for Ryzen 3 as they should.

 

Intel(ligence) Corporation right now: they are smart as a empty bottle of water.
Anyway,who cares about a synthetic bench?Benchies in games matter all the time.

 

Not sure what's wrong with using the available features of a cpu to make it go as fast as possible?

 

Whole problem of such tweet is that are half true at best.

Its simply new code, so nobody messed with old results that is important.. are new result different yes, is this strange yes? Should they look in to it yes, should suck results pass some internal testing and go public without some lengthy explanation, how its possible (it its invaliding sense of old result) - NO. Its Passmark to blame yes.. did they such thing intentionally? I doubt it, they will loose lots of credit.
Do remember that professional benchmarking company, which did some benchmarking fo Intel and disscuss it with Gamers Nexus? - That was different animal..

 

Oh dear, $ntel, to expand the use of your "$3 billion set aside for anti-competitive practices" does you no favour whatsoever. Hopefully - for the benefit of consumers - you'll find your rightful position, soon.

 

The point is what the benchmark aim to do.
If you write specific load for some specific instruction, that is fair, as long as those instructions have a practical application in the range of situations for which the benchmark is supposed to measure something.

A wrong measurement would be for example create a benchmark for system dedicated to MS office and then use the GPU as a score indicator.

So more in particular if the score is a `generic cpu score` yes you should use everything that is available, is the user then that knows if that benchmark is reliable or not.

 

The AVX-512 instruction set does not take part at all in the single threaded test.
It does have small impact in the multi threaded score but it's not the root cause for this big change in scores.
In my humble opinion currently it's not reflecting anywhere close the real performance delta between Ryzen 3xxx and i9xxx.
Not that I care so much since I'm not using it and not planning to either.

I agree as said above; it's a synthetic benchmark, if you don't know what it means don't look at it.
Bench your workload; gaming, scientific, encoding, etc

 

I don't see the problem here. First of all, synthetic benchmarks are pretty stupid to begin with and I don't understand why there are so many of them. They say very little about real-world performance. But even if this was a real-world benchmark, so what if Intel runs better? Intel put in the effort for these instructions and AMD didn't. Frankly, AMD doesn't really need to, because practically nothing depends on AVX512 (let alone AVX2) and I don't see that changing any time soon. Since these benchmarks serve no function other than bragging rights, wouldn't it be nice to know the theoretical performance boost of AVX512?

 

Isn't that kind of a chicken and egg problem though? Most people didn't expect AVX to catch on at all in processors and now some games require it. Up until recently AMD didn't have an AVX2 processor - so perhaps some workloads will start utilizing/requiring it, same with AVX512.

Also the article states that Intel processors that don't support AVX512 are getting large boosts in scoring.. so the entire update seems to favor Intel more than AMD. You can imagine at least some number of sales are based on these benchmarks and if the benchmarks are incorrectly skewing heavily towards one vendor over another then I think it is a big deal. The only way most people are going to know is if media/users call it out. It's not like everyone reads Guru3D forums all day or compare and contrast real world workloads with what they see with a quick google search. I also feel like bad press tends to keep these companies at least somewhat honest.

 

I actually used them a lot to quickly compare CPUs and GPUs

But yeah I won't use this page anymore. I want stable benchmarks. Benchmark stats that are changed from one day to another, are completely useless and untrustworthy and Passmark just joined this list to me.

Do you have any page to recommend as an alternative? I tried this, but I is not a good alternative for me: https://www.phoronix-test-suite.com
- No standalone official .zip to just execute and use
- It requires CMD/Powershell to be installed
- It requires admin rights with "Unknown" publisher in the install process --> this is where i stopped trying to install it.

So now I'm looking for a new benchmark alternative for CPU/GPU comparison.

 

AVX-512 barely improved performance in most cases anyway. Most of the time you're memory bottlenecked, and when you're not you get such a huge frequency reduction from AVX-512, the performance barely changes.
Having a non AVX-512 Intel chip suddenly increase in ranking smells of custom Intel code paths being executed.

 

I have never thought very highly of this benchmark and this news confirms my thoughts. To me it seems to be just another "why bother?" serious of tests. I am glad to see this site tends to agree since they do not use it in their CPU reviews. I also noticed Anandtech and Tom's Hardware don't use it either, or at least didn't in their latest series of reviews.

Oh well.

 

So, basically they did the same thing Userbenchmark did - improperly weighted single thread performance over multi thread, in a world where it's absolutely clear that multi thread is going to be the future.

Nice job there, guys. It's forward thinking like that that leads us to big things. Like climate change, and pandemics.

 

The difference being that userbenchmark is a nonsense website that never had an credibility and Passmark was something people actually used to reference. That said, Passmark hasn't been popular in 10 (15?) years and userbenchmark was always a goofy tool for the kiddos - just doubling down on their own irrelevance at this point.

 

Thanks for posting about the update. I was using Passmark 9, and thought my rig was pretty good.

But, now I realize it is pretty much crap.

I needz 2B LeEtZ!!!!1111!!!!. I guess I will throw my rig in the dumpster and lay out few KBucks for an UbErLeEtZ rig!!!!1111!!!!.