Discussion in 'Frontpage news' started by Hilbert Hagedoorn, May 4, 2017.
I bet it's the infinity fabric...such an evil name has to have something to do with it :banana:
I doubt Intel paid them anything, nor sent them any 1600 dollars CPUs. This CPU-Z benchmark is rather new and obscure. Like Hilbert has said, Intel doesn't give a damn about lots of things anymore, not even sending CPUs for reviewing. So, why would they care about CPU-Z?
I reckon the deeper problem is that this CPU-Z developer genuinely felt it's wrong that Ryzen is doing so fine, so he had to find a way to make it do less fine. Apparently that makes the benchmark more believable and relevant in his opinion. This is an unfortunately prevalent opinion among the masses as well, of course largely due to AMD's own disastrous previous architecture that satisfied nobody, but its roots lie also in Intel's actions in the more distant past when it infamously defamed AMD. The end result is the atmosphere where AMD appearing excellent means something is wrong.
"Hey guys, we have this useless algorithm that processes 30% faster than Skylake processors, core for core!"
BTW Jim Keller was not the most influential designer with Ryzen. He wasn't even mentioned in the Q&A of Ryzen's launch.
Well I guess my question would be, what do we as a community expect out of a benchmark? Should it be representative of what the CPU does in general - in the real world - in it's current state? Or should it be representative of what it could do, given a completely optimized scenario? Should the benchmark combine all the facets of a CPU into one score? Or should it separate them, to show it's strengths and weaknesses compared to another?
I'm of the understanding that the CPU-Z authors said "Hey our benchmark puts Ryzen too far ahead of Intel, which is not representative of real world performance" and thus they changed the weight of some of the results, to more accurately portray where both chips land overall.
Is that right? I don't know. It could be if you think that benchmarks should be representative of current performance. But if you think benchmarks should show what Ryzen could potentially be - then maybe you think it's wrong and misleading customers. Which is also fair, because IMO Ryzen is the better buy compared to four core Intel parts - but that's my opinion. CPU-Z guy might feel different - I don't think it makes him wrong.
I also don't think it makes him paid by Intel - which is my biggest complaint about this entire thing. That's what it immediately becomes - not just a difference of opinion, but sabotage.
They were not inflated. What happened was that previous benchmark was using operations in the way that were not optimal for maximum performance figures, Ryzen with a new scheduler and other optimizations handled that way better then Intel did.
I am fine with new benchmark routine, however, call previous one inflated is about as accurate as calling asynchronous compute cheating.
relative to real-world performance, it's inflated. it's categorized as synthetic. until AMD comes out with some use, or some software engineer figures out a use, it's going to remain that way.
And that analogy makes no sense.
My Ryzen went from
2306 to 431
and from 19256 to 4734
Who cares about CPUz benchmark anyway ? As far i remmber it was always usless and it still is ...
This was actually my reaction to this.
Not only do i agree with this but i have to add: who cares about synthetic benchmarks??? I just skip them , they are useless to me. I want benchmarks with real world software not software that measures something useless...
About CPU-Z scores, i found strange that Ryzen scored higher in STP than Intel but never cared about it.
That analogy between CPU scheduler or CPU logic that make it perform better in specific situations and GPU async compute which more or less do a similar thing? Unless you know what instruction sequence caused Intel to be slower and the reason why Ryzen was better in it, you can't say otherwise.
How you handle instructions is exactly what is behind most of IPC gains lately and in that particular benchmark, Ryzen was just better in it.
Yes, it is a synthetic benchmark and not single one of them make a realistic performance comparison between CPUs, there will be always a quite huge difference between results.
Each synthetic benchmark (or sequence of instructions) tells only how CPU compares in that specific situation and nothing more.
To answer your words, no synthetic benchmark shows you real world performance and so it cannot be inflated. All that you can do, is to select benchmark operation in the way that would have minimum differences....until something similar happens again, then you can change it again.
For me, regardless of real world performance this case just showed me that AMD did a good job and Ryzen handle at least some nonstandard situations better than all previous CPU designs on both sides.
I agree, it's a very superficial way to compare CPU's.
But for lazy people, it's easier to just compare 2 numbers.
Dude, async Compute has many uses, from reducing VR latency to improving performance, and thus ultimately graphics, especially on consoles.
CPU-Z is just a synthetic benchmark. That's it. End of story.
They cannot be compared when calling one useless and the other useful. One is already proven in multiple pieces of software on PC and console.
That doesn't make any sense. You're trying way too hard. The whole point of benchmarks is to have an indication of real world performance. This is literally WHY it was changed.
Ryzen is a fantastic processor.
I think there are much better benchmarks to show you numbers. CPUz should stay as information tool , nothing else. Save some space ...
Async compute is scheduler function on GPU to be able to switch compute tasks between GPU cores to speed up calculations by "optimaly" distributing the workload.
CPU scheduler does more or less same it tries to distribute the workload over CPU cores.
Both of them increase calculation throughput ad thus performance.
As I said before, none of the synthetic benchmarks is really useful to show real word performance. They are too specific and you would need to run same mathematical model as in real applications and there is just too many possibilities.
Synthetic benchmarks are only useful to give you an idea about possible performance and capabilities and only if you look at more of them.
If you run 10 DIFFERENT mathematical models and most of them are better on specific CPU then it gives you an idea.
I would not go that far, it's good CPU and for very good Price and yes its surprisingly good given its complete new architecture with a lot of new untested technologies in it.
If (or once) AMD manage to fix issues and weak points it has, then it would be the fantastic processor.
Just now hoping that new AGESA update and bugfixes coming this month finally make my memory to run on more than 2133 MHz (Damn Hynix chips)
BTW, I really not trust any Synthetic benchmarks as working in Enterprise ERP area taught me very well that even in same application workload and related CPU performance differs a lot.
They are working with Hynix.
Just out of curiosity, what processor would you consider a fantastic processor?
I've seen this floating around, but it sounds pretty retarded if you ask me.
Do you have a single fact to back that up?
No, I don't.
Nvidia asked to disable it for their GPUs since it was negatively impacting performance. They didn't ask to disable it for AMD too, quit inventing.
What I do remember however, is Oxide having no damn clue how Maxwell cards are supposed to work and throwing Radeon-specific optimizations at them and saying 'look, they don't work the same'. Oh surprise surprise, who would have thought that you need to optimize for both architectures individually?
Probably, yeah. CPUs are general purpose processors, chasing different corner cases or architecting for certain use cases doesn't make much sense - it only makes sense for GPUs which are highly specific pieces of hardware.
Some performance differences in CPUs are a byproduct rather than specific design.
1 Lisa Su herself said that they're under Skylake's IPC
2 Ryzen clocks lower than Kaby Lake
1 + 2 = Ryzen has a higher single-core performance than Kaby (like the old CPU-Z showed)?
Man, some of you should try using your synapses.
Goddamn man, you get the cake.
This is exactly why I've been avoiding some subreddits lately, fanboy infestation is abound.
Everything is a conspiracy against AMD now. For example my cat just farted. Probably because my neighbour bought a Ryzen.
Haha I'll look into this later. I need few laughs