Tom's Hardware reports that the Ryzen 9 3950X beats the Core i9-9980XE in Geekbench single core results.
This is starting to confirm a theory I had with Geekbench and why it scores well on Apple SoCs compared to other ARM processors and manages to sneak up on Intel's lower end offerings. But the short of it Skylake really only has four execution ports and they're shared between INT and FP operations. There's four other ports but they're only for load/store operations.
Contrast to the first Zen at least which has four INT pipes and four FP pipes. So if Zen 2 has better IPC, I can see that making up for the 200 MHz turbo boost deficit the 3950X has.
-
Should point out that while 200MHz might sound like a lot, when the i9 has 4.5GHz and the Ryzen 9 has 4.3GHz, this is only a ~4% improvement. This is well within the range that IPC improvements can bring.
-
Geekbench is bullshit, it also greatly favours ARM - comparing an ARM soc to an Intel chip using it is meaningless. I fully expect Zen 2 to beat Intel's IPC but I'll wait for a more meaningful benchmark to know by how much...
-
Unless you can provide a detailed analysis of the distributed executables including assembly dumps, compilers, and compile time options used, the whole "Geekbench favors ARM" argument is weak.
-
A good chunk of Geekbench integer scores are determined by SHA performance which in most devices is hardware assisted - meaning an Intel chip will perform about the same as any other chip in that sense and therefore won't really be able to capitalize on any performance advantage it may offer.
Every other benchmark doesn't show mobile arm chips getting close to Intel, so why should we take Geekbench seriously? https://www.pcworld.com/article/3006268/tested-why-the-ipad-pro-really-isnt-as-fast-a-laptop.html
Although, to be fair, while looking for sources I found out that Geekbench 4 supposedly addresses some of these concerns.
-
@Sauron If you're so opposed to the use of GeekBench, provide alternative metrics for use. Otherwise, complaining into the wind won't help...
-
-
@Sauron That's not providing enough metrics, from what I see here. To see what I mean, make 2 separate lists - one for GeekBench, and one for Cinebench R20. Then, populate the lists with all performance metrics that they measure (IPC, Render Time, etc.). If the two are very different, then you are looking at tools that cannot perform the same task. In that case, you can't simply drop in one as a replacement for the other. If the lists are very similar, then you can do so.
-
-
@Sauron Geekbench 3.4 added support for Intel's SHA instructions. (https://www.primatelabs.com/blog/)
Ultimately if anything, I'm pointing out a flaw in Intel's design.
-
@Mira Yurizaki and it shows, current geekbench scores for mobile arm chips are nowhere near Intel chips.
-
Quote
what do you mean? cinebench is regarded as an accurate and meaningful cpu benchmark for both single core and multicore performance. It's consistent with real world performance and stresses the processor as you would expect.
You missed the point. I never called Cinebench's accuracy or meaningful metrics into question. What I did question was whether it could be used as a drop-in for GeekBench, due to differences in what they measure and how they get their metrics. Read carefully.
- Just because both benchmarks are targeting the same component, doesn't mean that they measure the same metrics for said component
- Even if they measure the same metrics, the way in which said metrics are assessed is also a factor
The second point is one that you just made for me, yourself. The way GeekBench assesses SHA Integer performance, for instance (assuming that you believe what you said).
-
Whether it can exactly replace geekbench isn't really relevant if geekbench isn't good at what it does - if you want to measure SHA performance specifically that's fair but geekbench is used as a catch-all cpu benchmark and does not accurately distinguish how much of the result is due to SHA and how much of it isn't. I would also argue SHA performance isn't that important beyond a given minimum in general use, particularly on a mobile device.
-
Quote
if geekbench isn't good at what it does
Quotegeekbench is used as a catch-all cpu benchmark and does not accurately distinguish how much of the result is due to SHA and how much of it isn't
QuoteI would also argue SHA performance isn't that important beyond a given minimum in general use
Sounds like someone needs to start writing their own benchmarking software if that's the case. You can throw together a suite of tools to do the job if you're really that serious about accurate measurement of overall CPU Integer performance and Compute capabilities. Otherwise, can't help you there.
-
Also, @Sauron, I hope the SHA you're talking about isn't one of these:
- https://en.wikipedia.org/wiki/SHA-2
- https://en.wikipedia.org/wiki/SHA-3
- https://algs4.cs.princeton.edu/34hash/
Just making sure...
