Geekbench 6 - Geekbench Blog
Weird choice of baseline CPU and even weird is that the baseline score is 2500.
i7-12700 does hardly 2000 in GB5 with the fastest DDR5.
Weird choice of baseline CPU and even weird is that the baseline score is 2500.
i7-12700 does hardly 2000 in GB5 with the fastest DDR5.
What I'd like to see is some sort of a long, complex, hybrid workload. Like building and packaging a game, which can consist of multiple processing steps like texture generation, mesh optimisation, pre-backed retraced lighting, compilation, asset compression etc., all using a parallel build system. This kind of stuff goes in the direction you mention while still being practically relevant.
I don't know who came up with this test (Anand or Ganesh or someone else) but it's pure genius. Shows how 4800U/5800U trump ADL-mobile in concurrent workloads. 12th gen is worse than even Comet Lake!
It's not about favoring AMD. It's a valid use case. Lots of people will run some workload and start doing something else while the original workload chugs along in the background. AMD is just better at it and if there is more awareness about that through benchmarks exposing this benefit of AMD's architecture, it will also pressure Intel to improve the deficiencies in their architecture to address this common use case.
They run various benchmarks while looping video transcoding task. I am sorry, but quantifying effects of ONE specific concurrent task with ONE set of parameters (video formats, transcoding parameters) has very limited general significance.
I don't know who came up with this test (Anand or Ganesh or someone else) but it's pure genius. Shows how 4800U/5800U trump ADL-mobile in concurrent workloads. 12th gen is worse than even Comet Lake!
What if Alder lake CPU gets too distracted by one particular video format, while it could run some other format effortessly? Did they allow integrated graphics to help with the encoding, and if not, is it fair to test a product with a part intended for this specific workload disabled? Etc etc etc.
So yeah, I am not convinced that AMD CPUs are better in multitasking than Intel CPUs by this at all.
Whover Thinks a 7950X and 13900K are better than The ThreadRipper Pro 5995WX because GeekBench6 says so has gone Senile. Simple as that. GB6 is a light "Real World" Benchmark that is design for Phones and Laptops.
When Xeon W9 releases this will be the worst app to use to test it.
Simply put it, Do you want to test Workstation performance? Don't use GB6, but PugetSystem or something else.
When Xeon W9 releases this will be the worst app to use to test it.
Simply put it, Do you want to test Workstation performance? Don't use GB6, but PugetSystem or something else.
I 100% believe that a 7950x and a 13900K are better than a Zen3 Threadripper for most consumer applications.
Do you believe that an Android Phone is better than a 13900K?
So why don't you just run Geekbench 6 while VLC is encoding some video and see. It seems bizarre that you think Geekbench should implement this sort of test.
View attachment 76537
I don't know who came up with this test (Anand or Ganesh or someone else) but it's pure genius. Shows how 4800U/5800U trump ADL-mobile in concurrent workloads. 12th gen is worse than even Comet Lake!
You're pointing at a borked result for the Xiaomi. There are ways of spoofing Geekbench into giving unrealistically high scores; you've found one such result.
7 ZIP, Handbrake or Agisoft are all consumers apps, besides tests that focus on a single task with a low core count are useless to estimate the perfs since the CPU is underused while the task use very low time for exe, that s akin to microbenchmarks that say nothing about the perfs when there s a really demanding set of loads or a soft like the ones i quoted.
View attachment 76537
I don't know who came up with this test (Anand or Ganesh or someone else) but it's pure genius. Shows how 4800U/5800U trump ADL-mobile in concurrent workloads. 12th gen is worse than even Comet Lake!
This is the big-little design disaster. In theoretical test it's impossible that 1260P would lose to a 5800U and even not any better than a 4800U, but the multitasking while handled by crappy Thread Detector is the only explaination for this. Frequently switching between big and little cause too much latency and perf loss. Similar situations happened every time I heard from others are either multitask lagging or perf loss during aggressive switching, and sometimes even as bad as only the little cores working while big cores keep idle.
Last edited:
I said most. Not all. There's a name for this kind of fallacy where someone tries to argument a statement as not true by providing a few niche examples.
How much 7zip, Handbrake, Agisoft are people doing on a 13900k?
It depends. What application are we trying to use? An Android phone could be better than a 13900k. It depends, right?
Also, this erroneous result doesn't change anything.
Why are people so sensitive about this? This statement isn't controversial: A 7950x and a 13900k should run consumer applications better than a Zen3 Threadripper. Relax people.
The thing is that GB does no more provide the extent of the perfs, whatever has enough perf is stuck to about 20 000 pts, the fallacy is to state that such an arbitrary measurement is good enough hence anything that is better at some tasks will be counted as equal if not slower...
FI a 7700X@88W is quite faster in the MT test than a 5950X, yet in general usage ,like browsing, there wont be any difference while in heavy MT the latter will be way faster, so the usage perf is not reflected, a browser that react at 0.05s instead of 0.06s is a useless measurement because it s not that kind of tasks whose exe speed does matter to the user.
Edit : Computerbase made an article about GB6 and a graph populated by their forumers with ST et MT scores, i guess it s now clear for whom it was specifically re "designed", and certainly not for the users needs...
Geekbench 6: Die neue Benchmark-Suite im Leser-Benchmark
Der Geekbench 6 bringt einige neue Workloads mit, bestehende wurden mit neuer Datenbasis aktualisiert. Leser können Ergebnisse einreichen.
www.computerbase.de
Last edited:
poke01
Diamond Member
- Mar 8, 2022
- 3,433
- 4,713
- 106
Comparing a 13900k and a 7950x. The 13900k comes out ahead in MT. This makes sense because 13900K has more cores.
But was Geekbench never perfect in MT.
No non-x86 chip on GB6 is close to the i9 but Apple's. That's fake.
I think the "unrealistic expectations" part is spot on here. With current set of tests, there are not that many where AVX512 could help much with direct recompile of same code. Maybe photo library stuff, photo filter.
The other tests could hardly use AVX512 without proper handwritten code paths and that would defeat multi platform and vendor "agnostic" part of the tests - ARM guys would ask for SVE3 with 666bit vectors and so on. In previous GB5 we had retarded outliers like say FFT or encryption that were directly calculating "throughput" and would probably make sense to double FLops just by using some vendor library or some short code that gets autovectorized and support AVX512.
Even then CPU vendors found that it is easier to pad the score by including some 256bit V_AES instruction with ridiculous throughput that made some new laptop beat a whole server in AES encryption throughput ( and of course utterly suck in real world, as to actually serve up content for encryption and deliver it/from requires actual server).
So GB6 is my opinion a great desktop/workstation performance test that emphasizes the way people actually use CPUs in 2023 and is brave enough to shatter illusions of people who disagree that 8 strong cores are plenty and the rest ( be it 8 more strong cores or 16 marketing cores ) gives very diminishing gains. Kudos to them for not catering to Cinebench/DC runner crowd.
moinmoin
Diamond Member
- Jun 1, 2017
- 5,206
- 8,367
- 136
No, but there is a strong relation between the ST and the MT score, regardless of the actual amount of cores present. Which makes the latter effectively meaningless.
The numbers I see so far and their ratios/balance between ST and MT make me expect Zen 4 Threadripper to beat 7950X slightly, but by far not reflecting the additional amount of cores, rather reflecting the higher ST achievable on more cores across the chip.
Beyond a rather small amount GB6's MT score is highly tuned to be agnostic to the amount of cores present.
Yeah, single specific workload run in isolation. That's not what happens in reality. That's also not what users looking at 7950X, Threadripper, workstation or server chips usually want. Instead they want to run a higher amount of workloads concurrently without hitting bottlenecks. For that purpose GB6's MT score is completely misleading while GB4's and GB5's MT scores were rather serviceable.
Reminder that before Ryzen Intel actually started with Cinebench as marketing benchmark (wtftech article and slide from 2011). Cinebench was known as Intel-optimized to boot, so obviously once AMD could showcase beating Intel with it using Ryzen they did just that.
moinmoin
Diamond Member
- Jun 1, 2017
- 5,206
- 8,367
- 136
The current MT and ST scores with ratio between them, sorted by that.Geekbench 6: Die neue Benchmark-Suite im Leser-Benchmark
Der Geekbench 6 bringt einige neue Workloads mit, bestehende wurden mit neuer Datenbasis aktualisiert. Leser können Ergebnisse einreichen.
It's not even Good for phones, it measure only Burst performance, most phones nowadays can't even sustain 70% of their performance.
lightmanek
Senior member
- Feb 19, 2017
- 508
- 1,245
- 136
Geekbench 6 is good for ST and benchamrking OS/App responsiveness, but it is nowhere near good at indicating MT performance for prosumers. Luckily we have a ton of other, more useful benchmarks for that, so I'm OK with GB 6 showing MT performance as it is, as long as people understand that is not what you look at to evaluate CPU's for HPC/heavy CPU transcode/Code Compile/Scientific and many more.
One feature I would like to see in GB is to have ability to select duration of each task, so you could adjust the load on CPU and see how it can sustain high clocks at light MT loads and when you might be hitting power / cache limits with increased loads.
GB6 to me is like EPA car mileage to real world use - yes, it is a good indication between manufacturers and you can strive to achieve claimed results when using a car in very specific way, but your mileage will vary depending on a lot of factors in real world.
One feature I would like to see in GB is to have ability to select duration of each task, so you could adjust the load on CPU and see how it can sustain high clocks at light MT loads and when you might be hitting power / cache limits with increased loads.
GB6 to me is like EPA car mileage to real world use - yes, it is a good indication between manufacturers and you can strive to achieve claimed results when using a car in very specific way, but your mileage will vary depending on a lot of factors in real world.
Last edited:
Uhh, wot?
So your contention is that most people were buying GB5 to engage in dick-measuring, and are upset that it's now targeted at useful information rather than dick-measuring?
OK...
No, it just shows that multithreading has overhead and does not scale all that well with the number of cores. And it also shows that Intel's E-cores are only useful for trivial parallel workloads.
Not really. It's just how CPUs are marketed. Higher-end CPUs usually have both more cores and higher single-core boost. Which is why for many of the current models the two scores appear correlated. Well, because they are, that's how the chip is configured to begin with.
It becomes a bit more clear with Apple who is probably the only manufacturer that ships CPUs with consistent clock configuration across models. If you look at MT scores of M1, M1 Pro/Max, and M1 Ultra you will see that every additional cluster of four cores adds about 50% of the original MT score. That's your locking/cache coherency overhead. Makes sense to me.
For some workloads, sure. But GB6 also includes trivially parallelizable tasks that scale very well with the number of cores, like RT. And of course, if you have some specific use case in mind you should benchmark that use case.
Cinebench massively favours high core counts (trivially parallel work) as well as fast/wide SIMD units and caches (long data dependency chains, SIMD heavy). It overestimates performance in any kind of workload that doesn't use SIMD that much or actual needs some cooperation between the cores.
I don't think it's that much off for code compile. Compilation is known to show diminishing returns with increasing core count.
Timed LLVM Compilation Benchmark - OpenBenchmarking.org
Here you can see how doubling the amount of cores (with the same power/frequency per core) doesn't reduce the time in half. The scaling observed there isn't that different from what we see in GB6.
moinmoin
Diamond Member
- Jun 1, 2017
- 5,206
- 8,367
- 136
GB6 also including trivially parallelizable tasks is meaningless if the overall MT score barely reflects the actual core count. Ideally GB would split up MT scores between workload tests only using a limited amount of cores and parallelizable tests extending to all available threads.
Isolated benchmarks only make sense for isolated workloads. As you yourself point out at length workloads that scale well to any number of cores are the exception. But typical usage of chips that contain a massive amount of cores is not running isolated workloads, especially in servers it's running a lot of workloads concurrently. For that purpose GB6's MT score is just completely useless at best and misleading people at worst.
Geekbench was never meant to be, nor ever has been, a good workload by which to judge servers.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 24K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 20K
-
-
-
Facebook
Twitter