Geekbench Blog
www.geekbench.com
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.
That's exactly what Geekbench is aiming for. They even said "client applications" which is another way of saying "consumer level applications". If you want to test a 128c system, you need something else.So you are saying this benchmark is ONLY for use on desktop PC's ? not even high end ones that CAN use up to 32, or even 64 cores ?
Our concern is that most users, including those considering high-end workstations, don't know whether their workloads scale well.
We've seen several workstation workloads that do not scale well (where an 8-core system performs similarly to a 16-core system) and some that have negative scaling (where an 8-core CPU outperforms a 64-core CPU).
If users understand their workloads scale well, the workloads that scale well (e.g., Clang, Asset Compression, and Ray Tracer) can be used as a reference. Users can focus on those individual workload scores rather than the overall multi-core score.
Our assertion is that very few applications can take advantage of HEDT systems. Even developer workloads like compilation aren't guaranteed to take full advantage of HEDT systems (e.g., RISC-V toolchain compilation performance).
Our other assertion is that users, in general, do not know if their applications will scale. We've heard numerous reports of instances where users go off benchmark numbers and buy a HEDT system only to find that it's slower than their previous system.
Users who purchase these systems should understand that HEDT systems come with compromises (e.g., ThreadRipper systems have a non-uniform memory topology). Having a benchmark that reports best-case scaling does a disservice to users (and the industry in general). Instead, having a benchmark that uses a mix of workloads to measure average scaling (as Geekbench 6 does) gives users a better understanding of the trade-offs that come with HEDT systems.
Users who are sophisticated enough to understand these trade-offs and who are aware that their application will scale on HEDT systems can either use individual workloads in Geekbench 6 to measure performance or use other benchmarks entirely (even Geekbench 5). Adding a second multi-core score to Geekbench would make things more, not less, confusing for our users.
I'm not sure what the "gotcha" is supposed to be. Geekbench is, and always has been, a client-focused benchmark. That some people happen to rely on it for internet dick-measuring contests is not a compelling reason to abandon that focus. And as he says, Geekbench 6's new methodology better mirrors many professional use cases as well. Adobe suite, CAD, etc form the majority of the workstation market, and they don't demonstrate infinite scaling.John's latest thoughts on the matter:
The man himself recommends GB5 if you care about scalable MT workloads/benchmarks/systems. Exactly as it was suggested earlier in the support thread - took several messages back and forth before he admitted that.
So, the case is settled. GB6 MT bench targets an array of poorly scalable benchmarks for the workstation crowd. If you care about anything truly scalable, please refer to GB5.
It uses as many cores as workloads do for the target audience. So yes, as a benchmark, that's an improvement.GB6 development process:
It's using ALL the cores! Oh no! Something is very wrong.
Now it hardly touches any cores. Mission accomplished! Pop out the champagne!
John is going to get AMD conspiracy theorists all riled up.Our other assertion is that users, in general, do not know if their applications will scale. We've heard numerous reports of instances where users go off benchmark numbers and buy a HEDT system only to find that it's slower than their previous system.
Users who purchase these systems should understand that HEDT systems come with compromises (e.g., ThreadRipper systems have a non-uniform memory topology). Having a benchmark that reports best-case scaling does a disservice to users (and the industry in general). Instead, having a benchmark that uses a mix of workloads to measure average scaling (as Geekbench 6 does) gives users a better understanding of the trade-offs that come with HEDT systems.
That misalignment is a far bigger issue with GB5. The ideal GB5 CPU would be 1, maybe 2 big cores, and then as many small cores as can fit. But as GB6 shows, that would be a poor choice for most workloads, and I'd venture a guess you wouldn't be interested in it either. It's also, clearly, not what anyone's actually designing.Third kind: Those who fear that this could motivate idiotic design decisions on future CPU architectures, leading to scaling back on cores and putting 16 cores out of the reach of consumers for good.
Third kind: Those who fear that this could motivate idiotic design decisions on future CPU architectures, leading to scaling back on cores and putting 16 cores out of the reach of consumers for good.
Consumer applications perform better with fewer fast cores over many slow cores. Maybe this will finally force Intel and AMD to design CPUs that actually make consumer applications run faster instead of stuffing as many slower cores as possible.Third kind: Those who fear that this could motivate idiotic design decisions on future CPU architectures, leading to scaling back on cores and putting 16 cores out of the reach of consumers for good.
Normal people can sometimes find a great deal on used 16 core CPUs. Any 16 core CPU bought today will eventually end up in a normal person's PC sooner or later, either because it's too good a purchase at 2nd hand price or it's a hand down from a family member or friend. Just because they may not need them does not mean that developers shouldn't try to make the best use of them. Heck, what if a future Windows version requires 8 core minimum just for its own services? Those with 16 cores at that time will have plenty of cores left for their important tasks. We already should have been at 48 cores in the desktop space but thanks to Intel's misstep, we aren't there yet because AMD feels no pressure from Intel to do that. Heck, Intel managed 32 cores just last year. If the common man doesn't have more than 8 cores, there is no incentive for developers to try to figure out how to run their applications better with more cores. If you build em, they will come.Normal people do not need 16 cores. Period.
It's easy to increase MT. It's hard to increase ST. I'd rather have AMD and Intel focus on improving ST and then give us enough MT where consumer applications are satisfied.
Wow, this article is almost 20 years old, this article is a dinosaur.
Yes, it's harder to come by but it's more impactful. Everyone knows ST is harder. I even write this where you quoted me.
ST improvements are much harder to come by. My 12700K is barely twice the ST performance of my i7-5775C (at least in GB5). AMD would love to obliterate Intel in ST but even their best and brightest don't have the answer to that. Multicore is the only way and they need to work on improving multicore efficiency by reducing latency between cores. Well, unless Intel's Soft Machines project of reverse HT somehow bears fruit.
I think you are looking at this the wrong way.Let's take AMD as an example. AMD is very focused on server and cloud workloads with Zen. Thus, ST is not very important to AMD. They still manage to produce a core that has very good ST though. But what if AMD focused more on ST? Maybe create another core design that is consumer focused.
Those who have such fear obviously have no clue how teams developing CPU select and use benchmarks. GB5 and GB6 are just two among many benchmarks used. And individual scores of subtests are used rather than the aggregated score.Third kind: Those who fear that this could motivate idiotic design decisions on future CPU architectures, leading to scaling back on cores and putting 16 cores out of the reach of consumers for good.
And yet, ~20 years later, single thread performance remains the most important factor for client applications.
ST improvements are much harder to come by. My 12700K is barely twice the ST performance of my i7-5775C (at least in GB5). AMD would love to obliterate Intel in ST but even their best and brightest don't have the answer to that. Multicore is the only way and they need to work on improving multicore efficiency by reducing latency between cores. Well, unless Intel's Soft Machines project of reverse HT somehow bears fruit.
Normal people do not need 16 cores. Period.
Only the enthusiasts, such people who you may found on a forum like this one, may find use for such a larger number of cores.
If you take all the people on the planet who own a computer - laptop or a pc:
Normal people benefit from increasing performance per core, and they need just a few of powerful cores.
- 80% could get by with 4 cores just fine
- 19% could require for what they do on the computer 8 cores, but they could survive with 4 cores as well, occasionally feeling restricted
- 1% could require for what they do on the computer more than 8 cores, but they could survive with 8 cores as well, occasionally feeling restricted
All those chips with large number of cores which are now available on the consumer platforms will be on average massively underutilised.
Those few enthused consumers are lucky that they can purchase such powerful CPUs on the cheapest consumer platform.
Trying to apply a mindset of a miniscule sliver of all computer users to everybody is foolish.