Intel & AMD Processor Hierarchy

Page 6 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

M42

Reputable
Nov 5, 2020
99
48
4,560
As for multi score i would expect 12900k/ks to be approx double seeing as it is running double the cores. This is just logical . Then if you take into account the fact that the p cores are a newer generation and faster and the e cores are a little slower it still balances out about double! Its Logicical !
But multicore vs 12900k/ks is a completely different price and product segment.
I don't know how you have come to compare 12900ks to 5800x3d for multi core ? Its just not logical.
Its all about gaming for 5800x3d
The 12900 variants have 8 high-performance cores and 8 high-efficiency cores. The 8 high-efficiency cores add little to the multicore performance. In fact, on Windows 10 this can hurt performance in some situations because Windows 10 is not aware of the different CPU types. Thus 12900ks being 94.6% faster doesn't seem unreasonable.
 

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
The 12900 variants have 8 high-performance cores and 8 high-efficiency cores. The 8 high-efficiency cores add little to the multicore performance. In fact, on Windows 10 this can hurt performance in some situations because Windows 10 is not aware of the different CPU types. Thus 12900ks being 94.6% faster doesn't seem unreasonable.
That's what I said 94% faster in multicore ! But the 8 efficiency cores add a lot to the multi core score, not a little . Yes it can run bad in win10. But e cores add nothing to gaming at this point in time as a general rule although Hitman 3 has been coded to make use of them for some tasks to off load stuff.
But as I have already said its in a completely different price and product category .
 
Last edited:

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
If you fully read any of the recent in-depth reviews for the 5800X3D you should have seen that not only does it clock slower than the 5800X but it cannot be overclocked, so for regular applications the 5800X is faster than the 5800X3D.

Again the only reason the 5800X3D is faster in some games is that those particular games have tighter loops where the 5800X3D can access its internal cache instead of being slowed down by accessing RAM. In other games the 12900ks is much faster because the 5800X3D has fewer cache hits.

If you don't understand why this is so, then I suggest you read a little more on the topic of CPU caches.
Well yes but only by 5 % max in single core and maybe upto 10% in multi core but that is not for everything . As for overclocking it doesn't really matter for gaming as it makes no difference infact quite often its worse but the default boost clock still works ok but it is hard limited to 1.35v.
There are 2 points of comparison here i am using.
You don't seem to understand the points i am making or are just dicking me along!
Firstly I am comparing 5800X vs 5800X3D and 2ndly 5800X3D vs 12900k/ks with the focus on gaming!
You are saying because the Ryzen 5800X3D doesnt overclock its way slower ! For gaming its irrelevant but for apps you could get a further boost maybe 10 -15 % at best. ! But the whole point of the 5800X3D is to improve gaming performance in which it does quite well, better than I thought, but you mileage will vary!
The 3rd point I make is you are telling me 12900ks is 71% faster in passmark single core score. Although you have gone quiet on that this time?
I already stated logically why I think that score is flawed. The cpus are almost identical bar the slightly lower clocks and 3d v cache. So therefore it will be slightly slower in things that don't use the 3d v cache. But it wil get a boost in things that do use the 3dv cache. So it will be close. Its sample size is just 1 as compared to the 5800X sample size of 4290 . They even say it on the passmark website that its highly likely to have a high error factor cos sample size is just 1 sample!
I think you are just dicking me along tbh.
I have already stated 12900k will crush it in apps but that is logical and a given seeing it has twice as many cores. Again that's not what 5800x3d is for . Ryzen has 12 and 16 core cpus to compete with that.
And price of the halo 12900ks is stupidly expensive for slight gain and 12900k is the top of the line both different in price bracket and app segments .

When i reread your post you are saying that 5800X3D is no good and way slower cos it can't overclock and only works on a few games and is a one trick pony .
Well you could say that being really negative but in reality it turns out to be the fastest gaming cpu currently available on the planet. They have sacrificed a little clock speed to implement new 3d v cache technology. It has achieved a massive boost for gaming over the original 5800X so much so that Tomshardware rates it the best available currently. So they have achieved what they set out to do .
If you look at Hardware Uboxed they tested it on 40 odd games with the Ryzen 5800X3D.
Its a good spread of games. Still fares well !
 
Last edited:
:sleep:


Yes we shall wait and see when and where it ends up on the heirarchy chart .
My bet is going by past efforts, they will take forever to put it on ( if at all), but they will not put it at the top cos it is not as fast for other programs ie non gaming apps even though it is a supposed to be a CPU GAMING HEIRARCHY LIST .
Watch this space ! 😆
I hope they prove me wrong... but i wouldnt hold my breath!

The chart specifically says '...Gaming CPU Benchmark'.
The article mentions the 5800X3D.
The Windows 11 graph shows the 5800X3D on top.
The Windows 10 graph doesn't show it at all...?

I guess they didn't finish testing the 5800X3D on Windows 10...? That's one Hell of an omission though. If you are actually mentioning a new CPU in the article, one that is supposed to take the top spot in gaming performance, you should finish all your testing before you decide to just leave it off the charts/lists without any explanation whatsoever.
 
Last edited:
  • Like
Reactions: guru7of9

M42

Reputable
Nov 5, 2020
99
48
4,560
Well yes but only by 5 % max in single core and maybe upto 10% in multi core but that is not for everything . As for overclocking it doesn't really matter for gaming as it makes no difference infact quite often its worse but the default boost clock still works ok but it is hard limited to 1.35v.
There are 2 points of comparison here i am using.
You don't seem to understand the points i am making or are just dicking me along!
Firstly I am comparing 5800X vs 5800X3D and 2ndly 5800X3D vs 12900k/ks with the focus on gaming!
You are saying because the Ryzen 5800X3D doesnt overclock its way slower ! For gaming its irrelevant but for apps you could get a further boost maybe 10 -15 % at best. ! But the whole point of the 5800X3D is to improve gaming performance in which it does quite well, better than I thought, but you mileage will vary!
The 3rd point I make is you are telling me 12900ks is 71% faster in passmark single core score. Although you have gone quiet on that this time?
I already stated logically why I think that score is flawed. The cpus are almost identical bar the slightly lower clocks and 3d v cache. So therefore it will be slightly slower in things that don't use the 3d v cache. But it wil get a boost in things that do use the 3dv cache. So it will be close. Its sample size is just 1 as compared to the 5800X sample size of 4290 . They even say it on the passmark website that its highly likely to have a high error factor cos sample size is just 1 sample!
I think you are just dicking me along tbh.
I have already stated 12900k will crush it in apps but that is logical and a given seeing it has twice as many cores. Again that's not what 5800x3d is for . Ryzen has 12 and 16 core cpus to compete with that.
And price of the halo 12900ks is stupidly expensive for slight gain and 12900k is the top of the line both different in price bracket and app segments .

When i reread your post you are saying that 5800X3D is no good and way slower cos it can't overclock and only works on a few games and is a one trick pony .
Well you could say that being really negative but in reality it turns out to be the fastest gaming cpu currently available on the planet. They have sacrificed a little clock speed to implement new 3d v cache technology. It has achieved a massive boost for gaming over the original 5800X so much so that Tomshardware rates it the best available currently. So they have achieved what they set out to do .
If you look at Hardware Uboxed they tested it on 40 odd games with the Ryzen 5800X3D.
Its a good spread of games. Still fares well !

First, I didn't think I needed to say more on the 12900ks vs 5800X3D single -thread performance because I had already explained why there is a huge disparity and I thought you understood, or would at least try to research why this is so.

Let me try again, the only reason 5800X3D is fastest in some games is because of the extra memory cache. AMD even makes a big deal about this. This cache is useful in select games where there are tight loops in code. This boosts effective performance in those areas because the CPU doesn't have to wait as long to retrieve code or data as it would from RAM. That is independent of actual single-thread performance when having to access memory outside of cache, which is typically the case in most applications and many games where the 5800X3D's cache is not big enough to be as effective. Without changing anything else, if the 12900ks had equivalent cache it would likely decimate the 5800X3D in most titles. The only difference then is some titles favor Intel or AMD inherently.

I'm sure there are others, but here is another review that shows single-core performance of the 5800X3D, as well as 5800X and 12900ks. As you can see the 5800x's single thread AND multi-core performance in Geekbench is faster than the 5800X3D. The 12900ks is at the top. In this case the performance is only 39% faster in single-core performance, but that is still huge.

 
If you fully read any of the recent in-depth reviews for the 5800X3D you should have seen that not only does it clock slower than the 5800X but it cannot be overclocked, so for regular applications the 5800X is faster than the 5800X3D.

Again the only reason the 5800X3D is faster in some games is that those particular games have tighter loops where the 5800X3D can access its internal cache instead of being slowed down by accessing RAM. In other games the 12900ks is much faster because the 5800X3D has fewer cache hits.

If you don't understand why this is so, then I suggest you read a little more on the topic of CPU caches.
It's obvious that the Passmark single-threaded result for the 5800X3D is garbage, though. 3486 for the 5800X, 2509 for the 5800X3D. That's without overclocking or any other factors, supposedly, which means according to Passmark the 5800X runs 39% faster than the 5800X3D. If you look at our own benchmarks, however, the 5800X single-threaded result (geomean) was 337.5 and the 5800X3D result was 315.5. So our tests have the 5800X outperforming the 5800X3D by just 7%, which is much more in line with the difference in maximum boost clocks.

Basically, no one should ever trust Passmark as a reliable test. It's sometimes accurate, but clearly the 5800X3D figures are garbage right now.
 
  • Like
Reactions: King_V

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
It's obvious that the Passmark single-threaded result for the 5800X3D is garbage, though. 3486 for the 5800X, 2509 for the 5800X3D. That's without overclocking or any other factors, supposedly, which means according to Passmark the 5800X runs 39% faster than the 5800X3D. If you look at our own benchmarks, however, the 5800X single-threaded result (geomean) was 337.5 and the 5800X3D result was 315.5. So our tests have the 5800X outperforming the 5800X3D by just 7%, which is much more in line with the difference in maximum boost clocks.

Basically, no one should ever trust Passmark as a reliable test. It's sometimes accurate, but clearly the 5800X3D figures are garbage right now.
Touche ! That's what I been saying all along !
 

M42

Reputable
Nov 5, 2020
99
48
4,560
Touche ! That's what I been saying all along !
Saying it doesn't make it true. Geekbench is garbage too, then?

As a general CPU, the 5800X3D is slower than the 5800X, as reported from multiple sources and benchmarks. Do your research! :)
 
Last edited:

M42

Reputable
Nov 5, 2020
99
48
4,560
And by the way, regarding multi-threaded performance, the 5800X3D has 16 logical cores, which includes the hyperthreaded cores. The 12900-series only has 24 logical cores, because the 8 p-cores are hyperthreaded (16-cores total) and the 8 e-cores are not. Thus, the 12900 series only has 50% more logical cores than the 5800X3D, and those extra cores are not as performant.
 
Saying it doesn't make it true. Geekbench is garbage too, then?

As a general CPU, the 5800X3D is slower than the 5800X, as reported from multiple sources and benchmarks. Do your research! :)
Actually, yes, Geekbench is also garbage as far as a truly representative benchmark showcasing CPU performance. Just about any fully synthetic test will fall into the same category.

But yes, 5800X3D tends to be slightly slower than the 5800X outside of games. Only memory intensive applications like 7-zip compression and y-cruncher will run faster thanks to the larger L3 cache. If you see more than a 10% deficit running stock, however, then I'd definitely question the result.
 
  • Like
Reactions: King_V and guru7of9

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
First, I didn't think I needed to say more on the 12900ks vs 5800X3D single -thread performance because I had already explained why there is a huge disparity and I thought you understood, or would at least try to research why this is so.

Let me try again, the only reason 5800X3D is fastest in some games is because of the extra memory cache. AMD even makes a big deal about this. This cache is useful in select games where there are tight loops in code. This boosts effective performance in those areas because the CPU doesn't have to wait as long to retrieve code or data as it would from RAM. That is independent of actual single-thread performance when having to access memory outside of cache, which is typically the case in most applications and many games where the 5800X3D's cache is not big enough to be as effective. Without changing anything else, if the 12900ks had equivalent cache it would likely decimate the 5800X3D in most titles. The only difference then is some titles favor Intel or AMD inherently.

I'm sure there are others, but here is another review that shows single-core performance of the 5800X3D, as well as 5800X and 12900ks. As you can see the 5800x's single thread AND multi-core performance in Geekbench is faster than the 5800X3D. The 12900ks is at the top. In this case the performance is only 39% faster in single-core performance, but that is still huge.

Yes 5800x is faster than 5800x3d and its upto about 5% at most single and multi thread apps . I have never disputed that !
You seem to not like the 3dv cache for some reason not sure why tbh? Its not perfect but nothing ever is but it works a treat for gaming!
I will dispute 12900ks single core 71% in passmark but, till the cows come home . On average It's around 25% faster as a rough guide.
I have looked at lots of different app benchmarks and I still say as a general rule the 12900k/ks is around 25% faster single thread and almost double in multi thread.
You need to look at a lot of different testing web sites . They vary a bit but that is the general trend.
 

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
And by the way, regarding multi-threaded performance, the 5800X3D has 16 logical cores, which includes the hyperthreaded cores. The 12900-series only has 24 logical cores, because the 8 p-cores are hyperthreaded (16-cores total) and the 8 e-cores are not. Thus, the 12900 series only has 50% more logical cores than the 5800X3D, and those extra cores are not as performant.
Now you are starting to sound silly . You should probly research that a bit me thinks
You should know hyperthreading is not 100% same as a physical core performance . Infact far from it.
 

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
If you fully read any of the recent in-depth reviews for the 5800X3D you should have seen that not only does it clock slower than the 5800X but it cannot be overclocked, so for regular applications the 5800X is faster than the 5800X3D.

Again the only reason the 5800X3D is faster in some games is that those particular games have tighter loops where the 5800X3D can access its internal cache instead of being slowed down by accessing RAM. In other games the 12900ks is much faster because the 5800X3D has fewer cache hits.

If you don't understand why this is so, then I suggest you read a little more on the topic of CPU caches.
Well yes that is the whole reason for 3dvcache! To make games run faster! It doesn't work for every game but it works on a lot more than people first thought it would, which is quite a lot.
 

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
Saying it doesn't make it true. Geekbench is garbage too, then?

As a general CPU, the 5800X3D is slower than the 5800X, as reported from multiple sources and benchmarks. Do your research! :)
Which part of I have already agreed with you on this many times Don't you get?? ?
But its mininlmal and the 3dvcache gives you a good performance boost gaming for the most part !
Geekbench tends to favour Intel but the single thread is still only around 25% faster .
 

guru7of9

Reputable
Jun 1, 2018
58
7
4,545
Actually, yes, Geekbench is also garbage as far as a truly representative benchmark showcasing CPU performance. Just about any fully synthetic test will fall into the same category.

But yes, 5800X3D tends to be slightly slower than the 5800X outside of games. Only memory intensive applications like 7-zip compression and y-cruncher will run faster thanks to the larger L3 cache. If you see more than a 10% deficit running stock, however, then I'd definitely question the result.
Totally agree but I was trying to be gentle! 😊
 

M42

Reputable
Nov 5, 2020
99
48
4,560
It's obvious that the Passmark single-threaded result for the 5800X3D is garbage, though. 3486 for the 5800X, 2509 for the 5800X3D. That's without overclocking or any other factors, supposedly, which means according to Passmark the 5800X runs 39% faster than the 5800X3D. If you look at our own benchmarks, however, the 5800X single-threaded result (geomean) was 337.5 and the 5800X3D result was 315.5. So our tests have the 5800X outperforming the 5800X3D by just 7%, which is much more in line with the difference in maximum boost clocks.

Basically, no one should ever trust Passmark as a reliable test. It's sometimes accurate, but clearly the 5800X3D figures are garbage right now.
Hi Jarred, in general I believe that most people posting scores to Passmark represent high-end users trying to tweak their computers to get the highest performance. so the scores probably represent performance under the most ideal conditions, overclocking included. I agree that there are not enough 5800X3D scores yet to produce an accurate assessment of the 5800X3D's performance, but even after more results are in, its single-threaded performance should be well behind the Alder Lakes desktop CPUs in almost any benchmark available.

As I am sure you know the primary reason for the 5800X3D's fantastic performance in some games is the extra cache and not improved performance in its CPU instructions or increased CPU clock speed. And because the 5800X3D cannot be easily overclocked, I believe that the 5800X, which can be overclocked, will likely end up with more than a 7% CPU advantage in the long run. I guess we'll find out when more results are in. And, thanks for taking to time to reply!
 

M42

Reputable
Nov 5, 2020
99
48
4,560
Actually, yes, Geekbench is also garbage as far as a truly representative benchmark showcasing CPU performance. Just about any fully synthetic test will fall into the same category.
So, every synthetic bench test is invalid for testing? I totally disagree with this statement. I've been a software developer for over 40 years and my company has used many synthetic tests to measure hardware and software performance. We regularly benchmark software and hardware and we can accurately project software performance using various CPUs, Intel and AMD, from various synthetic tests. These tests are just as valid as measuring frame rates in games. This site is called "Tom's Hardware", and I thought its mission is for tech enthusiasts, which includes other areas than games? :unsure:
 

M42

Reputable
Nov 5, 2020
99
48
4,560
Which part of I have already agreed with you on this many times Don't you get?? ?
But its mininlmal and the 3dvcache gives you a good performance boost gaming for the most part !
Geekbench tends to favour Intel but the single thread is still only around 25% faster .
I understand that, but I didn't make up the Passmark numbers either. The low 5800X3D score might have been from a poorly configured system, with slow memory, other applications running in the background, or whatever. It doesn't mean that Passmark is flawed, just that the test system probably was not performing well.
 

M42

Reputable
Nov 5, 2020
99
48
4,560
Now you are starting to sound silly . You should probly research that a bit me thinks
You should know hyperthreading is not 100% same as a physical core performance . Infact far from it.
Of course hyperthreaded cores only contribute 25-30% more performance, but that still is a factor. You said the 12900ks has double the cores, but the cores are not the same so it made little sense to compare them like that. Not only are e-cores less performant, they don't get the extra 25% from hyperthreading.
 
So, every synthetic bench test is invalid for testing? I totally disagree with this statement. I've been a software developer for over 40 years and my company has used many synthetic tests to measure hardware and software performance. We regularly benchmark software and hardware and we can accurately project software performance using various CPUs, Intel and AMD, from various synthetic tests. These tests are just as valid as measuring frame rates in games. This site is called "Tom's Hardware", and I thought its mission is for tech enthusiasts, which includes other areas than games? :unsure:
Read the words: Any fully synthetic test. You can maybe learn some things about an architecture, but you should not base your opinion on stuff that easily fits in the L1 cache, because most workloads aren’t like that.

Want to know pure hypothetical integer throughput? Then a synthetic test might tell you that. But the synthetic results won’t explain to you whether a system works well for general office use, web browsing, playing games, or other workloads.

In a similar vein, using results from something like AIDA64‘s memory tests and assuming that they will actually matter for all applications is idiotic. Just like looking at the results from gaming performance and assuming that they apply to encryption, or cryptography in general, or video editing or… Hopefully you get the point.

Geekbench isn’t totally worthless, neither is Passmark, but no one should take their results as representative of anything other than a small snippet of what performance looks like. Full system test, like PCMark, give a far better picture of how I processor performs in the real world. But even those results need to be interpreted correctly.
 
  • Like
Reactions: King_V
So, every synthetic bench test is invalid for testing? I totally disagree with this statement. I've been a software developer for over 40 years and my company has used many synthetic tests to measure hardware and software performance. We regularly benchmark software and hardware and we can accurately project software performance using various CPUs, Intel and AMD, from various synthetic tests. These tests are just as valid as measuring frame rates in games. This site is called "Tom's Hardware", and I thought its mission is for tech enthusiasts, which includes other areas than games? :unsure:
The problem is, synthetic tests are not necessarily going to be representative of performance in real-world applications, especially when comparing different hardware architectures to one another. Different processor architectures are going to be faster at different things, and any single benchmark is inevitably going to favor one over another. Even when a benchmark tries to cover a wide variety of algorithms to come to its final score, the portion of the score weighted to each algorithm is not likely to be representative of any given piece of software. And not everyone makes use of the same software, so ranking processors by a single score (or pair of scores, in the case of single and multithreaded performance) in a one-size-fits-all approach is never going to produce particularly accurate results.

It tends to be much better to look at results of tests in actual software, focusing on applications that are more relevant to your needs, while downplaying those that are not. So, one should be looking through reviews, rather than just scanning some chart based on an arbitrary score derived from who knows where. Passmark seems to be pretty vague about what they are actually testing, and how that is compiled into a final score. All I can gather from their performance test page is that their CPU test "executes complex mathematical calculations involving compression, encryption and physics simulations". How relevant is that to any given user who might be browsing the web, or gaming, or editing photos, or rendering in 3D software? Probably not very.

And it could be considered a similar case for this hierarchy chart at Tom's, even in the context of gaming performance. They base the numbers here on benchmark results from just a handful of games that, on average, tend to be more demanding on the CPU than most game releases. So are the results representative of the typical gaming performance that most will experience? Probably not. Nor do I expect these results to necessarily be representative of performance in future games. And I know a few of these titles show disproportionately larger gains from 3D V-Cache than what the vast majority of titles do. So I would fully expect the 5800X3D to top this chart. But outside a handful of titles, it's probably not going to be noticeably better for gaming than other processors costing significantly less.

As for a 12900KS supposedly being 71% faster than a 5800X3D in single-threaded performance according to Passmark (based on 1 sample), anyone who's looked at a review of the processor could tell you that's complete nonsense, and doesn't even come close to representing reality. There's clearly something wrong there.

And for overclocking, recent high-end CPUs don't tend to overclock much, as Intel and AMD perform binning at the factory and design the processors to reach much closer to their limits at stock. So the lack of overclocking support for the 5800X3D isn't likely to affect much.
 
  • Like
Reactions: King_V

M42

Reputable
Nov 5, 2020
99
48
4,560
The problem is, synthetic tests are not necessarily going to be representative of performance in real-world applications, especially when comparing different hardware architectures to one another. Different processor architectures are going to be faster at different things, and any single benchmark is inevitably going to favor one over another. Even when a benchmark tries to cover a wide variety of algorithms to come to its final score, the portion of the score weighted to each algorithm is not likely to be representative of any given piece of software. And not everyone makes use of the same software, so ranking processors by a single score (or pair of scores, in the case of single and multithreaded performance) in a one-size-fits-all approach is never going to produce particularly accurate results.

It tends to be much better to look at results of tests in actual software, focusing on applications that are more relevant to your needs, while downplaying those that are not. So, one should be looking through reviews, rather than just scanning some chart based on an arbitrary score derived from who knows where. Passmark seems to be pretty vague about what they are actually testing, and how that is compiled into a final score. All I can gather from their performance test page is that their CPU test "executes complex mathematical calculations involving compression, encryption and physics simulations". How relevant is that to any given user who might be browsing the web, or gaming, or editing photos, or rendering in 3D software? Probably not very.

And it could be considered a similar case for this hierarchy chart at Tom's, even in the context of gaming performance. They base the numbers here on benchmark results from just a handful of games that, on average, tend to be more demanding on the CPU than most game releases. So are the results representative of the typical gaming performance that most will experience? Probably not. Nor do I expect these results to necessarily be representative of performance in future games. And I know a few of these titles show disproportionately larger gains from 3D V-Cache than what the vast majority of titles do. So I would fully expect the 5800X3D to top this chart. But outside a handful of titles, it's probably not going to be noticeably better for gaming than other processors costing significantly less.

As for a 12900KS supposedly being 71% faster than a 5800X3D in single-threaded performance according to Passmark (based on 1 sample), anyone who's looked at a review of the processor could tell you that's complete nonsense, and doesn't even come close to representing reality. There's clearly something wrong there.

And for overclocking, recent high-end CPUs don't tend to overclock much, as Intel and AMD perform binning at the factory and design the processors to reach much closer to their limits at stock. So the lack of overclocking support for the 5800X3D isn't likely to affect much.
An analogy for "synthetic" benchmarks might be an automobile engine's horsepower and torque, measured via a dynometer. The actual horsepower and torque transmitted to the ground can be vastly different than the dynometer's measurements. The vehicles 0-60 time, for example, can vary with the same engine because of the automobile's weight, driveline, gear ratio's, tire sizes, road conditions, etc. But that doesn't make the horsepower and torque ratings meaningless. It's a valid and required way for manufacturers to provide performance numbers of an engine. The same applies to most benchmarking applications for measuring CPU performance.

So, if I want to look at application performance in some applications, I might head over to Puget Systems website and look through their reviews. The interesting thing is that over time the single-thread performance numbers on Passmark usually match up well to the CPU rankings from Puget's test results.

Plus, we do benchmark our scientific applications, and the Passmark single-thread performance is usually an accurate indicator for the expected performance of a given CPU running our applications on a well-balanced workstation (i.e. well-chosen CPU, cooling, memory, disk, and graphics components.)

Having a balanced system is very important. As a well-heeled tech enthusiast, each year I usually build several high-performance systems with carefully selected components. I change out parts and measure with various benchmarks and our own applications to get an idea of what works best. The results from these benchmarks are usually well above the average Passmark single and multithreaded scores for a given CPU. And, BTW, I've had much more luck teasing performance out of Alder Lake CPUs than I have had with Zen3 CPUs.

That said, the 5800X3D now has 5 samples, which I admit is still not enough to produce a good performance estimate and it could be that the submissions were from computers with poorly balanced components. The 5800X3D, now has a single-thread performance of 2655. The 5800x's score is now only 31% higher and the 12900ks is only 61% higher.
 
The 5800X3D, now has a single-thread performance of 2655. The 5800x's score is now only 31% higher and the 12900ks is only 61% higher.
That's still completely unrepresentative of reality. Look at any review. Depending on what software is being tested, the 5800X3D's application performance typically ranges anywhere from being nearly identical to the 5800X, to being around 6% behind in the worst-case scenarios. That aligns with the difference in boost clocks between the two processors. Workloads that don't benefit from the additional cache can fall behind by up to that amount. Anything more than that doesn't make much sense. And overclocking is not increasing the 5800X's single-threaded performance by more than a few percent.

And while the different architectural design of Intel's processors makes the results more variable depending on what software is being tested, there aren't many outliers that come close to that difference, let alone that being the norm. And the 12900KS pretty much does not overclock beyond it's stock single-threaded boost clocks, since it's already clocked near its limits from the factory.

More samples seemingly aligning with those results just supports the suggestion that the benchmark is not always going to be representative of real-world performance. Or perhaps the handful of tested systems were just prebuilts with bad RAM configurations or something.
 
  • Like
Reactions: guru7of9 and King_V

M42

Reputable
Nov 5, 2020
99
48
4,560
That's still completely unrepresentative of reality. Look at any review.

More samples seemingly aligning with those results just supports the suggestion that the benchmark is not always going to be representative of real-world performance. Or perhaps the handful of tested systems were just prebuilts with bad RAM configurations or something.
Yes, it could be a poorly configured system. Or, it could be that there is a pretty wide variation in the quality of the 5800X3D. If reviewers got their sample from AMD they were likely cherry-picked to show the best possible performance.

So let's say that the 5800X3D is 6% slower than the 5800X in single-thread performance. The 5800x scores 3,486 with a whopping 4314 samples, so it should be very accurate. So doing the calculation, 3,486 * 0.94x (100% - 6%) of that is 3,277. And since the 12900ks scores 4,308, it makes the 12900ks "only" 31.5% faster in single-threaded performance.

Fair enough?