Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[tamz_msc]

He wasn't talking about POV-Ray. At least they are good FP benchmarks, which is why they became popular in the first place. Though they are not the only benches I want to see, I'd much rather see them than Geekbench since it doesn't mirror anything I do.

The vast majority of everyday computing involves the kind of workloads that Geekbench runs, not renderers like Cinebench. To say that Geekbench doesn't mirror anything you do with a computing device is profoundly ignorant.

 

[DrMrLordX]

To say that Geekbench doesn't mirror anything you do with a computing device is profoundly ignorant.

I don't queue up a series of random tasks that are barely identified by the software publisher and then run them in specific sequence.

 

[igor_kavinski]

Geekbench serves as a common denominator for comparing the performance of dissimilar architectures. It isn't able to predict the maximum performance potential of an architecture that could be unlocked in an optimized application. For that purpose, something like a benchmark script using LibreOffice/Firefox etc. could be used. An open source benchmark comprised of real world open source applications would be a great incentive for the different CPU vendors to put in the best optimizations for their own architectures and that could serve as a really good, real world benchmark.

 

[DrMrLordX]

An open source benchmark comprised of real world open source applications would be a great incentive for the different CPU vendors to put in the best optimizations for their own architectures and that could serve as a really good, real world benchmark.

Good benchmark software costs time to produce. If there's no financial incentive to continue maintaining a piece of software, it's difficult to get anyone to put in the effort necessary to keep such a benchmark up-to-date. There's always the Phoronix Test Suite, but it has enough odd choices that reasonable parties should object to it being a good performance indicator for many PC users.

Right now there's no one catch-all solution to benchmarking software choices. Certainly not Geekbench 5. A true performance picture of Raptor Lake will require several benchmarks to be run on it.

 

[Asterox]

The vast majority of everyday computing involves the kind of workloads that Geekbench runs, not renderers like Cinebench. To say that Geekbench doesn't mirror anything you do with a computing device is profoundly ignorant.

Still, it is classic CPU benchmark or not your everyday app.

One interesting example, can you specify or explain why such a result with FastStone app?This is still not clear to me, nor was it clear to the people who tested it.This is not high multithread test, any we all now that in the test used Intel 8/16 CPU has highest Singlecore Geekbench score.

Page 31 - Zen 2 APUs/"Renoir" discussion thread

Page 31 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[moinmoin]

An open source benchmark comprised of real world open source applications would be a great incentive for the different CPU vendors to put in the best optimizations for their own architectures and that could serve as a really good, real world benchmark.

https://openbenchmarking.org/ does just that.

 

[igor_kavinski]

https://openbenchmarking.org/ does just that.

I know but can end users create their own tests there? There is no Firefox browsing functions test and LibreOffice test completes in 5 seconds or less.

 

[tomatosummit]

90% do, but the few don't.

Heck it stands for Instructions Per Clock. Hence the confusion. Performance per clock confuses nobody. And it doesn't care things like memory or cache, it's all taken into account. It's arguing about horsepower versus top speed/acceleration/towing capacity. The latter is what really matters.

AMD and Intel also does silly things like 19% gain. Really? Are you sure it's not 18.5% or 20%? In the sane days they used to say 15-20%. I am going to outdo your 19% with 21%.

Give me a break, seriously.

I would disagree even further.
IPC is missappropriated and missused but so is performance per clock and even clockspeed. Sure they're all very interesting things to discuss but in the end they're all academic. Especially when it comes down to a product instead of just an architecture for example.
Performance is A1 and then Power efficiency as that's what has to be designed around are mostly what's really important to the end customer.

 

[moinmoin]

I know but can end users create their own tests there?

Yes, it's completely open.

"OpenBenchmarking.org is completely extensible. Any free software project can take advantage of any of these features for their own testing purposes by simply writing a test profile / suite that compiles with our public specifications."

OpenBenchmarking.org Features - OpenBenchmarking.org

openbenchmarking.org

And I don't think it actually requires you to be a project to be able to write your own tests. Not sure about the specifications that have to be met though.

 

[tomatosummit]

I don't queue up a series of random tasks that are barely identified by the software publisher and then run them in specific sequence.

That's a pretty good flaw with lots of these aggregate useful software benchmarks. The most appropriate ones would probably just cover a slew of the top three most popular adobe suite programs and that'd be dandy.
The most common thing most users do is sit there with the computer idle while they think about what to do next, you know, if the computer isn't busy doing some hard core processing because that's the kind of thing you wait around for.

The thing with cinebench is it's more like a horsepower test than anything else.
It might not be repressentative of common use cases but it is representative of what a cpu will be doing when it's under full tilt, even other software.
Also everything has a cinebench run so you can compare against everyting else, although with how rapid the c23 release was I'd argue it's going against that now.

While we're on car comparisons, geek bench is the south african car safety regulations test, video games are track times and alder/raptorlake are hybrid vehicles.

 

[tamz_msc]

I don't queue up a series of random tasks that are barely identified by the software publisher and then run them in specific sequence.

Again spoken out of ignorance:

https://www.geekbench.com/doc/geekbench5-cpu-workloads.pdf

The integer workloads are some of the most common tasks that people do with computers. I'd go so far as to argue that many of the FP workloads in Geekbench are more relevant to everyday computing than SPEC, whose FP suite involves programs that were developed with HPC in mind.

 

[mikk]

Cinebench R23 from Raichu (13900K apparently)

https://twitter.com/x/status/1556288587985227777

 

[shady28]

Geekbench serves as a common denominator for comparing the performance of dissimilar architectures. It isn't able to predict the maximum performance potential of an architecture that could be unlocked in an optimized application. For that purpose, something like a benchmark script using LibreOffice/Firefox etc. could be used. An open source benchmark comprised of real world open source applications would be a great incentive for the different CPU vendors to put in the best optimizations for their own architectures and that could serve as a really good, real world benchmark.

As far as relevant to real users, the best benchmark is going to be the applications that most users use.

Geekbench is pretty well recognized as a decent bench because it uses a lot of well known algorithms in different areas and benchmarks based off of that.

Like most things, people don't pay attention to detail though. You can look inside Geekbench results and see if something is throwing off the results.

For example - without even revealing what these two chips are, you can see it. The single core AES performance of one chip is almost entirely responsible for the higher single core performance here.

Now, see if you can guess what two generations are being compared :

 

[shady28]

I would disagree even further.
IPC is missappropriated and missused but so is performance per clock and even clockspeed. Sure they're all very interesting things to discuss but in the end they're all academic. Especially when it comes down to a product instead of just an architecture for example.
Performance is A1 and then Power efficiency as that's what has to be designed around are mostly what's really important to the end customer.

Totally agree. IPC, performance/clock, these things are irrelevant to any kind of real investigation of how well a particular platform and CPU perform.

And the platform matters. You can get a 10% difference on the same chip/ram/storage just by swapping the motherboard out. You can get 20% differences in some loads just from RAM.

This just highlights the fact that you can't test "IPC" if other components are involved in the test. I have not seen true core "IPC" type tests - instruction speed tests that fit in the L1 cache - used regularly in over two decades. Sites discarded them because they weren't relevant to users. And they *are not relevant to users*, at all.

And now, everyone talks about IPC, using benchmarks that are most decidedly not showing you IPC. Clown world.

 

[Makaveli]

This PBO 5800X it's getting 1718 at 4.6 Ghz Boost. You can extrapolate the difference on Speed and IPC and it should be a match to the Raptor Lake ST score.

View attachment 65469

I do 5.05Ghz with PBO which you can see in the maximum frequency below

 

[szrpx]

Cinebench R23 from Raichu (13900K apparently)

https://twitter.com/x/status/1556288587985227777

Not bad for ~250 W, E cores definitely pulling more weight in that MT score. Wonder how much it can improve with a tune. I think the 12900KS can do 31K at ~ 250 W. So 13900K/KS should be able to do a decent bit better with some TLC.

Curious to see how 7950x stacks up.

 

[shady28]

Not bad for ~250 W, E cores definitely pulling more weight in that MT score. Wonder how much it can improve with a tune. I think the 12900KS can do 31K at ~ 250 W. So 13900K/KS should be able to do a decent bit better with some TLC.

Curious to see how 7950x stacks up.

From that link :



250W is a bit much. I have an Arctic Liquid Freezer II 240 AIO and it can keep up with about 190W continuous. It's certainly fine for short periods of 240W or so but we're talking like 10s.

I think a very good 360mm AIO is probably good to 225-250W, but past that you'll need custom loop. Most people with a 360 AIO will wind up overheating at 250W.

Of course, this is a 13900K so that's not your norm, such a user likely has at least a 280 and probably a 360 AIO. I'd like to see what happens to a 13700K or 13600K when power unlocked on a more mundane AIO or good air cooler.

 

[Abwx]

...

From that link :

View attachment 65522

The comparison with RPL@250W and 5950X is wrong since the latter does 26196 pts at stock, that s 38% or so, not 48%...

With RPL@350W the difference is 55%, not 67%..

Intel Core i9-12900K, i7-12700K & i5-12600K im Test: Benchmarks in Anwendungen

Intel Alder Lake im Test: Benchmarks in Anwendungen / Ein rekordverdächtiger Testumfang / Windows 10 vs. Windows 11

www.computerbase.de

 

[shady28]

...



The comparison with RPL@250W and 5950X is wrong since the latter does 26196 pts at stock, that s 38% or so, not 48%...

With RPL@350W the difference is 55%, not 67%..

Intel Core i9-12900K, i7-12700K & i5-12600K im Test: Benchmarks in Anwendungen

Intel Alder Lake im Test: Benchmarks in Anwendungen / Ein rekordverdächtiger Testumfang / Windows 10 vs. Windows 11

www.computerbase.de

So you can change the numbers if you pick a different sites numbers, using a different motherboard, different RAM, different cooling and so on.

What a surprise. Was there a point to your post?

 

[Abwx]

So you can change the numbers if you pick a different sites numbers, using a different motherboard, different RAM, different cooling and so on.

What a surprise. Was there a point to your post?

Computerbase tests are done at manufacturer s stock settings, including the RAM speed and latencies, so my point is that the numbers exhibited by Raichu for the 5950X are plain wrong.

 

[Markfw]

So you can change the numbers if you pick a different sites numbers, using a different motherboard, different RAM, different cooling and so on.

What a surprise. Was there a point to your post?

Well, I think that with all these variables, unless its a respected review site with all the details, and its a fair comparison, that all of these should be taken with a great deal of salt. I personally don't believe most of them, as clickbait.

 

[shady28]

Computerbase tests are done at manufacturer s stock serttings, including the RAM speed and latencies, so my point is that the numbers exhibited by Raichu for the 5950X are plain wrong.

Riiiigght. So they do what, pair a 12900K with DDR5-4800 JEDEC using an air cooler. And then power unlock it. Probably using an ASRock Z690 Phantom Gaming motherboard too.

Yep, that'll give us a meaningful result. You got me.

Anything else you'd like to add?

 

[Abwx]

Riiiigght. So they do what, pair a 12900K with DDR5-4800 JEDEC using an air cooler. And then power unlock it. Probably using an ASRock Z690 Phantom Gaming motherboard too.

Yep, that'll give us a meaningful result. You got me.

Anything else you'd like to add?

Nice deflection since we re talking of the 5950X score at Computerbase compared to the score from Raichu, think that Computerbase use stock RAM settings for the 5950X, so your argument is backfiring...

Edit : So the numbers are 38% and 55% at 250 and 350W respectively, not 48% and 67%, the latter are made up numbers but if you agree that a forum should be used as a misleading tool that s another story...

 

[shady28]

Nice deflection since we re talking of the 5950X score at Computerbase compared to the score from Raichu, think that Computerbase use stock RAM settings for the 5950X, so your argument is backfiring...

Edit : So the numbers are 38% and 55% at 250 and 350W respectively, not 48% and 67%, the latter are made up numbers but if you agree that a forum should be used as a misleading tool that s another story...

We? Who's we? You decided to use computerbase.

What if we use TPU? We know what their config is. DDR4-3600 C14 on AMD. DDR5-6000 C36 on Intel Alder Lake.

And the score here is, 25813.

 

[desrever]

The vast majority of everyday computing involves the kind of workloads that Geekbench runs, not renderers like Cinebench. To say that Geekbench doesn't mirror anything you do with a computing device is profoundly ignorant.

Problem with Geekbench is it doesn't test much that actually stress a high end desktop level CPU enough to matter. Who cares if any of the micro benchmarks that make up Geekbench is fast when you use a desktop where it wouldn't be noticeable at all in everyday use? You can notice your frame rate is 10% lower, you can't really notice if takes 100 nanoseconds more to open a PDF. A lot of the Geekbench benchmarks aren't all that useful and they are so small it barely matters. There is very few use cases on desktop where the CPU is actually stressed and they are very specific to the person using the computer.

You can say everyone does web browsing so it's the best benchmark but I'm pretty sure nobody can tell 1 cpu from another with a blind test just doing web browsing. Gaming, compiling, rendering, transcoding, productivity, etc are pretty specific to people, this means people need to check benchmarks that matter to them. Geekbench doesn't really allows for that and just does their own weights of a whole bunch of things and gives a score.