AMD Ryzen 9 9950X Review: We've Seen This Before...

[Nobina]

I'm not saying they are false, since performance depends on used software, hardware, settings...

However, it has been proven multiple times that Ryzen performance on certain workloads is excellent. That means Zen5 is not small but very big improvement from Zen4. If that does not show on TS workloads, it still doesn't mean Zen5 is "flop".



That your POV. However, if you want advancement, you also have to accept sacrifices. Good example:

x86-64 compilers make only SSE2 or higher FPU code, not x87. You can still use x87, but compilers do not make x87 code. So, if 64-bit software is used, why bother with x87 performance at all? Well, someone wants Pifast to run much faster but price is either hgher CPU price or lower performance on other areas. Choose:

- Superior performance in Pifast, more expensive CPU OR less performance onother areas
- Worse performance on Pifast but better performance on 64-bit software AND no additional CPU cost

You cannot please everyone.

So you think TS reviews obscure software and in reality people generally use AVX512?

 

[HardReset]

As I said, that software doesn't exist. No one should buy a new CPU hoping that someday it might be faster in future software. I'd be especially cautious in this case since AVX512 has been around for a while and it hasn't seen wide adoption in consumer or even professional software.

Then no-one should have bought Pentium 4 (poor x87, good SSE2) or Athlon64 (useless 64-bit support at that time). See? CPUs Always have some aspects that May be good in future, or not. Basically you're saying nobody should ever buy CPU.

As for AVX512, there are only poor excuses not to use it. Here's one: https://x.com/IanCutress/status/1327358373373898752

You want fast CPU, you use AVX512 software. If you are good with slow CPUs, why you complain about lack of speed on Zen5? Double standards at it's finest.

So you think TS reviews obscure software and in reality people generally use AVX512?

Not necessarily. But again, it has already been proven that Zen5 is much faster than Zen4, threfore there is nothing badly wrong on Zen5. However if your software is bad, then you may not get full advantage of speedup Zen5 offers.

Problem is, AVX512 has been around nearly decade already and there are compilers that allow to take advantage of it. It's not that developers Cannot use it. It's about they don't Want to use it. Therefore blaming AMD is just plain stupid.

 

[Squid Surprise]

Problem is, AVX512 has been around nearly decade already and there are compilers that allow to take advantage of it. It's not that developers Cannot use it. It's about they don't Want to use it. Therefore blaming AMD is just plain stupid.

I think the real problem is that AMD decided to optimize their CPU for software very few want to use then…
Your argument tends to fall flat - if you build a new CPU, it needs to run the stuff people use - and run it better than the previous generation CPU… otherwise, it’s a flop.

 

[HardReset]

I think the real problem is that AMD decided to optimize their CPU for software very few want to use then…
Your argument tends to fall flat - if you build a new CPU, it needs to run the stuff people use - and run it better than the previous generation CPU… otherwise, it’s a flop.

You basically understood idea.

My argument is valid. AMD could do new CPU that runs old software much better than Zen5 while maintaining almost every advantage Zen5 has. But, then CPU would be bigger and price would be higher. Then you would complain even more about high price, right?

To put it another way: Zen5 is a flop if because it don't run old software much better than Zen4. But even if it would do, then it would be flop because it isn't enough future proof. Even if Zen5 would be future proof AND runs old software much better, it would be flop because it would be too expensive.

Overall: Zen5 would be flop, no matter what.

Now that is out of the way, another POV. It takes billions of dollars and around 5 years to create complex CPU like Ryzen 9950X. However it takes few seconds from developer to add AVX512 support on software. You really say AMD is one to blame ""

 

[Squid Surprise]

Overall: Zen5 would be flop, no matter what.

Yep!

Now that is out of the way, another POV. It takes billions of dollars and around 5 years to create complex CPU like Ryzen 9950X. However it takes few seconds from developer to add AVX512 support on software. You really say AMD is one to blame ""

Yes, they are to blame. As you’ve already stated, AVX512 has existed for awhile now - and very few are using it. Why waste years and billions of dollars exclusively supporting it instead of supporting the software people actually use?

 

[Nobina]

Not necessarily. But again, it has already been proven that Zen5 is much faster than Zen4, threfore there is nothing badly wrong on Zen5. However if your software is bad, then you may not get full advantage of speedup Zen5 offers.

Problem is, AVX512 has been around nearly decade already and there are compilers that allow to take advantage of it. It's not that developers Cannot use it. It's about they don't Want to use it. Therefore blaming AMD is just plain stupid.

It has been proven here in this TS review that Zen5 is actually slower in some cases, most of the time there's a few % increase. And since you said software tested in TS review is "not necessarily" obscure, and I agree, then it counts. I think the vast majority of people use software TS tested and not software from the review you have linked, so why did AMD decide to make CPUs specifically for software almost nobody uses?

 

[cfenton]

Then no-one should have bought Pentium 4 (poor x87, good SSE2) or Athlon64 (useless 64-bit support at that time). See? CPUs Always have some aspects that May be good in future, or not. Basically you're saying nobody should ever buy CPU.

As for AVX512, there are only poor excuses not to use it. Here's one: https://x.com/IanCutress/status/1327358373373898752

You want fast CPU, you use AVX512 software. If you are good with slow CPUs, why you complain about lack of speed on Zen5? Double standards at it's finest.



Not necessarily. But again, it has already been proven that Zen5 is much faster than Zen4, threfore there is nothing badly wrong on Zen5. However if your software is bad, then you may not get full advantage of speedup Zen5 offers.

Problem is, AVX512 has been around nearly decade already and there are compilers that allow to take advantage of it. It's not that developers Cannot use it. It's about they don't Want to use it. Therefore blaming AMD is just plain stupid.

What do you not understand about that software not existing for my work? I don't care why it doesn't exist, I care that it doesn't exist.

Do you really think that if developers could increase performance dramatically by recompiling their software that they wouldn't bother? Either AVX512 doesn't provide any advantages to the software or it would be a major undertaking to implement it.

As for the no one should buy any new CPU bit, that's just nonsense. Look at the improvements AMD made between Zen 3 and Zen 4. That is what I want to see. Zen 4 was much faster in everything, not just a few obscure enterprise focused workloads.

 

[Inthenstus]

They should have just added more X3D chips to the 7000 series and called it a day. They lied about their marketing, and no one is happy. These are DOA, unless they match the price of the 7000 series. Sorry Lisa, you’re becoming Pat. Don’t be Pat.

 

[HardReset]

Yes, they are to blame. As you’ve already stated, AVX512 has existed for awhile now - and very few are using it. Why waste years and billions of dollars exclusively supporting it instead of supporting the software people actually use?

Very few are using it because developers thing it sucks because there has been a few Intel CPUs where it really sucked. Now it's 2024 and it doesn't suck on Zen5.

Why support it? Well, It has already shown it can bring triple digit performance gains. But you say we should stick with small double digit performance gains instead? You are allowed to have your opinion but don't be surprised if someone disagrees.

It has been proven here in this TS review that Zen5 is actually slower in some cases, most of the time there's a few % increase. And since you said software tested in TS review is "not necessarily" obscure, and I agree, then it counts. I think the vast majority of people use software TS tested and not software from the review you have linked, so why did AMD decide to make CPUs specifically for software almost nobody uses?

You never update your software? I doubt you have much software (not updated or anything) you used like 5 years ago? Right?

Again, adding AVX512 support on software takes around few seconds from developer, so this software problem should be very easy to fix. But still you think AMD should stick with low gains because wasting those few seconds is way too much to demand from developers? You are allowed to have your opinion but don't be surprised if someone disagrees.

 

[HardReset]

What do you not understand about that software not existing for my work? I don't care why it doesn't exist, I care that it doesn't exist.

Then get better software? I doubt you never update your software.

Do you really think that if developers could increase performance dramatically by recompiling their software that they wouldn't bother? Either AVX512 doesn't provide any advantages to the software or it would be a major undertaking to implement it.

Now let's see:

"Either AVX512 doesn't provide any advantages to the software". It has already proven several times on this thread that AVX512 provides huge advantages. Even triple digit gains have been seen.

"Major undertaking to implement it" means, in practice, change compiler flags. That takes literally few seconds, just adjust compiler flags to create AVX512 code instead just, say, AVX2.

As for the no one should buy any new CPU bit, that's just nonsense. Look at the improvements AMD made between Zen 3 and Zen 4. That is what I want to see. Zen 4 was much faster in everything, not just a few obscure enterprise focused workloads.

But Zen5 is ten times faster than Zen4 if used properly. Again, you don't need "enterprise focused" workloads, you need software that takes advantage of AVX512. And adding that support takes, as already said, few seconds.

 

[Shirley Dulcey]

A report on another site suggests that the problem, or at least a big part of it, is that the Windows support for Zen 5 isn't quite ready for prime time. Testing on Linux has given much more favorable results for Zen 5 and Ryzen 9000, even when using it to run Windows games.

 

[Squid Surprise]

Then get better software? I doubt you never update your software.

I shouldn’t have to! Nor should anyone else! A new CPU should run everything better… not just new stuff.

Now let's see:

"Either AVX512 doesn't provide any advantages to the software". It has already proven several times on this thread that AVX512 provides huge advantages. Even triple digit gains have been seen.
"Major undertaking to implement it" means, in practice, change compiler flags. That takes literally few seconds, just adjust compiler flags to create AVX512 code instead just, say, AVX2.

Then why hasn’t it? And it doesn’t matter what the reason is… the point is that it hasn’t been widely implemented- even if the reason is an Intel conspiracy, the FACT is that few things use it… so why spend billions on a CPU series that only performs using it?

But Zen5 is ten times faster than Zen4 if used properly. Again, you don't need "enterprise focused" workloads, you need software that takes advantage of AVX512. And adding that support takes, as already said, few seconds.

You sound like Apple telling users “they’re holding their phone wrong” when they don’t get reception…

A report on another site suggests that the problem, or at least a big part of it, is that the Windows support for Zen 5 isn't quite ready for prime time. Testing on Linux has given much more favorable results for Zen 5 and Ryzen 9000, even when using it to run Windows games.

Even if this is true (I doubt it), then the fault is with AMD. Windows is a known entity and has been around for decades. Any new CPU better be optimized for the latest version upon its release!