a look at (ry)Zen's IPC - from Excavator to KabyLake

[zMeul]

Just now, tlink said:

its very logical that programs optimize for the clearly superior product for that workload.

that's a lie

programs do not optimise for nothing, the same compilers that show poor gaming performance for Zen are used when shown good multithreaded workload performance for Zen

compilers generate exes that work on the x86 arch, nothing more, nothing less - you can instruct compilers to use / not use a particular instruction set

but those instruction sets are already established - if AMD hasn't implemented them correctly, that's on AMD

[3DOSH]

14 minutes ago, zMeul said:

that's a lie

programs do not optimise for nothing, the same compilers that show poor gaming performance for Zen are used when shown good multithreaded workload performance for Zen

compilers generate exes that work on the x86 arch, nothing more, nothing less - you can instruct compilers to use / not use a particular instruction set

but those instruction sets are already established - if AMD hasn't implemented them correctly, that's on AMD

I'll leave this here http://www.anandtech.com/show/5448/the-bulldozer-scheduling-patch-tested/3. 
Non-Mild gains with only windows optimizations. You were saying ??

[zMeul]

9 minutes ago, 3DOSH said:

I'll leave this here http://www.anandtech.com/show/5448/the-bulldozer-scheduling-patch-tested/3. 
Non-Mild gains with only windows optimizations. You were saying ??

what are we talking here?!!?

are we comparing compilers with Windows' scheduler now?! what is this!? you give me apples and want orange juice?!

 

Bulldozer had a particular problem with the scheduling (not exactly) and they fixed it with MS' help - the perf gains are minimalistic and fringe case scenarios

it didn't made Bulldozer better, not then, and not 5y later

[3DOSH]

12 minutes ago, zMeul said:

what are we talking here?!!?

are we comparing compilers with Windows' scheduler now?! what is this!? you give me apples and want orange juice?!

 

AMD had a particular problem with the scheduling and they fixed it with MS' help - the perf gains are minimalistic and fringe case scenarios

it didn't made Bulldozer better, not then, and not 5y later

Quote

What the compiler cannot do:
– Understand dependencies between data with indirect addressing
– Non-integer or complex strength reduction
– Optimize by Unrolling/Merging/Blocking with
• Calls to functions or subroutines
• I/O statements or calls within the code
– Optimize variables with values known only at run-time
• Compilers are generally conservative, i.e. will not optimise if strong
risk of obtaining incorrect results unless forced to by the user
either with compiler directives or options.
• Still sometimes happens that codes give wrong results, even with
minor optimisations
 

I'm not knowledgeable enough when it comes to compilers unlike you which you seem to be guru at so I don't get into it, but little googling turned up this. 
Zen might and might not be the above but you are saying it's absolutely not. I don't have anything to back up my argument since i don't really know, on the other hand you must have evidence to be this sure.

[Okjoek]

I hope AMD can get their crap together. I was considering either an R3 because it's only 129 dollars or an APU whenever they come out which is probably towards the end of the year.

 

If not I'll just have to go with Kaby Lake or FM2+

 

 

[zMeul]

16 minutes ago, 3DOSH said:

I'm not knowledgeable enough when it comes to compilers unlike you which you seem to be guru at so I don't get into it, but little googling turned up this. 
Zen might and might not be the above but you are saying it's absolutely not.

here's what you need to take away from that list you quoted:

Quote

Compilers are generally conservative, i.e. will not optimise if strong risk of obtaining incorrect results unless forced to by the user either with compiler directives or options

what does that mean: generic compilers produce exes that work on any x86 machine and meets instruction set dependencies

you can force compilers to generate exes a certain way, but that problem is on the developer's head

 

---

 

ps: I'm not a guru, I'm knowledgeable and have a degree in computer programming (very old and might even worth shit today)

I haven't done any serious programming maybe for more than a decade (I think I stopped soon after returning from the army), but the principles of computer programming have not changed

[3DOSH]

2 minutes ago, zMeul said:

here's what you need to take away from that list you quoted:

what does that mean: generic compilers produce exes that work on any x86 machine and meets instruction set dependencies

you can force compilers to generate exes a certain way, but that problem is on the developer's head

I'm sorry and what is AMD doing by working with game devs to optimize code? what you said above correct ?

[orangecat]

I have a hard time believing this. If Zen was as bad as this post says it is intel wouldn't have adjusted pricing. intel obviously knows Zen is competitive. Saying its only 6% faster than their previous part seems stupid when just about every other benchmark seems to prove otherwise. i don't know enough about how exactly this benchmarking was done or what exactly they were benchmarking but we already know Zen is behind in a few small insignificant areas but it's nothing that will hurt end users.

 

I'm not saying Zen is perfect because it's not but it's a serious step up for AMD and it's what AMD has always been good at. They never really give intel the business but they do give them competition. If AMD can make a CPU that is within 5% performance of a 6900K in 95% of workloads I think it's safe to say it's good enough. Does AMD have the best CPU on the market? No they don't. I don't think any of these benchmarks show Zens strong point only it's weaknesses. I'm sure intel has a few instructions on their chips that AMD might laugh at but it all comes down to what you expect from the chip and what you actually use it for. AMD is marketing their new CPUs as desktop chips for consumers so if its a bit behind when it comes to scientific work who cares because that wasn't its main selling point.

[zMeul]

5 minutes ago, 3DOSH said:

I'm sorry and what is AMD doing by working with game devs to optimize code? what you said above correct ?

and with AMD going to work with game developers is going to achieve what?! 1-2-3 FPS more on average!?

once you build the CPU that's it, there's no magic to make it better; the gaps that you see now in games will still be there

 

also, the same thing AMD said when discussing Bulldozer issues, that things will get better .. well ?!?! they didn't!

[orangecat]

15 hours ago, TOMPPIX said:

why is everyone so fixed on the single core performance, it's multi thread performance destroys every i7 out there in the same price range.

Because single threaded performance still matters allot in 2017. not everything is multithreaded and better single threaded helps multi threaded performance too.

[zMeul]

4 minutes ago, Kilobytez95 said:

I have a hard time believing this. If Zen was as bad as this post says it is intel wouldn't have adjusted pricing.

and yet, Intel hasn't adjusted any pricing

what were you reading? Trump news channel? 

 

Quote

Saying its only 6% faster than their previous part

you read wrong and I strongly suggest you go back to the OP and read again

I said it's 6% faster on top of their initial prediction of 40% - that's totalling 46% faster on IPC compared to Bulldozer - and the data shows exactly that

[orangecat]

Just now, zMeul said:

and yet, Intel hasn't adjusted any pricing

what were you reading? Trump news channel? 

I heard they changed the price of their X99 chips. I'm not 100% sure if it's true but obviously intel has reacted in some way. if Zen couldn't come close to anything they has intel would have completely ignored Zen and acted like it never happened.

[zMeul]

Just now, Kilobytez95 said:

I heard they changed the price of their X99 chips. I'm not 100% sure if it's true but obviously intel has reacted in some way.

you heard wrong

 

also, I updated my previous post with other wrong points you said

[3DOSH]

Just now, zMeul said:

and with AMD going to work with game developers is going to achieve what?! 1-2-3 FPS more on average!?

once you build the CPU that's it, there's no magic to make it better; the gaps that you see now in games will still be there

 

also, the same thing AMD said when discussing Bulldozer issues, that things will get better .. well ?!?! they didn't!

Yet again you make assumptions with nothing to back them up. where did you get 1-2-3 FPS estimate. 
What makes you think what was said about Bulldozer well apply to Zen when its total departure from Bulldozer. 
 

[orangecat]

2 minutes ago, zMeul said:

you heard wrong

 

also, I updated my previous post with other wrong points you said

Ok or you could have just replied that way i don't need to go looking for your edited post.

[StealthArsenal]

To be clear, Intel hasn't price dropped any of their processors.  What you might have heard and/or saw on line was Microcenter dropped processors well below MSRP during the launch of Ryzen.  I picked up an 1800x and a 7700k (7700K was $299 USD).  They have now pushed the 7700k back to $319. 

[zMeul]

5 minutes ago, Kilobytez95 said:

Ok or you could have just replied that way i don't need to go looking for your edited post.

I've already edited the post before I saw your next reply

[zMeul]

8 minutes ago, 3DOSH said:

Yet again you make assumptions with nothing to back them up. where did you get 1-2-3 FPS estimate. 

in the chart and link from Anand you quoted me with  wtf mate

 

Quote

What makes you think what was said about Bulldozer well apply to Zen when its total departure from Bulldozer

because it's a general truth you don't seem to understand / comprehend / accept

[3DOSH]

4 minutes ago, zMeul said:

in the chart and link from Anand you quoted me with  wtf mate

 

because it's a general truth you don't seem to understand / comprehend / accept

 

That was windows optimizations not program specific. Face palm
If im not understand / comprehend / accept its because your argument is weak to say the least.

[Imglidinhere]

1 hour ago, zMeul said:

I don't know what you think I'm painting but you are wrong

 

the tests I picked are FMA related and all of them (Bullet, Himeno, NBody, Linpack) show the same thing - ZEN doesn't perform good

add these to those in the OP

 

and yes, if you go and read all the data and charts, it all adds up to what I said and shown - Zen barely reaches Haswell in IPC

And exactly what picture are you trying to paint? Kaby Lake has worse IPC than Haswell by about 5%. It's already been proven. The only reason KL is faster is due to the higher frequencies. That's all. Besides, the instruction sets within each Intel CPU allow it to run more efficiently and, by relation, faster in those application scenarios than without. Hell, some programs flat out drop dead and don't know what to do with the processor without certain instruction sets present.

 

Also the first graph shows the instruction sets that KL has over Haswell. If you used that as a baseline, then you're claiming that Kaby Lake has an IPC of 100% over Haswell which is as fucking FALSE as the earth being flat. Your source's graphs show that Zen is exactly where it needs to be. The ABSOLUTE IPC means it includes the clockspeed increases. It's the ONLY way that KL can manage a 30% boost over Haswell, and we all know that Haswell is NOT slow. Intel took a step backward from IPC performance when they jumped from Haswell to Skylake, and IPC improvements are literally 0% going from SL to KL. That's also known too.

[79wjd]

6 hours ago, Glenwing said:

 

No, the scores here show Kaby Lake to be about 17.3% higher than Haswell. It's just because they're percentages, not absolute numbers, and percentages depend on reference point.

I know, I accounted for it being percentages -- Kaby Lake still isn't anywhere near 17% ahead of Haswell.

[Glenwing]

3 minutes ago, djdwosk97 said:

I know, I accounted for it being percentages -- Kaby Lake still isn't anywhere near 17% ahead of Haswell.

I think I quoted you by accident lul

 

And yeah it's definitely a bit off, I just wanted to clarify the percentage issue.

[zMeul]

10 minutes ago, Imglidinhere said:

-

what the fuck are you on about

 

the topic title is "a look at (ry)Zen's IPC"

and it compares Zen, Haswell, KabyLake IPC versus Excavator - that's it!

and that's exactly what it shows - Zen has a 46% IPC gain over Excavator, but still lacks behind Haswell (as predicted many months ago) and further still behind KabyLake

 

if you're seeing something else, that's on you

[Imglidinhere]

2 minutes ago, zMeul said:

what the fuck are you on about

 

the topic title is "a look at (ry)Zen's IPC"

and it compares Zen, Haswell, KabyLake IPC versus Excavator - that's it!

and that's exactly what it shows - Zen has a 46% IPC gain over Excavator, but still lacks behind Haswell (as predicted many months ago) and further still behind KabyLake

 

if you're seeing something else, that's on you

Annnnnnnd......? No really I'm confused as to what the point of your argument is here. This article shows that it's right where it's supposed to be... and yet most of the graphs show that it's beating Haswell and KL in just about every test save for like... three or four.  

[zMeul]

4 minutes ago, Imglidinhere said:

Based on what findings? No really I'm confused as to what the point of your argument is here. This article shows that it's right where it's supposed to be... and yet most of the graphs show that it's beating Haswell and KL in just about every test save for like... three or four.  

really?! I wonder what are you reading

 

explain to me how it beats Haswell in IPC, "in just about every test" (you said) when the absolute IPC compared to Excavator sums up to: