Overclockers Australia Forums

OCAU News - Wiki - QuickLinks - Pix - Sponsors  

Go Back   Overclockers Australia Forums > Manufacturer-specific Forums > AMD x86 CPUs and chipsets

Notices


Sign up for a free OCAU account and this ad will go away!
Search our forums with Google:
Reply
 
Thread Tools
Old 8th November 2015, 10:46 AM   #121
Apokalipse
Member
 
Apokalipse's Avatar
 
Join Date: May 2006
Location: Melbourne
Posts: 4,058
Default

Quote:
Originally Posted by mAJORD View Post
As much as I agree it's not a true 8 core processor, the fact you can't quantify how fast a 'core' must be in isolation makes this lawsuit completely April fools worthy. In fact I almost thought it was an old April fools joke you accidentally dug up
It is a true 8 core processor, and it can process 8 threads at a time.

If people are disappointed with its performance, they obviously didn't consider that there are more factors that determine performance than the number of cores.
__________________
Q: How many Turnbull NBN Review Panel members does it take to change a lightbulb?
A: None. Their cost-benefit analysis concluded that living in darkness provides better value to the public.
Apokalipse is offline   Reply With Quote

Join OCAU to remove this ad!
Old 8th November 2015, 11:01 AM   #122
newlife
Member
 
newlife's Avatar
 
Join Date: Oct 2012
Location: Adelaide, SA
Posts: 2,079
Default

Quote:
Originally Posted by Apokalipse View Post
It is a true 8 core processor, and it can process 8 threads at a time.

If people are disappointed with its performance, they obviously didn't consider that there are more factors that determine performance than the number of cores.
Yeah and afaik poor cache design plays a large part in that also it's surprising how many people think it's not a core without a fpu
newlife is offline   Reply With Quote
Old 8th November 2015, 11:08 AM   #123
Apokalipse
Member
 
Apokalipse's Avatar
 
Join Date: May 2006
Location: Melbourne
Posts: 4,058
Default

Quote:
Originally Posted by newlife View Post
Yeah and afaik poor cache design plays a large part in that also it's surprising how many people think it's not a core without a fpu
If core count was determined by the number of FPU's, then some systems would have zero CPU cores.
__________________
Q: How many Turnbull NBN Review Panel members does it take to change a lightbulb?
A: None. Their cost-benefit analysis concluded that living in darkness provides better value to the public.
Apokalipse is offline   Reply With Quote
Old 8th November 2015, 11:35 AM   #124
newlife
Member
 
newlife's Avatar
 
Join Date: Oct 2012
Location: Adelaide, SA
Posts: 2,079
Default

Exactly my point but a lot of people think otherwise
newlife is offline   Reply With Quote
Old 8th November 2015, 12:43 PM   #125
Sologuy
Member
 
Sologuy's Avatar
 
Join Date: Oct 2015
Location: Northern Tasmania
Posts: 169
Default

Back to the title of this thread...
Perhaps this article from some 5 yrs ago will go some way to explain why bulldozer/piledriver cpus perform poorly in benchmarks compared to the opposition products from Intel... and from this, its easy to see why AMD's market share is poor compared to Intel today.

This excerpt from the article is particularly worrying for AMD and any other cpu manufacturer for that matter:

Quote:
"Many software developers think that the compiler is compatible with AMD processors, and in fact it is, but unbeknownst to the programmer it puts in a biased CPU dispatcher that chooses an inferior code path whenever it is running on a non-Intel processor," Fog writes, "If programmers knew this fact they would probably use another compiler. Who wants to sell a piece of software that doesn't work well on AMD processors?"
Even though this article is over 5 yrs old, how do we know for sure that these problems are non-existent today?
Sologuy is offline   Reply With Quote
Old 8th November 2015, 1:11 PM   #126
Kurosaki
Member
 
Kurosaki's Avatar
 
Join Date: Dec 2005
Location: Australia
Posts: 4,569
Default

Quote:
Originally Posted by Apokalipse View Post
It is a true 8 core processor, and it can process 8 threads at a time.
By your definition Intel should advertise i7 CPUs are 8 core and 12 core CPUs?

Quote:
Originally Posted by Apokalipse View Post
If core count was determined by the number of FPU's, then some systems would have zero CPU cores.
Most of those are quite ancient. I struggling to think of a desktop CPU past 486 that didn't have a 1:1 ratio of ALU:FPU or had an option of a "math co-processor". An argument can be made its an natural evolution in technology, much the same way we our expectation and definition of certain things are different to how it once was.
Kurosaki is offline   Reply With Quote
Old 8th November 2015, 3:20 PM   #127
Apokalipse
Member
 
Apokalipse's Avatar
 
Join Date: May 2006
Location: Melbourne
Posts: 4,058
Default

Quote:
Originally Posted by Kurosaki View Post
By your definition Intel should advertise i7 CPUs are 8 core and 12 core CPUs?
Hyperthreading is not even remotely the same as how Bulldozer works.

With hyperthreading, there is one physical core, but when one thread is running, it does not use 100% of the execution resources at once (even with OoOE allowing it to run more instructions ahead of time from the same thread). So Hyperthreading allows another thread to operate in the core, using some of the unused execution resources, but not to the full performance of another core, since both threads might want to use the same execution units.

With Bulldozer, however, it actually does physically have extra hardware. It is two cores that just happen to share some resources that aren't normally shared. To say that a module has two cores is correct, but there are of course performance penalties to sharing those resources when both cores are being used.

Quote:
Originally Posted by Kurosaki View Post
Most of those are quite ancient. I struggling to think of a desktop CPU past 486 that didn't have a 1:1 ratio of ALU:FPU or had an option of a "math co-processor". An argument can be made its an natural evolution in technology, much the same way we our expectation and definition of certain things are different to how it once was.
But if you want to define what a "core" is, you should define it by the number of integer units. Every "core" has one. It doesn't matter how old they are.

Mind you, the FPU is Bulldozer can and does allow 2x128-bit instructions to run at once, so there's that.
It's not so simple to say that an 8-core Bulldozer CPU has 4 FPU's, because each FPU is actually made of two 128-bit FPU pipelines that can work independently, or combine to execute 256-bit FPU instructions.
__________________
Q: How many Turnbull NBN Review Panel members does it take to change a lightbulb?
A: None. Their cost-benefit analysis concluded that living in darkness provides better value to the public.

Last edited by Apokalipse; 8th November 2015 at 3:36 PM.
Apokalipse is offline   Reply With Quote
Old 8th November 2015, 3:39 PM   #128
mAJORD
Member
 
mAJORD's Avatar
 
Join Date: Jun 2002
Location: Griffin , Brisbane
Posts: 8,589
Default

Quote:
Originally Posted by Apokalipse View Post
Hyperthreading is not even remotely the same as how Bulldozer works.

With hyperthreading, there is one physical core, but when one thread is running, it does not use 100% of the execution resources at once (even with OoOE allowing it to run more instructions ahead of time from the same thread). So Hyperthreading allows another thread to operate in the core, using some of the unused execution resources, but not to the full performance of another core, since both threads might both want to use the same execution units.

With Bulldozer, however, it actually does physically have extra hardware. It is two cores that just happen to share some resources that aren't normally shared. To say that a module has two cores is correct, but there are of course performance penalties to sharing those resources.

But if you want to define what a "core" is, you should define it by the number of integer units. Every "core" has one. It doesn't matter how old they are.

Mind you, the FPU is Bulldozer can and does allow 2x128-bit instructions to run at once, so there's that.

CMT is is in essence an extension of hyperthreading. Whilst Intel shares all execution hardware between 2 threads, the orignal Bulldozer shared only some the Hardware - i.e The Front end (Fetch+decode) and Floating point hardware (FP scheduler and FPU).

Hence Not a 'true' Octocore from a hardware pt of view.
mAJORD is offline   Reply With Quote
Old 8th November 2015, 4:22 PM   #129
Cask
Member
 
Cask's Avatar
 
Join Date: Nov 2011
Location: Gold Coast
Posts: 757
Default

Quote:
Originally Posted by n2p4u View Post
But how could you expect other people to buy AMD, but it is not good for yourself. What the point in AMD even bothering if they do make a better product but you still will not buy?
Sadly bad experience usually doesn't bring anyone back regardless of the good reviews.

If your car - let's say it's a Toyota - had too many problems from day one and then 5 Toyota's later, you still didn't have a reliable brand new car; would you ever buy another Toyota or would you move over to a different brand?
__________________
X79S-UP5-WiFi | i7 3820 @ 4.5GHz (water) | 64GB RAM |
Pentax K-x | Canon 60D (ML) |
OCAU Pentax Users | Graphics Designer | Trades
Cask is offline   Reply With Quote
Old 9th November 2015, 12:08 AM   #130
demiurge3141
Member
 
Join Date: Aug 2005
Location: Melbourne 3073
Posts: 828
Default

Funny how AMD's marketing actually put themselves in a worse light. How embarrassing is it to have your six-core beaten up by a dual core i3?
demiurge3141 is online now   Reply With Quote
Old 9th November 2015, 12:39 AM   #131
mAJORD
Member
 
mAJORD's Avatar
 
Join Date: Jun 2002
Location: Griffin , Brisbane
Posts: 8,589
Default

Sorry I don't follow. Can you point to the Q1 2013 i3 that beats an FX6350 in MT?
mAJORD is offline   Reply With Quote
Old 9th November 2015, 9:53 AM   #132
Perko
Member
 
Perko's Avatar
 
Join Date: Aug 2011
Location: NW Tasmania
Posts: 1,773
Default

Quote:
Originally Posted by mAJORD View Post
Sorry I don't follow. Can you point to the Q1 2013 i3 that beats an FX6350 in MT?
Why just MT?
__________________
Main: Corsair 600C/Enermax Platimax 850/MSI X99A SLI Plus/i7-5820k @ 4GHz/Be Quiet! Silent Loop 280mm AIO/Corsair Vengeance 3000MHz Low Profile/Galax GTX 1080 EXOC/Creative SB-Z + Beyer T70/Samsung 950 Pro 512GB + 1TB Caviar Black/Win 10 Pro/MK Fission - MX Whites/Mionix Naos 3200/AOC AGON AG271QG
Perko is online now   Reply With Quote
Old 9th November 2015, 10:54 AM   #133
mAJORD
Member
 
mAJORD's Avatar
 
Join Date: Jun 2002
Location: Griffin , Brisbane
Posts: 8,589
Default

Quote:
Originally Posted by Perko View Post
Why just MT?
Because the entire discussion revolves around the issue of core count's
mAJORD is offline   Reply With Quote
Old 9th November 2015, 12:38 PM   #134
Perko
Member
 
Perko's Avatar
 
Join Date: Aug 2011
Location: NW Tasmania
Posts: 1,773
Default

Quote:
Originally Posted by mAJORD View Post
Because the entire discussion revolves around the issue of core count's
No, it's based around what counts as a "true" core. I'm guessing that the plaintiff's ill-thought-out case will be based on overall performance in real world applications, and that the shared FPU in each module means that each module is not really two cores in a colloquial sense. As much as AMD are probably bullet-proof in a straight up false advertising case, the god-awful performance of Bulldozer could probably move a jury in the right court room with the right lawyers, assuming it's even possible to go to a jury, (not all that familiar with the US civil courts), but judges generally aren't known for their tech savvy either.

I really hope that Zen puts as much of this bullshit behind us as possible. I noticed that they have some 14nm samples done, but it looks like they're for GPU. Still, if I were in the market for non-X99 kit at the moment, I would be waiting, Z170 is overpriced to shit and current AMD offerings would struggle to beat much that the average OCAU member would have in their cases from 2011 or so.
__________________
Main: Corsair 600C/Enermax Platimax 850/MSI X99A SLI Plus/i7-5820k @ 4GHz/Be Quiet! Silent Loop 280mm AIO/Corsair Vengeance 3000MHz Low Profile/Galax GTX 1080 EXOC/Creative SB-Z + Beyer T70/Samsung 950 Pro 512GB + 1TB Caviar Black/Win 10 Pro/MK Fission - MX Whites/Mionix Naos 3200/AOC AGON AG271QG
Perko is online now   Reply With Quote
Old 9th November 2015, 3:05 PM   #135
mAJORD
Member
 
mAJORD's Avatar
 
Join Date: Jun 2002
Location: Griffin , Brisbane
Posts: 8,589
Default

Quote:
Originally Posted by Perko View Post
No, it's based around what counts as a "true" core. I'm guessing that the plaintiff's ill-thought-out case will be based on overall performance in real world applications, and that the shared FPU in each module means that each module is not really two cores in a colloquial sense. As much as AMD are probably bullet-proof in a straight up false advertising case, the god-awful performance of Bulldozer could probably move a jury in the right court room with the right lawyers, assuming it's even possible to go to a jury, (not all that familiar with the US civil courts), but judges generally aren't known for their tech savvy either.

I really hope that Zen puts as much of this bullshit behind us as possible. I noticed that they have some 14nm samples done, but it looks like they're for GPU. Still, if I were in the market for non-X99 kit at the moment, I would be waiting, Z170 is overpriced to shit and current AMD offerings would struggle to beat much that the average OCAU member would have in their cases from 2011 or so.
Yes, and that naturally progresses to number of cores advertised , and the resultant vs expected total performance with those cores fully utilized.


Quote:
As much as AMD are probably bullet-proof in a straight up false advertising case, the god-awful performance of Bulldozer could probably move a jury in the right court room with the right lawyers,
No, because performance is not defined anywhere. It's the same old argument from the Ghz war days, just with cores.

With the move to arbitrary model numbers for SKU's and decreasing the emphasis on clockspeed, most people have realized it's a poor indicator. This just needs to happen with core counts also.

The formula to determine performance has become incredibly complex. The following all significantly affect performance:

-uArch IPC (Now not so easily identified with multiple, uARChs of vastly different IPC sharing the same name - e.g current Pentium lines)
-Base frequency
-Turbo frequency
-# cores
-SMT and associated scaling
-software thread count and scaling
-IGP type/performance

It is a mess, with no simple solution from a model naming and marketing pt of view

The only scenario where this would hold up, is if AMD launched their 8 core FX for a higher price than the outgoing 6 core, but with no MT performance advantage across a range of benchmarks. Or worse still, pitched it as an alternative to Intel's 6 core.

Whilst BD was disappointing, and was beaten by its predecessor on occasion in single / lightly threaded tasks, it did hold a sizable enough advantage in MT performance to justify its price.
mAJORD is offline   Reply With Quote
Reply

Bookmarks

Sign up for a free OCAU account and this ad will go away!

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +10. The time now is 1:48 PM.


Powered by vBulletin® Version 3.8.9
Copyright ©2000 - 2017, vBulletin Solutions, Inc.
OCAU is not responsible for the content of individual messages posted by others.
Other content copyright Overclockers Australia.
OCAU is hosted by Micron21!