Why the Bulldozers L3 cache is so slow ? - Page 2 - Overclock.net

Forum Jump: 
Reply
 
Thread Tools
post #11 of 31 Old 05-13-2013, 06:42 AM
Audiophile
 
Join Date: Mar 2011
Location: Wallachia
Posts: 1,829
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 107
I think the L3 adds it s latency to the pocesing chain.The RAM latency is already there and the L3 adds even more.

Most probable the RAM or a part of it is used/reserved as L3 cache ,when there isn t one on the die for a multicore CPU liek the Athlons.

Looking at the toms hardware we see that in multithreaded apps the Athlon 2 does better than the Phenom 2 .

What about FM2 motherboards with 2 APU sockets and GDDR5 on-board AMD ?! smile.gif

Maybe i am to sarcastic sometimes smile.gif
Dromihetes is offline  
Sponsored Links
Advertisement
 
post #12 of 31 Old 05-13-2013, 06:50 AM
Ryzen from the ashes for EPYC yo
 
DaveLT's Avatar
 
Join Date: Dec 2012
Posts: 8,602
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 278
Quote:
Originally Posted by Dromihetes View Post

I think the L3 adds it s latency to the pocesing chain.The RAM latency is already there and the L3 adds even more.

Most probable the RAM or a part of it is used/reserved as L3 cache ,when there isn t one on the die for a multicore CPU liek the Athlons.

Looking at the toms hardware we see that in multithreaded apps the Athlon 2 does better than the Phenom 2 .
No smile.gif
L3 cache is a memory area for high speed buffer, RAM just isn't that. Read up on cache on google. If L3 was really harming the latency then L2 would have never been used either
DaveLT is offline  
post #13 of 31 Old 05-13-2013, 07:03 AM
Bod mods rock
 
Bitemarks and bloodstains's Avatar
 
Join Date: Apr 2006
Location: Durham-England-UK
Posts: 22,701
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 16 Post(s)
Liked: 567
Bitemarks and bloodstains is online now  
Sponsored Links
Advertisement
 
post #14 of 31 Old 05-13-2013, 07:07 AM
Audiophile
 
Join Date: Mar 2011
Location: Wallachia
Posts: 1,829
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 107
Quote:
Originally Posted by Bitemarks and bloodstains View Post

Please keep this thread to L3 cache and not your opinions on Bulldozer.

I mean no disrespect ,but the thread is about Bulldozer L3 cache though smile.gif

What about FM2 motherboards with 2 APU sockets and GDDR5 on-board AMD ?! smile.gif

Maybe i am to sarcastic sometimes smile.gif
Dromihetes is offline  
post #15 of 31 Old 05-13-2013, 07:08 AM
New to Overclock.net
 
Join Date: Sep 2007
Posts: 6,632
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 485
Quote:
Originally Posted by senna89 View Post

Why the Bulldozers L3 cache is so slow ? i saw a very minimal difference from FX4300 and APU 5800K in performance.
What is the cause ?

Technically the FX-4300 is Piledriver and not Bulldozer, but you are right. Comparing the A10-5800k and the FX-4300 show the L3 cache has very little impact. In fact with the exception of gaming, the A10-5800K has a slight (insignificant?) lead in many benchmarks.

http://www.anandtech.com/bench/Product/675?vs=700

AMD needs to publish a Business Manual called "How Not to Launch a Product: Learn from our Mistakes and Succeed!".
AMD probably wont tho, because that would be a guaranteed money maker.

PassMark System Score: Passmark Rating 5,710, CPU Mark 19,985
CPU-Z Validation: LINK AIDA64: LINK
Cinebench15: LINK Geekbench3 scores: LINK Geekbench4.1 scores: LINK
UserBenchmarks: CPU: 105.5%, GPU: 111.7% MEM: 127.9%
BinaryDemon is offline  
post #16 of 31 Old 05-13-2013, 07:09 AM
Technician / Systems Builder
 
malmental's Avatar
 
Join Date: Jun 2010
Location: TYS
Posts: 8,181
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 377
Quote:
Originally Posted by Dromihetes View Post

Quote:
Originally Posted by Bitemarks and bloodstains View Post

Please keep this thread to L3 cache and not your opinions on Bulldozer.

I mean no disrespect ,but the thread is about Bulldozer L3 cache though smile.gif
agreed..

(A.S.) - IT | CompTIA A+, Net+ | MCDST | PMP | DSCE (Dell)
US Navy Vet - GS-E (PO3)

daily driver | surround boi

Clubz... (Click to show)
malmental is offline  
post #17 of 31 Old 05-13-2013, 12:26 PM
New to Overclock.net
 
Artikbot's Avatar
 
Join Date: Jul 2010
Location: Catalunya
Posts: 14,957
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 3 Post(s)
Liked: 613
Quote:
Originally Posted by Dromihetes View Post

I think the L3 adds it s latency to the pocesing chain.The RAM latency is already there and the L3 adds even more.

It doesn't work that way.

Processes don't go from RAM to L3 to L2 to L1 registers. The CPU works all the way around. Fetches data from L1, if it finds it, excellent because it is extremely fast. If it misses the L1, tries to fetch it from L2. If it is there, fine because it's still very fast. If it doesn't goes into L3. If it's there, well, good enough, because L3 is still not that bad. If it misses, then all hell and heaven shakes because it has to retrieve it directly from the system memory, which is INCREDIBLY slow compared to the on-die caches.

Those cache misses produce very large performance losses to the ALU/FPU, because it means they have to literally stop what they are doing to fetch the data from a slow source.

Take a look at this, the Overviews section is explained in very plain english: http://en.wikipedia.org/wiki/CPU_cache
Grmadness likes this.

Chimp Challenge Participant 

Artikbot is online now  
post #18 of 31 Old 05-13-2013, 12:58 PM
New to Overclock.net
 
MrPerforations's Avatar
 
Join Date: Sep 2010
Location: Milton Keynes,England
Posts: 1,876
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 78
hello,
thought it was just about dividing data sets over different cpu, both ways?

have old hard drive with 2mb cache and they shunt 3gb/s and there old drives, so it can do the job with that little ram.

Chimp Challenge Participant 

MrPerforations is offline  
post #19 of 31 Old 05-13-2013, 03:51 PM
Audiophile
 
Join Date: Mar 2011
Location: Wallachia
Posts: 1,829
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 107
Quote:
Originally Posted by Artikbot View Post

It doesn't work that way.

Processes don't go from RAM to L3 to L2 to L1 registers. The CPU works all the way around. Fetches data from L1, if it finds it, excellent because it is extremely fast. If it misses the L1, tries to fetch it from L2. If it is there, fine because it's still very fast. If it doesn't goes into L3. If it's there, well, good enough, because L3 is still not that bad. If it misses, then all hell and heaven shakes because it has to retrieve it directly from the system memory, which is INCREDIBLY slow compared to the on-die caches.

Those cache misses produce very large performance losses to the ALU/FPU, because it means they have to literally stop what they are doing to fetch the data from a slow source.

Take a look at this, the Overviews section is explained in very plain english: http://en.wikipedia.org/wiki/CPU_cache


Thanks man for the links.

I took a look here and there and i have come to the conclusion that in spite of the latency ,a bigger L2 instead of an additional L3 is not a bad strategy from AMD , to lower TDP and costs.

It also seems that Intel does the thing different as L1 and L2 are used different as same data can be in both of them (which with AMD doesnt happen) as such an L3 is more necessary for them.

For desktop pc-s the L3 cache doesn t seem to be such a big deal at least with Bulldozer.

The L3 in FX is faster than RAM , but either it s not getting filled ,either the information stored there never or rarely gets to be used smile.gif

Maybe thats why AMD took some time for Vishera to optimize L1 and L2 only.

What about FM2 motherboards with 2 APU sockets and GDDR5 on-board AMD ?! smile.gif

Maybe i am to sarcastic sometimes smile.gif
Dromihetes is offline  
post #20 of 31 Old 05-13-2013, 05:03 PM
Ryzen from the ashes for EPYC yo
 
DaveLT's Avatar
 
Join Date: Dec 2012
Posts: 8,602
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 278
Cache is always filled. It simply means that either the L3 cache is missed so frequently or that everything can be found in the L2
DaveLT is offline  
Reply

Quick Reply

Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may post new threads
You may 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
Trackbacks are Off
Pingbacks are Off
Refbacks are Off