Overclock.net › Forums › AMD › AMD CPUs › How much does lack of L3 cache impact performance?
New Posts  All Forums:Forum Nav:

How much does lack of L3 cache impact performance?

post #1 of 13
Thread Starter 
I know that it isn't an across the board thing. And I don't expect a cut and dried answer. With the new APU's not having any how is real world performance affected? Should we ever expect to see any in an APU? Will we see more cores before L3? Or L3 before more cores? Or will we see more cores and L3 if AMD ever makes a successor to the FX line?
post #2 of 13
Check this
http://www.extremetech.com/computing/100583-analyzing-bulldozers-scaling-single-thread-performance/3


http://superuser.com/questions/317771/is-the-cache-size-or-number-of-cores-more-important-when-weighing-cpu-performanc

for AMD, 4 are the keywords
The Future is Fusion(look for HSA)
Edited by PontiacGTX - 3/11/14 at 5:54am
post #3 of 13
Quote:
With the new APU's not having any how is real world performance affected?

Not by much, if at all in most cases. This is in stark contrast to K10, where there were often distinctive gaps between the Athlon II's and Phenom II's at the same clocks.
Quote:
Should we ever expect to see any in an APU?

No. L3 cache takes up a lot of die space, and doesn't grant much of a performance benefit. This would increase the die size pretty substantially, making the chips more expensive to produce. It wouldn't make sense from a financial or practical standpoint. Further down the line, maybe post-Carrizo, who knows.
Quote:
Will we see more cores before L3? Or L3 before more cores? Or will we see more cores and L3 if AMD ever makes a successor to the FX line?

Right now there is no chance for L3$, or more cores. Leaked maps show that Carrizo & Toronto are still quad-core max, so nothing more than that for the foreseeable future.
post #4 of 13
Thread Starter 
Quote:
Originally Posted by NaroonGTX View Post

Not by much, if at all in most cases. This is in stark contrast to K10, where there were often distinctive gaps between the Athlon II's and Phenom II's at the same clocks.
No. L3 cache takes up a lot of die space, and doesn't grant much of a performance benefit. This would increase the die size pretty substantially, making the chips more expensive to produce. It wouldn't make sense from a financial or practical standpoint. Further down the line, maybe post-Carrizo, who knows.
Right now there is no chance for L3$, or more cores. Leaked maps show that Carrizo & Toronto are still quad-core max, so nothing more than that for the foreseeable future.


I am planning on doing a build in April. Probably FM2+ board and either the Richland Athlon 760K with discrete graphics or the Kaveri 7850K. Thinking of the HSA. I'm not a big gamer.
post #5 of 13
Get kaveri.you will get a nice cpu but a bit overpriced cpu for its non-HSA performance
post #6 of 13
Personally I'd get an Athlon 760K + HD 7770/7790 as it would be better performance and not much more than a 7850K smile.gif CPU performance for Kaveri isn't a huge step. However you'd have that FM2+ platform for Excavator later on smile.gif It'll pave the way if HSA takes off.
post #7 of 13
Regarding L3.

L3 makes a significant difference in performance if it is substantially faster than system RAM (higher bandwidth and lower latency). Unfortunately bulldozer/piledriver architecture represents a regression in L3 performance compared to the K10 architecture. When we factor in the regression in L3 performance, combined with the improved memory controller (can run higher speeds/bandwidth, lower effective latency), the delta in performance between the L3 cache and just using system RAM instead, is minimal on PD.

Unfortunately, when the CPU is sharing a memory controller with the GPU, the performance of the system memory (from the CPUs perspective) tanks, and there would be significant advantages to having L3 available to the CPU in this configuration. Pretty silly eh? On FX, L3 isn't important because the system memory is dedicated and the L3 is too slow to offer advantages. Where L3 would have the best chance of improving performance, is on the platform that doesn't have room on the die for it.


In contrast, the L3 performance on the Intel platform, is comparable to the L2 performance on the AMD platform, as such, L3 makes a BIG difference in performance on a haswell chip.


When you factor in that the 760K will clock higher (5ghz is easy on these), and the HD7770 class of GPU has 5X the video memory bandwidth, Kaveri doesn't look very good. I don't see how HSA can reduce overhead enough to compensate for these enormous differences in bandwidth except in very limited fringe workloads. I don't expect to see it competitive until it's running quad channel system RAM on DDR4.
Edited by mdocod - 3/11/14 at 1:23pm
     
CPUMotherboardGraphicsRAM
FX-8350 990X EVO R2.0 Sparkle GTX460 768MB ballistix tactical 2 x 8GB 
Hard DriveHard DriveOptical DriveCooling
Toshiiba THNSNH 256GB Enterprise RE3 1TB Asus BD combo drive Artic A30 
OSMonitorMonitorMonitor
Manjaro Linux Samsung 21.5" LCD E2009WFP E2009WFP 
PowerCase
Seasonic G 550W Modular Fractal Design Core 3500 
CPUMotherboardGraphicsRAM
FX-6300, 4.7 GHZ@1.43V GA-970A-UD3P GTX 460 768MB Mixed DIMMs. 2x4GB + 2x8GB @ 1600-8-8-8 
Hard DriveHard DriveHard DriveHard Drive
Toshiba THNSNH 19nm 256GB 1TB Spinpoint F3 WD RE3 1TB WD RE3 1TB 
Optical DriveCoolingOSOS
yes CM Seidon 120V SolydK OpenSuse 13.1 
OSOSMonitorMonitor
Linux Mint 9-32 bit // Linux Mint 17-64 bit  Manjaro Xfce Samsung 21.5" HannsG 21.5" sideways! 
KeyboardPowerCaseMouse
Sticky ATNG Rosewill Green 630W NZXT Gamma Basic Microsoft corded 
CPUMotherboardGraphicsRAM
A10-6800k 4.8GHZ @ 1.375V, 1.2GHZ iGPU Gigabyte GA-F2A85XN-WIFI HD8670D Ripjaws 
Hard DriveOptical DriveCoolingOS
Seagate ST1000DM003 Asus BC-12B1ST/BLK/B/AS Zalman CNPS5X Linux Mint 15 
MonitorKeyboardPowerCase
LG IPS224V-PN Logitec K360 FSP 400W Aurum S 80+ gold Prodigy 
Mouse
logitec M235 
  hide details  
Reply
     
CPUMotherboardGraphicsRAM
FX-8350 990X EVO R2.0 Sparkle GTX460 768MB ballistix tactical 2 x 8GB 
Hard DriveHard DriveOptical DriveCooling
Toshiiba THNSNH 256GB Enterprise RE3 1TB Asus BD combo drive Artic A30 
OSMonitorMonitorMonitor
Manjaro Linux Samsung 21.5" LCD E2009WFP E2009WFP 
PowerCase
Seasonic G 550W Modular Fractal Design Core 3500 
CPUMotherboardGraphicsRAM
FX-6300, 4.7 GHZ@1.43V GA-970A-UD3P GTX 460 768MB Mixed DIMMs. 2x4GB + 2x8GB @ 1600-8-8-8 
Hard DriveHard DriveHard DriveHard Drive
Toshiba THNSNH 19nm 256GB 1TB Spinpoint F3 WD RE3 1TB WD RE3 1TB 
Optical DriveCoolingOSOS
yes CM Seidon 120V SolydK OpenSuse 13.1 
OSOSMonitorMonitor
Linux Mint 9-32 bit // Linux Mint 17-64 bit  Manjaro Xfce Samsung 21.5" HannsG 21.5" sideways! 
KeyboardPowerCaseMouse
Sticky ATNG Rosewill Green 630W NZXT Gamma Basic Microsoft corded 
CPUMotherboardGraphicsRAM
A10-6800k 4.8GHZ @ 1.375V, 1.2GHZ iGPU Gigabyte GA-F2A85XN-WIFI HD8670D Ripjaws 
Hard DriveOptical DriveCoolingOS
Seagate ST1000DM003 Asus BC-12B1ST/BLK/B/AS Zalman CNPS5X Linux Mint 15 
MonitorKeyboardPowerCase
LG IPS224V-PN Logitec K360 FSP 400W Aurum S 80+ gold Prodigy 
Mouse
logitec M235 
  hide details  
Reply
post #8 of 13
L3 cache runs at the speed of CPU-NB. Unlike Phenom II, you don't see many FX chip users running high NB clocks since it isn't needed as much to take advantage of RAM speed. I can confirm what mdocod said about L3 cache performing worse with bulldozer. Even with 3GHz NB my FX-4100 had worse L3 latency than my Phenom IIs with 3GHz NB. Now I'm running a 760K and I don't even miss the L3 because my RAM runs so fast. L3 on die also adds quite a bit of heat. 750K (or 760K if it's on sale) is an obvious choice for a gaming platform at $80 or less. 7850K costing twice as much doesn't make sense unless for some reason you really don't want a dedicated GPU. The 65/W45W Kaveri is a really good deal though for an HSA platform.
Light Gamer #1
(12 items)
 
Light Gamer #2
(10 items)
 
 
CPUMotherboardGraphicsGraphics
i3-6100 ASRock Fatal1ty Z170 Gaming K4/D3 MSI r7850 TFIII OC MSI r7850 TFIII OC 
RAMHard DriveCoolingOS
G,Skill TridentX OCZ Vector 150 TPC-612 Windows 10 
KeyboardPowerCaseMouse
G105 Rosewill Capstone 550M HAF XB M570 
CPUMotherboardGraphicsRAM
Athlon 860K MSI A88X-G45 MSI GTX 460 G.Skill RipjawsZ 
Hard DriveCoolingOSKeyboard
PNY Optima Hyper T4 Windows 10 Devastator 2 
PowerCase
Rosewill Valens DIYPC Ranger-R4-R 
  hide details  
Reply
Light Gamer #1
(12 items)
 
Light Gamer #2
(10 items)
 
 
CPUMotherboardGraphicsGraphics
i3-6100 ASRock Fatal1ty Z170 Gaming K4/D3 MSI r7850 TFIII OC MSI r7850 TFIII OC 
RAMHard DriveCoolingOS
G,Skill TridentX OCZ Vector 150 TPC-612 Windows 10 
KeyboardPowerCaseMouse
G105 Rosewill Capstone 550M HAF XB M570 
CPUMotherboardGraphicsRAM
Athlon 860K MSI A88X-G45 MSI GTX 460 G.Skill RipjawsZ 
Hard DriveCoolingOSKeyboard
PNY Optima Hyper T4 Windows 10 Devastator 2 
PowerCase
Rosewill Valens DIYPC Ranger-R4-R 
  hide details  
Reply
post #9 of 13
It doesn't seem like FM2/+'s lack of L3 cache makes much difference in benchmarks...except discrete gaming.

http://www.xbitlabs.com/articles/cpu/display/amd-a10-a8-a6_7.html#sect0

I think there's still an argument to be made for the Athlon 750k/760k based upon price - the $30/$40 saved over FX-4xxx/6xxx can bump you up a video card tier - and future Kaveri/Carrizo Athlon upgrades. Nevertheless, with a video card better than one you can use in dual graphics with an APU, the FX series should still return higher fps.
Grassroots
(14 items)
 
  
CPUMotherboardGraphicsRAM
FX-8320 @ 4.4ghz Asus M5A97 R2.0 Sapphire Nitro R9 380 4gb 2x4gb Samsung @ 1866 CL9 
Hard DriveHard DriveCoolingOS
Samsung Spinpoint F3 1TB Samsung 850 Evo 250gb Coolermaster Hyper 212 Evo Windows 7 Ultimate 
MonitorMonitorPowerCase
Nixeus NX-VUE24A Asus VH222H EVGA SuperNova G2 750w Antec 300 Illusion 
MouseAudio
E-BLUE Cobra II Cambridge Soundworks Microworks 
  hide details  
Reply
Grassroots
(14 items)
 
  
CPUMotherboardGraphicsRAM
FX-8320 @ 4.4ghz Asus M5A97 R2.0 Sapphire Nitro R9 380 4gb 2x4gb Samsung @ 1866 CL9 
Hard DriveHard DriveCoolingOS
Samsung Spinpoint F3 1TB Samsung 850 Evo 250gb Coolermaster Hyper 212 Evo Windows 7 Ultimate 
MonitorMonitorPowerCase
Nixeus NX-VUE24A Asus VH222H EVGA SuperNova G2 750w Antec 300 Illusion 
MouseAudio
E-BLUE Cobra II Cambridge Soundworks Microworks 
  hide details  
Reply
post #10 of 13
Quote:
Originally Posted by Captainpabst View Post

It doesn't seem like FM2/+'s lack of L3 cache makes much difference in benchmarks...except discrete gaming.

http://www.xbitlabs.com/articles/cpu/display/amd-a10-a8-a6_7.html#sect0

I think there's still an argument to be made for the Athlon 750k/760k based upon price - the $30/$40 saved over FX-4xxx/6xxx can bump you up a video card tier - and future Kaveri/Carrizo Athlon upgrades. Nevertheless, with a video card better than one you can use in dual graphics with an APU, the FX series should still return higher fps.[/quo


Thanks for the link. Info like this is sorely lacking. This is pretty telltale, I was hoping for more from the X4-750/760K.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD CPUs
Overclock.net › Forums › AMD › AMD CPUs › How much does lack of L3 cache impact performance?