Overclock.net › Forums › AMD › AMD CPUs › Analysis of Kaveri's Downclocking CPU to 3.0GHz when iGPU is under full load
New Posts  All Forums:Forum Nav:

Analysis of Kaveri's Downclocking CPU to 3.0GHz when iGPU is under full load - Page 3

post #21 of 38
No, though a good benchmark is to try running the training course. It's single-player so you won't have to worry about server connection issues or anything.

Even a basic recording with something like plays.tv or shadowplay will be worth the trouble.
post #22 of 38
Decided to go with the integrated Farcry2 benchmark, these are the results:

Settings: Demo(Ranch Medium), 1920x1080 (60Hz), D3D10, Fixed Time Step(No), Disable Artificial Intelligence(No), Full Screen, Anti-Aliasing(None), VSync(No), Overall Quality(Optimal), Vegetation(High), Shading(High), Terrain(High), Geometry(High), Post FX(High), Texture(High), Shadow(High), Ambient(High), Hdr(Yes), Bloom(Yes), Fire(Very High), Physics(Very High), RealTrees(Very High)

[3GHz]

Average Framerate: 52.70
Max. Framerate: 78.72
Min. Framerate: 33.64


[3.5GHz]

Average Framerate: 53.91
Max. Framerate: 81.68
Min. Framerate: 34.80

[4GHz]

Average Framerate: 51.93
Max. Framerate: 80.13
Min. Framerate: 33.18
Edited by Feimitsu - 8/20/16 at 5:12am
post #23 of 38
I have to agree that 3Dmark is a bad choice for bottlenecks since it focuses on GPU or CPU on each test.
But testing at low-res low-settings to find if it's bottlenecked is bad.

1.If i intentionally test super-low res with low settings of course i'm going to find myself CPU-limited and underclocking the CPU will affect my results a lot.(Physics tests)
2.If i intentionally test super-high res with ultra settings of course i'm going to find myself GPU-limited and underclocking the CPU will not affect my results.(Graphics tests)

For both cases, my choice of settings/games is the cause and the effect is being bottlenecked at CPU/GPU,
Case 1 has already sky-high FPS so who cares if we get 300 or 400fps?
Case 2 is probably unplayable at <15 fps so what's the reason to test it?

This method is only useful if i'm comparing different GPUs on a specific CPU and even then, if under normal conditions(60-150fps) there is no difference, who cares what happens at hundreds of FPS?

The APU is configured to be a balance between those 2 extreme cases and the OP is trying to figure out if the GPU is actually bottlenecked by the CPU's drop in clock speed.

3DMark is a poor choice to test mixed workloads, unless you're running the combined tests, normal games would be the better choice for APUs.

Thanks to Feimitsu above, you can see that under FC2 (1080p high, nice), the CPU is not actually the bottleneck.

Of course you can test games like MMOs that will have a big effect on the CPU, but then you're looking for CPU bottlenecks in CPU bottlenecked games, CRAZY RIGHT?
post #24 of 38
Well I'm not sure if this is relevant or not here but when my A8 6410 apu downclocks it bottlenecks the gpu portion really bad.

The setup is constrained to single channel memory and only get's about 14gb/s for bandwidth w/ ddr3 1866. Max turbo speed or the CPU is 2.4ghz, during gaming it downclocks to 1.6-1.8ghz and my frames dip quite a bit. This is with both high and low settings, regardless of res. So being that a puma cpu (which is faster clock vs clock over kaveri) is dropping that much performance where it is bottlenecking a radeon r5 gpu I'd imagine dropping 700mhz on kaveri would indeed bottleneck a much faster gpu.

At least this has been my experience on games like CS:S
Edited by SoloCamo - 8/20/16 at 6:17am
The Struggle (4k)
(20 items)
 
File Server
(12 items)
 
Lenovo G50-45
(6 items)
 
CPUGraphicsRAMOS
Athlon II X2 250u Nvidia 6150SE  2gb DDR3 1066mhz Windows 10 Home 64 bit 
CPUGraphicsRAMHard Drive
AMD A8-6410 AMD R5 Crucial Ballistix 8GB DDR3L 1866 CAS10 Crucial BX100 250gb 
Optical DriveOS
DVD Windows 10 Home 
  hide details  
Reply
The Struggle (4k)
(20 items)
 
File Server
(12 items)
 
Lenovo G50-45
(6 items)
 
CPUGraphicsRAMOS
Athlon II X2 250u Nvidia 6150SE  2gb DDR3 1066mhz Windows 10 Home 64 bit 
CPUGraphicsRAMHard Drive
AMD A8-6410 AMD R5 Crucial Ballistix 8GB DDR3L 1866 CAS10 Crucial BX100 250gb 
Optical DriveOS
DVD Windows 10 Home 
  hide details  
Reply
post #25 of 38
Quote:
Originally Posted by Feimitsu View Post

Decided to go with the integrated Farcry2 benchmark, these are the results:

Interesting, 3.5 GHz came out better than 3 or 4 GHz. It might be that you're GPU-constrained at that res, though I have my doubts (that is an old title). There is also the issue of canned benchmarks/timedemos doing a poor job of assessing "real world" performance (timedemos are frequently light on the CPU). What kind of FPS do you get in FarCry 2 playing the actual game?

Also, would you be willing to re-run it at a lower res like 1024x768 if possible?

Quote:
Originally Posted by Marios145 View Post

But testing at low-res low-settings to find if it's bottlenecked is bad.

1.If i intentionally test super-low res with low settings of course i'm going to find myself CPU-limited and underclocking the CPU will affect my results a lot.(Physics tests)

When assessing the effects of CPU clockspeed on a title, you generally want to at least start with a CPU-limited scenario so you can see best case/worst case results based on clockspeed. It eliminates outlying factors so you can at least observe a trend. Once you've found the trend, you run the software you actually want to run (read: a game you'd really want to play) at the settings you'd actually use and see if you can replicate the trend.
Quote:
Case 1 has already sky-high FPS so who cares if we get 300 or 400fps?

To identify the trend, and to avoid numbers so small that differences in performance may amount to statistical noise. If result A is 400 fps and result B is 300 fps, we can discount variance/statistical noise, since it's clear that the two results are different. Also, the initial inquiry was mostly academic in nature, so all results are relevant to the discussion.
Quote:
Thanks to Feimitsu above, you can see that under FC2 (1080p high, nice), the CPU is not actually the bottleneck.

. . . well, maybe. Or it's a case that reinforces gapottberg's point that the CPU is effectively at 3 GHz for the entire run thanks to GeAPM throttling. We needed results at a lower res where GPU constraint would be impossible so that differences in CPU performance would be exaggerated.
Quote:
Of course you can test games like MMOs that will have a big effect on the CPU, but then you're looking for CPU bottlenecks in CPU bottlenecked games, CRAZY RIGHT?

MMOs aren't the only games that are heavily-dependent on CPU power. Many UE3 titles are like that, which is why I suggested a f2p title based on UE3 (Dirty Bomb). It doesn't feature a timedemo though, so it would require recording standard gameplay sessions for comparison. If Feimitsu doesn't want to do that, I might have to step in and do it myself, though my 7700k has gotten crash-happy and when my 7870k arrives, I'll be running a GeAPM-disabled UEFI (thanks Stilt) making the issue of throttling moot.
Edited by drmrlordx - 8/20/16 at 3:16pm
post #26 of 38
Okay, since the only data we have thus far is a canned Far Cry 2 benchmark and a synthetic (3DMark), I'll kick in with some "real world" gameplay vids. Files are large-ish so you have been warned. These were recorded using Plays.tv through the Spectre iGPU. There is stuttering from the recording process. Keep your eye on the fps counter and try to ignore the poor video quality. They are unedited, so go ahead and skip past the boring parts near the beginning.

Game: Dirty Bomb

A10-7700k, 1.8 GHz NB, DDR3-1600 CAS9

All CPU clockspeeds are static (no turbo). GeAPM throttling is enabled (2.8 GHz). Game settings are defaults chosen by the game.

800x600

3.4 GHz: https://www.dropbox.com/s/hcs1l70qqsez0p6/2016_08_20_21_39_25-ses.mp4?dl=0

2.8 GHz: https://www.dropbox.com/s/sqt0q5phmxxffbb/2016_08_20_21_19_09-ses.mp4?dl=0

1024x768

3.4 GHz: https://www.dropbox.com/s/r2ibi70kl70tijn/2016_08_20_20_10_44-ses.mp4?dl=0

2.8 GHz: https://www.dropbox.com/s/z5jk38jmge7cgb0/2016_08_20_20_35_19-ses.mp4?dl=0

Make of those what you will. My analysis is that there wasn't much difference between the two settings at 1024x768, but at 800x600, the 3.4 GHz setting did feel a bit smoother during actual gameplay. Recording via Plays.tv really put the hit on the iGPU making 1024x768 nearly unplayable. I also had three mining clients running in the background, but they don't take much CPU time and they aren't touching the Spectre.
post #27 of 38
thy gaming wench
(15 items)
 
server(old gamer)
(10 items)
 
 
CPUMotherboardGraphicsRAM
AMD FX-8350 Vishera 4.0GHz (4.2GHz Turbo) Socke...  ASUS SABERTOOTH 990FX R2.0 AM3+ AMD 990FX SATA...  SAPPHIRE 100354OCL Radeon HD 7870 GHz Edition ... Mushkin Enhanced Redline 16GB (4 x 4GB) 240-Pin... 
Hard DriveOptical DriveCoolingOS
samsung 840 256GB 500GB seagate some cheap asus  Thermaltake CLW0217 Water 2.0 Extreme/All-In-O... winduhs pro 64 
MonitorKeyboardPowerCase
ASUS VS Series VS247H-P Black 23.6" 2ms LED Ba... slick looking logitech for 50ish  RAIDMAX RX-1000AE 1000W ATX12V v2.3 / EPS12V S...  IN WIN Dragon Rider Black 1.0 - 0.8mm SECC Ste... 
MouseMouse PadAudio
some 25 dollar logitech  pfft Logitech 5.1 surround + sub  
CPUMotherboardGraphicsRAM
intel core 2 duo e8400 biostar tpower i45 some old 4000 series ati card that i didnt want... 8gb of gskill ddr2 800 
Hard DriveCoolingOSMonitor
3x1TB in raid z1 and 4x3TB in raidz2 cooler master hyper 212 freenas 8.3.1(freeBSD) headless 
PowerCase
corair 750 watt cooler master centurion 5 
  hide details  
Reply
thy gaming wench
(15 items)
 
server(old gamer)
(10 items)
 
 
CPUMotherboardGraphicsRAM
AMD FX-8350 Vishera 4.0GHz (4.2GHz Turbo) Socke...  ASUS SABERTOOTH 990FX R2.0 AM3+ AMD 990FX SATA...  SAPPHIRE 100354OCL Radeon HD 7870 GHz Edition ... Mushkin Enhanced Redline 16GB (4 x 4GB) 240-Pin... 
Hard DriveOptical DriveCoolingOS
samsung 840 256GB 500GB seagate some cheap asus  Thermaltake CLW0217 Water 2.0 Extreme/All-In-O... winduhs pro 64 
MonitorKeyboardPowerCase
ASUS VS Series VS247H-P Black 23.6" 2ms LED Ba... slick looking logitech for 50ish  RAIDMAX RX-1000AE 1000W ATX12V v2.3 / EPS12V S...  IN WIN Dragon Rider Black 1.0 - 0.8mm SECC Ste... 
MouseMouse PadAudio
some 25 dollar logitech  pfft Logitech 5.1 surround + sub  
CPUMotherboardGraphicsRAM
intel core 2 duo e8400 biostar tpower i45 some old 4000 series ati card that i didnt want... 8gb of gskill ddr2 800 
Hard DriveCoolingOSMonitor
3x1TB in raid z1 and 4x3TB in raidz2 cooler master hyper 212 freenas 8.3.1(freeBSD) headless 
PowerCase
corair 750 watt cooler master centurion 5 
  hide details  
Reply
post #28 of 38
Quote:
Originally Posted by SoloCamo View Post

Well I'm not sure if this is relevant or not here but when my A8 6410 apu downclocks it bottlenecks the gpu portion really bad.

The setup is constrained to single channel memory and only get's about 14gb/s for bandwidth w/ ddr3 1866. Max turbo speed or the CPU is 2.4ghz, during gaming it downclocks to 1.6-1.8ghz and my frames dip quite a bit. This is with both high and low settings, regardless of res. So being that a puma cpu (which is faster clock vs clock over kaveri) is dropping that much performance where it is bottlenecking a radeon r5 gpu I'd imagine dropping 700mhz on kaveri would indeed bottleneck a much faster gpu.

At least this has been my experience on games like CS:S

Playing GTA V on my A8-6410 laptop with a discrete M230 (five times more powerful than the iGPU in the CPU) results in 50% GPU use and 15-20fps. If the CPU remained at 2.4GHz it would perform better but not by much as it's still only a 15W ULV chip.

 

Interestingly an anniversary Pentium performs the same, showing that a dual core is no longer a viable gaming chip.

The girlfriend.
(15 items)
 
The Mistress
(13 items)
 
Media Server
(11 items)
 
CPUMotherboardGraphicsRAM
A8-6410 Lenovo Lancer 4B2 K16.3 R5 128 Shaders/M230 Hynix 8GB DDR3 1600 
Hard DriveHard DriveOSMonitor
Samsung 840 120 GB SSD Seagate Momentus 1TB 5400rmp Win 8.1 CMN1487 TN LED 14" 1366*768 
KeyboardPowerMouseMouse Pad
Lenovo AccuType 2900mAh/41Wh Elan Trackpad/Logitech M90 Super Flower 
Audio
AMD Avalon(Connexant) 
  hide details  
Reply
The girlfriend.
(15 items)
 
The Mistress
(13 items)
 
Media Server
(11 items)
 
CPUMotherboardGraphicsRAM
A8-6410 Lenovo Lancer 4B2 K16.3 R5 128 Shaders/M230 Hynix 8GB DDR3 1600 
Hard DriveHard DriveOSMonitor
Samsung 840 120 GB SSD Seagate Momentus 1TB 5400rmp Win 8.1 CMN1487 TN LED 14" 1366*768 
KeyboardPowerMouseMouse Pad
Lenovo AccuType 2900mAh/41Wh Elan Trackpad/Logitech M90 Super Flower 
Audio
AMD Avalon(Connexant) 
  hide details  
Reply
post #29 of 38
My tests were ran with GeAPM disabled.
post #30 of 38
DIRT 2

Settings: Ultra preset, Post processing set to Medium (High setting takes 20fps alone)

3GHz
1600x900: av_fps="61.129467" min_fps="48.234791"
1920x1080: av_fps="48.447468" min_fps="37.399837"

3.5GHz
1600x900: av_fps="60.792843" min_fps="49.109402"
1920x1080: av_fps="48.993683" min_fps="40.595753"
Edited by Feimitsu - 8/21/16 at 4:39am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD CPUs
  • Analysis of Kaveri's Downclocking CPU to 3.0GHz when iGPU is under full load
Overclock.net › Forums › AMD › AMD CPUs › Analysis of Kaveri's Downclocking CPU to 3.0GHz when iGPU is under full load