FX-8310: Initial Impressions

[janeuner]

== Background ==
This part is intended as a modest CPU upgrade to an existing system with the following components:
• FX-6300
• MSI 970A-G43
• 16GB DDR3-1833

The main use of the system is software development and video encoding. By moving from the FX-6300 to the FX-8310, I believed that I would see a small improvement in single thread performance (4.1Ghz vs 4.3Ghz Turbo) and a ~30% improvement in video encoding speed.

The FX-8310 was attractive due to its low upgrade cost (~$115 up front, minus $25 MIR, and sell the FX-6300 for ~70 on ebay) and the 95W TDP. My particular motherboard is known to have issues with high TDP parts (it has no VRM heatsink), so I was also very interested in net power consumption.


== Power Comparison ==
Power consumption was measured using a Kill-a-Watt meter to compare total power consumption, before and after the upgrade. Idle power was determined from Windows 7 Desktop with no applications. Peak CPU power was calculated with the Prime95 small-FFT torture test, and turbo core was monitored with CPU-Z.

FX-6300 Idle: 128W
FX-6300 Peak: 216W

FX-8310 Idle: 133W
FX-8310 Peak: 228W


== Handbrake Encoding ==
CPU performance was calculated using Handbrake to encode a 75-second video sample. Measurements were taken using a single thread and using all available cores, the best of three runs was recorded.

Single thread - FX-6300: 10.11 fps
Single thread - FX-8310: 9.90 fps

All threads - FX-6300: 45.84 fps
All threads - FX-8310: 57.73 fps

 

[janeuner]

(Post 1, cont'd)
== Further Investigation ==
The numbers above did not meet my expectations, so I fired up CPU-Z and watched the AMD's Turbo Core tech while it boosted to turbo speeds. I did not take quantitative measurements, but several trends quickly emerged.

1. With just 1 thread scheduled, FX-8310 spent less than half the time at 4.3Ghz. Most of the CPU time was spent at 3.7 Ghz, which is the FX-8310's secondary boost state. This is in contrast to FX-6300, which spent much more time boosted to 4.1 Ghz.

2. With all 8 threads scheduled, FX-8310 would occasionally throttle to 2.9Ghz, which is supposedly a power save state. Strangely, it would do so with Cool 'n Quiet disabled. Even more strange, it does not throttle when Turbo Core is disabled.

3. Simply by turning off turbo core, I can improve the FX-8310's multi-core encoding benchmark to 60.7 fps.

4. I briefly tried overclocking to each of the boost speeds and voltages. System power was around 280 watts at 3.8 Ghz. The system booted at 4.3 ghz, but I dared not do a 8-thread load test with this motherboard.

== Summary ==
This experiment did not go as expected. Although I will probably keep the FX-8310, I would not say that this exercise was worth the meager dollars invested. From a little playing, it is clear this chip is fantastic for overclocking, but the Turbo Core configuration leaves much to be desired at 95W.

TL;DR - FX-8310 performs like a 3.2Ghz chip that will turbo to 4.0 Ghz. It is very likely to overclock to 4.3Ghz or beyond. YMMV.

 

[SlowSpyder]

I'm not all that happy with how my CPU uses turbo either. Using a load that stresses just a single core, I'll see it bounce between 4.4Ghz - 4.7GHz, but it doesn't seem to ever lock into the turbo speed as I thought it would. The good news for you is that it sounds like the jobs you run will be able to take advantage of the extra cores, so you should see some benefit. Not bad for very little money or effort.

 

[Kalessian]

Nice thread, I may upgrade one of my 6300 to this eventually (for entertainment more than performance)

 

[exar333]

Nice thread, I may upgrade one of my 6300 to this eventually (for entertainment more than performance)

Agree.

This seems like a cost-friendly option for a drop-in replacement for someone using a 4xxx or 6xxx CPU today.

 

[Jovec]

Even for AMD, that idle is quite high. I can get my 8350 down to 62w at the wall with 8350, 16GB, 2 SSDs, 7950, 990FX, 7 fans, and a gold PSU. In fact, tuning for a low voltage had little effect on idle power consumption, as pure stock settings were maybe 2-3w higher.

 

[SlowSpyder]

Even for AMD, that idle is quite high. I can get my 8350 down to 62w at the wall with 8350, 16GB, 2 SSDs, 7950, 990FX, 7 fans, and a gold PSU. In fact, tuning for a low voltage had little effect on idle power consumption, as pure stock settings were maybe 2-3w higher.

Yea, it does look a little higher than I'd expect. With a Swiftech 317GPH pump, 4 x 200mm fans on my radiator, other case fans, 2 x SSD's, 7970, etc. I idle at about 125-130 watts at completely stock factory settings with a 1300 watt power supply, according to my Kill-o-Watt.

 

[soccerballtux]

(Post 1, cont'd)
== Further Investigation ==
The numbers above did not meet my expectations, so I fired up CPU-Z and watched the AMD's Turbo Core tech while it boosted to turbo speeds. I did not take quantitative measurements, but several trends quickly emerged.

1. With just 1 thread scheduled, FX-8310 spent less than half the time at 4.3Ghz. Most of the CPU time was spent at 3.7 Ghz, which is the FX-8310's secondary boost state. This is in contrast to FX-6300, which spent much more time boosted to 4.1 Ghz.

2. With all 8 threads scheduled, FX-8310 would occasionally throttle to 2.9Ghz, which is supposedly a power save state. Strangely, it would do so with Cool 'n Quiet disabled. Even more strange, it does not throttle when Turbo Core is disabled.

3. Simply by turning off turbo core, I can improve the FX-8310's multi-core encoding benchmark to 60.7 fps.

4. I briefly tried overclocking to each of the boost speeds and voltages. System power was around 280 watts at 3.8 Ghz. The system booted at 4.3 ghz, but I dared not do a 8-thread load test with this motherboard.

== Summary ==
This experiment did not go as expected. Although I will probably keep the FX-8310, I would not say that this exercise was worth the meager dollars invested. From a little playing, it is clear this chip is fantastic for overclocking, but the Turbo Core configuration leaves much to be desired at 95W.

TL;DR - FX-8310 performs like a 3.2Ghz chip that will turbo to 4.0 Ghz. It is very likely to overclock to 4.3Ghz or beyond. YMMV.

thanks for writing about #1, I have the same problem but with even worse behavior-- about 90-95% of the time it spends itself at 3.7ghz. I have an outstanding ticket that I had sent directly to Asus engineers about this. They probably have a backlog, or will check it out when they get around to another update. I ping them regularly and they said they'll contact me when they have something. I'm skeptical, but I do believe they will. Their support is believably competent.

I made a thread about this here, see post #38 for resolution. Basically, I use AMDMsrTweaker to write the 4.3ghz speed to the MSRs responsible for the 3.7ghz settings. I believe the TDP code for the 4.3ghz is borked resulting in the poor turbo behavior. When both the 3.7ghz and 4.3ghz slot are written to turbo to 4.3ghz, on 1-2 thread workloads (what I want the 4.3ghz for most) I spend pretty much all time at 4.3ghz. On 2 heavy threads (Prime is chattier than Fritz Chess Bench 13 so use that for loading 2 and only 2 cores) it's at 4.3ghz most of the time. On 3 it starts TDP throttling a bit more.

The downside to this I don't get to use 3.7ghz for 3-4 threaded workloads, which is why I filed the Asus ticket, but I'm holding out faith that they'll come through. If they don't, I'll warranty with 1 month on the warranty.

#2 happens for me as well, very very rarely, like maybe 5% duty cycle. Enough that I don't care. Are you sure you're disabling Turbo Core? Disabling APM will prevent it from happening as it locked it into 3.4ghz for me. I think. Logically, they decided APM is like the parent of CnQ and Turbo. Which we didn't really need, it just made things confusing again.

#3 that's unfortunate, I guess you're spending more than the 5% duty cycle I am at 2.9ghz.

Honestly, I want them to open source the TDP throttling module because even in my limited knowledge I know that, at the architecture level, I could come up with something [rock solid that would make everybody happy] that wouldn't fall back to 2.9, and would consistently stay at 4.3ghz for 1 and 2 thread workloads.

This lack of refinement is what tells me that internally AMD just cant deliver. They have valuable IP and products and the hardware itself is great, but they lack a management that really ties all the details together. Intel, by contrast, does, which is why their first SSD and controller to market was the absolutely most reliable, best SSD controller anyone has ever released. I am an underdog fanboy, but sometimes underdogs never learn.

 

[janeuner]

Even for AMD, that idle is quite high.

That has nothing to do with the AMD processor.

Before I started testing, I locked all three cooling fans to 100% to reduce transients during testing. Switching them back to auto drops the idle back down to 125w. But the biggest idle power draw comes from the R9 280X graphics card, which is driving three monitors. If I drop to one monitor on the DVI link, idle power plummets to 85w. I guess the DP and/or HDMI drivers are unusually power hungry.

 

[janeuner]

#2 happens for me as well, very very rarely, like maybe 5% duty cycle. Enough that I don't care. Are you sure you're disabling Turbo Core? Disabling APM will prevent it from happening as it locked it into 3.4ghz for me. I think. Logically, they decided APM is like the parent of CnQ and Turbo. Which we didn't really need, it just made things confusing again.

#3 that's unfortunate, I guess you're spending more than the 5% duty cycle I am at 2.9ghz.

I am sure. I did several repetitions toggling Turbo core on and off, making sure Cool 'n Quiet was still enabled, because I was stunned at the behavior. Turbo Core was the only factor that caused the chip to throttle down to 2.9 Ghz.

I forgot to mention this in the writeup - with turbo disabled, the Prime95 peak power rose to 250w and the chip held steady at 3.4Ghz. If not for that, I would be convinced it was the motherboard that was throttling the CPU.

This lack of refinement is what tells me that internally AMD just cant deliver. ... I am an underdog fanboy, but sometimes underdogs never learn.

For what its worth, this is not a retail chip, so rough edges are to be expected. My experience with FX-6300 was much more in line with what was advertised. For all we know, it may be that FX-8310 was a failed TurboCore experiment, and that's why it is so cheap.

Thanks for the tip about the turbo core speeds. I'll give that a shot this weekend.

 

[Ramses]

The faster the FX, the more useless turbo is ime. I saw that going from an 8350 to a 9590 even. Have seen the idle dual screen power usage too, blamed it on AMD GPU.

 

[Erenhardt]

4. I briefly tried overclocking to each of the boost speeds and voltages. System power was around 280 watts at 3.8 Ghz. The system booted at 4.3 ghz, but I dared not do a 8-thread load test with this motherboard.

Pfff.... What a sissy. I'm just after stress test session with my MSI G760 P43 an FX6300 4.8Ghz@1.55V


Thanks for sharing! Shame these are not available in my country. Would love to improve BF4 performance for cheap.

 

[Jovec]

Turbo (And APM, if the bios exposes it) is how AMD keeps their CPUs within TDP spec.

 

[Jovec]

That has nothing to do with the AMD processor.

Before I started testing, I locked all three cooling fans to 100% to reduce transients during testing. Switching them back to auto drops the idle back down to 125w. But the biggest idle power draw comes from the R9 280X graphics card, which is driving three monitors. If I drop to one monitor on the DVI link, idle power plummets to 85w. I guess the DP and/or HDMI drivers are unusually power hungry.

Not the drivers, it's more a hardware limitation. The GPU clocks up the core and VRAM when more than one monitor is connected. For example, my 7950 idles at 300 core, 150 Mem with a single monitor, but jumps to 500 core, 1200 Mem with two connected, for about a 30w increase.

 

[janeuner]

I meant the line drivers, so Jovec and I are in agreement.

Per soccerballtux, I fiddled with the boost states in bios settings. I settled on 3.3/4.0/4.3 for stock/boost2/boost1 and left voltage config on auto. The cpu is still hanging out near 95w for 2-6 thread workloads with those settings, and the single-thread encode benchmark is up to 10.16 fps.

I'd prefer an across-the-board stock upgrade, but this will do.

 

[SlowSpyder]

I meant the line drivers, so Jovec and I are in agreement.

Per soccerballtux, I fiddled with the boost states in bios settings. I settled on 3.3/4.0/4.3 for stock/boost2/boost1 and left voltage config on auto. The cpu is still hanging out near 95w for 2-6 thread workloads with those settings, and the single-thread encode benchmark is up to 10.16 fps.

I'd prefer an across-the-board stock upgrade, but this will do.

$20 upgrade.

 

[soccerballtux]

I meant the line drivers, so Jovec and I are in agreement.

Per soccerballtux, I fiddled with the boost states in bios settings. I settled on 3.3/4.0/4.3 for stock/boost2/boost1 and left voltage config on auto. The cpu is still hanging out near 95w for 2-6 thread workloads with those settings, and the single-thread encode benchmark is up to 10.16 fps.

I'd prefer an across-the-board stock upgrade, but this will do.

fiddling isn't what I said, I said setting them both to 4.3ghz. The code that manages the TDP is borked, the only way to get 4.3 out of the chip is to set them both to 4.3

 

[lord_emperor]

I got this CPU over xmas as well. Mine's running at 4.0/4.8 on a GA-970A-DS3P with a CM212.

 

[DrMrLordX]

These are good chips even on junk boards if your target is no higher than ~4.4 ghz. And really they are good for more than that if you want to go there, but you'll need a good enough board to get you there.

 

[GOLI@TH]

Sorry for bringing this up, but i do really curious about it.

Is it true AMD FX series not really 6 or 8 cores chip?
It rather like 3 or 4 core with HT-ish feature.(?)

From what i red in various forums that the last real 6 core from AMD is 10xxT series.

 

[DrMrLordX]

Yes and no. AMD chips that rely on CMT share FPU (floating point unit) resources. So, for integer workloads, it's like the chip has 8 cores, but for floating point, maybe not so much.

Though my own experiences with Steamroller (the most modern CMT chip AMD has yet released) show that the shared FP resources are more powerful than you might think.

Sadly, FX doesn't use Steamroller. It uses Piledriver, or actually a quasi-Bulldozer that people call Piledriver. FX has some . . . interesting issues as a result.

 

[GOLI@TH]

Yes and no. AMD chips that rely on CMT share FPU (floating point unit) resources. So, for integer workloads, it's like the chip has 8 cores, but for floating point, maybe not so much.

Though my own experiences with Steamroller (the most modern CMT chip AMD has yet released) show that the shared FP resources are more powerful than you might think.

Sadly, FX doesn't use Steamroller. It uses Piledriver, or actually a quasi-Bulldozer that people call Piledriver. FX has some . . . interesting issues as a result.

Interesting, short explanation that needs advanced thinking to understand. LOL.
Thanks, just give me time to process.

 

[Abwx]

TL;DR - FX-8310 performs like a 3.2Ghz chip that will turbo to 4.0 Ghz.

Not exactly, this is true only if you load it with Prime 95, and this latter use 20-25% more power than any regular heavy multithreaded soft.

It should actualy perform like a 3.5-3.6GHz chip that turbo to 4.0, do a test with Fritzchess bench wich is integer heavy and load the CPU more than Cinebench, and you can select the number of threads to check the turbo behaviour.

http://www.hardware.fr/marc/cwiz/fritz.exe

 

[Abwx]

Yes and no. AMD chips that rely on CMT share FPU (floating point unit) resources. So, for integer workloads, it's like the chip has 8 cores, but for floating point, maybe not so much.

Though my own experiences with Steamroller (the most modern CMT chip AMD has yet released) show that the shared FP resources are more powerful than you might think.

Sadly, FX doesn't use Steamroller. It uses Piledriver, or actually a quasi-Bulldozer that people call Piledriver. FX has some . . . interesting issues as a result.

It has a shared FPU for operations executions but the FP threads are handled by the Integer cores ALUs wich manage the load/store unit for said threads, there s a confusion done between shared execution ressources and threads managements, as such a module has two cores that manages each both integer and FP related threads.