Observations with an FX-8350

[The Alias]

IdontCare: Excellent job! I enjoy my 8350. However, if power draw for performance is the prime issue, Intel is light years ahead of AMD. Wth the cost of extras (better cooling etc) to make an AMD 8350 competitive with a 2500k/3570k the price differential is so close.

Right now for a new build, it's very difficult to recommend even the 8350 over the Intel 2500k/3570k (forget comparisons to 3770k as it is more expensive). It may come down to the type of software you are using.

Is the Vishera an improvement over the Zambezi? Sure, but honestly that isn't a difficult challenge.

I own 2 2500k rigs Oc'd and have no experience with the 3570k. However, all indications are that it improves on the 2500k.

For new gamers building a rig it is easy. If you are debating between the 3570k and the 8350, Intel appears to be the best buy. If you do other things (video editing etc and occasionally a game) the choice is closer. Also, if you are the owner of an AM3+ mb and want to "broaden" your horizens the 8350 might be worthwile. Don't expect miracles.

IDC, thank you for your objective approach to testing. Incredibly thorough. If ANYONE can maximize the performance of the FX8350, my money is on you!

to me the 8350 was never a good option because of the 8320 which is the same chip just underclocked and the 6300 which provides the same gaming capability as the 8350 except cheaper . I'd never suggest anyone buy an 8350 tbh the ones below that are golden at their price points though

 

[Idontcare]

@IDC
Check out this link .

That is a very nice 8350 chip compared to mine, and a very nice job on isolating the power usage of the CPU. My 8350 at 4GHz reboots when running LinX if the voltage is below 1.32V. And that guy's 8350 can go all the way down to 1.18V stable!

I am surprised the VRM's can get away with being passively cooled considering how many watts they are dissipating per his numbers. I trust ASUS knows what they are doing though.

It sure doesn't help AMD's reputation though, taking a power hungry chip and then pairing it with a rather inefficient VRM makes for a platform-level power hog like no other :\

It is a pity ASUS didn't use the same VRMs they use on the Intel boards for the AMD boards.

 

[Idontcare]

to me the 8350 was never a good option because of the 8320 which is the same chip just underclocked and the 6300 which provides the same gaming capability as the 8350 except cheaper . I'd never suggest anyone buy an 8350 tbh the ones below that are golden at their price points though

OC'ing AMD chips busts their warranty, no performance tuning plan option like with Intel.

The FX-8320 costs $180 at the egg, I bought my FX-8350 for $200 at the egg. (it is now $210 but it is up and down every other week)

I am not really all that worried about $20, but your $20 gets you another 500MHz of validated clockspeed headroom and a warranty that need not be invalidated.

I looked at the prices, and the trade-offs, and came to the exact opposite conclusion. Considering the minimal price differential, I would be hard-pressed to recommend an 8320 over an 8350. You give up a lot just to save those $20.

 

[thilanliyan]

Hmmm..undervolting may be an option I guess if I go with a 8350/8320.

Those results from XS are CRAZY!! 1.1875v?!

 

[Gikaseixas]

IDC where's the oc attempt? Can't wait for your conclusion on that.

 

[AtenRa]

OK there must be a "problem" with LinX64 and Bulldozer/Vishera Turbo Core.

I have just disabled the Turbo Core in the bios in order to examine the behavior of the CPU under LinX64. With the Turbo Core enabled the CPU was working at 4100MHz with all 8 cores and output ~77GFlops, power consumption was at 279W.
When i disabled the Turbo Core, the CPU was working at the base frequency of 4.0GHz with 1.272V(Voltage on AUTO) BUT it produced 84GFlops.
Power consumption was at 247W.

Although I havent noticed any throttling activity when Turbo mode was enabled, it seams that the CPU was drawing more power than it should and the performance was sub-par. This behavior could be attributed to the Motherboards BIOS, or to the the application(LinX64) or to the actual CPU.

Any way i will continue to investigate and report any more findings.

 

[guskline]

OC'ing AMD chips busts their warranty, no performance tuning plan option like with Intel.

The FX-8320 costs $180 at the egg, I bought my FX-8350 for $200 at the egg. (it is now $210 but it is up and down every other week)

I am not really all that worried about $20, but your $20 gets you another 500MHz of validated clockspeed headroom and a warranty that need not be invalidated.

I looked at the prices, and the trade-offs, and came to the exact opposite conclusion. Considering the minimal price differential, I would be hard-pressed to recommend an 8320 over an 8350. You give up a lot just to save those $20.

Also from the reading I have done, the binning of the 8350 vs 8320 seems to more strict than with the 8150 vs 8120 chips. For the OCers, the consensus is that the 8350s have the most OCing headroom. I made the same decision when choosing the 8350 vs 8320. The price differential was so close that I opted for the higher clock.

 

[dkm777]

Anybody here fold? I'm very curious what kind of PPD is FX-8350 pulling and at what power consumption. Because you see I'm debating whether to drop one into my secondary rig which was built just for this reason - to play with FX chips. But if the power consumption is anywhere near what LinX pulls then I'll have to scratch that idea and keep my X4 980 at 4GHz/2.8GHz.

 

[Idontcare]

IDC where's the oc attempt? Can't wait for your conclusion on that.

I haven't touched OC'ing yet because I don't want to invalidate my warranty until I know for certain the chip is functioning correctly.

With the holidays coming, I probably won't be getting into much OC'ing until early Jan TBH.

OK there must be a "problem" with LinX64 and Bulldozer/Vishera Turbo Core.

I have just disabled the Turbo Core in the bios in order to examine the behavior of the CPU under LinX64. With the Turbo Core enabled the CPU was working at 4100MHz with all 8 cores and output ~77GFlops, power consumption was at 279W.
When i disabled the Turbo Core, the CPU was working at the base frequency of 4.0GHz with 1.272V(Voltage on AUTO) BUT it produced 84GFlops.
Power consumption was at 247W.

Although I havent noticed any throttling activity when Turbo mode was enabled, it seams that the CPU was drawing more power than it should and the performance was sub-par. This behavior could be attributed to the Motherboards BIOS, or to the the application(LinX64) or to the actual CPU.

Any way i will continue to investigate and report any more findings.

What is your stock VID under load? (CoreTemp will tell you) Mine is 1.3875V.

I too get the 77GFlops vs. 85GFlops variation from run to run, but in my case the power usage does not flip-flop like you observe. In my case the power usage for 77GFlops is ~280W whereas the power usage for 85GFlops is ~290W.

I disabled turbo-core in the BIOS but left everything else enabled (CnQ, APM, C6, etc). If you are getting 77 vs 85, same as me, then that would suggest to me that turbo-core isn't really boosting your clocks to 4.1GHz, that they really are at 4GHz otherwise your GFlops should be higher than mine.

(4.1GHz/4GHz)*85GFlops = 87GFlops

You would notice if your GFlops were breaking 87 versus coming in at 85.

So I think the turbo-core situation is a red-herring.

When I get the 77GFlops result I usually have to reboot my system before I can get the 85GFlops result. But once the system is "in the zone" and churning out 85GFlop results then it is really consistent about it.

For what its worth, I have been able to get GFlop values as low as 63GFlops at the same clockspeed (4GHz) just by playing around with my Digi+ settings in AI Suite. Change the CPU and CPU/NB phases, LLC, and response factors and you'll see the GFlops values move up and down.

I suspect the NB is changing clockspeed...but I have not found any programs that will tell me the NB clockspeed outside of the BIOS screen. If the NB is changing clockspeed then we would expect the performance to vary all while the core clockspeeds remain the same.

Do you know of any windows utilities that can track (or change) the NB clockspeed in real-time?

 

[Idontcare]

idle power usage is highly dependent on the rest of the PC, and the motherboard,
as for the cost, for the same $430 you could go with the K i5 and a good enough MB for high OC (at the end of the day the i5 OCed requires a lot less "quality" from the VRMs I think), now the K i5 will clearly beat the FX in some aspects, and be beaten in others, for the average gamer/overclocker I think the i5 just makes more sense at the moment...

That is a more an argument against the value of the motherboards than the value of an FX vs. i5/i7.

One does not buy a $230 CVF-Z or $350 MIVE-Z if one's goal is simply to have maximal fps/$.

That tier of mobos were created for an entirely different purpose. And all I was getting at was that for the purposes in which one finds themselves buying a CVF-Z or MIVE-Z, the aggregate expense of mobo+cpu compares favorably for the FX-8350 vs. i7-2600k/3770k.

51C HSF? While software reports the CPU to be 61C? Ouch.

Atleast certain chips looks to be way way beyond 125W. Unless mobo manufactors directly screwed up. But that should essentially demand a huge recall. Either CPU to MBs.

I believe the TJmax for the FX-8350 is 90°C, as reported by CoreTemp, but I have not been able to confirm it.

With the Intel chips it is easy enough to confirm TJmax, you just load up LinX and unplug the HSF fan's power cord and watch the CoreTemp (or Realtemp) temperatures versus clockspeed and you'll notice quite readily when the CPU is throttling from hitting TJmax.

However, in the case of my FX8350 all attempts to get the chip to throttle have been met with failure. What I mean by that is that as soon as the chip gets to 87°C, 3°C below the reported TJmax, the system shuts down (physically shuts off).

It doesn't throttle, it doesn't reboot, and it isn't a stability issue because I've underclocked the cpu and it does the same thing.

It also isn't a current-draw issue as I reduced the linx loading to just 2 threads, leaving the system drawing a mere 150W from the socket at 3GHz and the system still just shuts down as soon at it hits 87°C.

Is this normal behavior for AMD chips when they hit TJmax? They don't throttle, they just flat out shut down?

 

[AtenRa]

The NB dont change within Windows, it stays at the frequency you have set it.

You can only change the NB values in BIOS, you can see the NB frequency in CPUz (Memory). Default frequency is 2200MHz.

With the ASUS M5A97 R2.0 i cannot change the values in the LLC, i can only ENABLE the CPU LLC. With Turbo Core disable and CPU LLC enable and CPU Voltage at AUTO, CPUz was showing 1.272V under load in LinX64. With Turbo Core enable it was rising up to 1.4V.

I have also found out that i could run LinX with 1.2875V with Turbo Core enabled and CPU LLC BUT, linX was producing very low results. It was like the CPU was under performing but it was stable.
Also, i could run Cinebench 11.5 with 1.25V, Turbo Core ON, but linX would produce an error.

I dont know how LinX is working but it strains the CPU more than everything else. It reminds me OCCT with GTX480, it was the only application that GTX480 would use that much power because it would not throttle down.

 

[blckgrffn]

I think it begs the question, then, if LinX is a worthwhile test for AMD if it shows results that are so far removed from 99.9999% of tasks that folks actually use a CPU for.

I know how that sounds, but it's like some of those GPU power draw tests... I guess its nice to see how badly a chip might behave, but numbers from even other synthetics that more closely resemble real world scenarios are much more interesting.

 

[(sic)Klown12]

Is this normal behavior for AMD chips when they hit TJmax? They don't throttle, they just flat out shut down?

I believe that depends on the motherboard. Some will shut down before you can reach the CPU's throttle point while others will let it reach it and then do what you're used to. But all my experience comes with Thuban and before so something might have changed since then.

 

[inf64]

I believe that depends on the motherboard. Some will shut down before you can reach the CPU's throttle point while others will let it reach it and then do what you're used to. But all my experience comes with Thuban and before so something might have changed since then.

I think you are correct. We see AtenRa's board/CPU combo just throttles down but I'm not sure what kind of temperatures is he seeing in that scenario.

 

[Idontcare]

The NB dont change within Windows, it stays at the frequency you have set it.

You can only change the NB values in BIOS, you can see the NB frequency in CPUz (Memory). Default frequency is 2200MHz.

With the ASUS M5A97 R2.0 i cannot change the values in the LLC, i can only ENABLE the CPU LLC. With Turbo Core disable and CPU LLC enable and CPU Voltage at AUTO, CPUz was showing 1.272V under load in LinX64. With Turbo Core enable it was rising up to 1.4V.

I have also found out that i could run LinX with 1.2875V with Turbo Core enabled and CPU LLC BUT, linX was producing very low results. It was like the CPU was under performing but it was stable.
Also, i could run Cinebench 11.5 with 1.25V, Turbo Core ON, but linX would produce an error.

I dont know how LinX is working but it strains the CPU more than everything else. It reminds me OCCT with GTX480, it was the only application that GTX480 would use that much power because it would not throttle down.

CoreTemp will tell you what the VID is, VID is not the CPU's voltage, it is the target value towards which LLC is attempting to drive the CPU's voltage. Knowing the VID of your 8350 will tell you how much Vdroop your mobo is allowing under the various LLC settings.

LinX will produce lower results on Intel rigs of the CPU is undervolted too much. No one knows why, exactly, but the prevailing theory is that the CPU's built-in ECC circuitry is catching the errors and rerunning calculations intermittantly, keeping the results correct but resulting in fewer "good" computations per second and thus a lower GFlops.

Not sure if AMDs cpus have a similar ECC circuitry that allows for the chip to run in the gray-area of being unstable but not crashing or producing invalid results.

I think it begs the question, then, if LinX is a worthwhile test for AMD if it shows results that are so far removed from 99.9999% of tasks that folks actually use a CPU for.

I know how that sounds, but it's like some of those GPU power draw tests... I guess its nice to see how badly a chip might behave, but numbers from even other synthetics that more closely resemble real world scenarios are much more interesting.

The purpose of LinX is to give us enthusiasts one tool which can detect errors in the computed results of our chips when we use them in non-stock (non-validated) configurations. Be it under-volted or over-clocked, we are blind to whether or not our chips are actually computing 1+1=2 unless we run a progam like LinX which checks the mathematical results rigorously under very stressed conditions.

The other thing is that LinX is not a unique program. It is a proxy for any program that runs matrix algebra. DC for example, Gaussian98 is another. When I run Gaussian (a program I like to use) my FX8350 hits the same power-consumption at the wall (280W) as it does when running LinX.

The difference is that with LinX I have a built-in error checker that readily tells me if the chip is doing the math correctly, gaussian98 doesn't do that for me.

I believe that depends on the motherboard. Some will shut down before you can reach the CPU's throttle point while others will let it reach it and then do what you're used to. But all my experience comes with Thuban and before so something might have changed since then.

I wouldn't expect ASUS to make their flagship Intel ROG mobo (MIVE-Z) function differently than their flagship AMD ROG mobo (CVF-Z). If it were different mobo makers, or even different brands within the same maker then I could see it.

Makes me wonder if there is some BIOS setting that I need to enable/disable to change the throttling behavior.

A computer that just shuts down at TJmax (loses all your work) is not very satisfying in comparison to a computer that actually throttles itself but remains functional and does not crash your OS.

I think you are correct. We see AtenRa's board/CPU combo just throttles down but I'm not sure what kind of temperatures is he seeing in that scenario.

AtenRa's mobo is not thermal throttling, at least not in any of the screenshots and tests he posted about so far. His, and mine, are experiencing some other kind of throttle - perhaps a current draw throttle - at temperatures well below TJmax.

 

[jvroig]

I have just disabled the Turbo Core in the bios in order to examine the behavior of the CPU under LinX64. With the Turbo Core enabled the CPU was working at 4100MHz with all 8 cores and output ~77GFlops, power consumption was at 279W. When i disabled the Turbo Core, the CPU was working at the base frequency of 4.0GHz with 1.272V(Voltage on AUTO) BUT it produced 84GFlops. Power consumption was at 247W.

Ah, TurboCore. I do not own an FX processor, but I have a Thuban. I have noticed the same phenomenon (TurboCore / CnQ makes the cpu draw more power under IBT Max):

Everything at Stock: Power consumption is 291-297W
Everything at Stock, except CnQ & Turbo Off: Power consumption is 243-247W

Just wanted to chime in to let you know you are not an isolated case, and that it isn't even isolated to the FX chips.

I have more observations and rants against CnQ/TurboCore, but that's for a different thread.

 

[jvroig]

Is this normal behavior for AMD chips when they hit TJmax? They don't throttle, they just flat out shut down?

This will be completely anecdotal until I actually run controlled tests specifically to catch this behavior, but my instinct / experience tells me: No throttling, just shuts down

Throughout the course of owning a 965 BE and 1090T (both still alive and kicking until now, and still tinkering with them regularly), I have never ever seen them throttle at all. It just shuts down and/or reboots. When it is unstable, the reboot is preceded by a BSOD (as expected), but other times, when the temps (software monitored) reach above 62 (such as 64-66C, the highest I've seen from CoreTemp during some tests), the PC just blacks out, no BSOD.

Completely anecdotal and relies on "what I remember is...", so take it for what it's worth (not much, or at all). But I am more or less convinced this is right, because throughout various motherboards, coolers, voltages, and clocks, I have never ever encountered even one scenario where I saw either of the chips throttle.

 

[thilanliyan]

My CPU will start throttling once it hits 65C (I have that number set in BIOS somewhere)...a combination of settings got me to throttling rather than just shutting off, which did happen before with different BIOS settings. I can see it lowering the clock speed and voltage in ASUS Aisuite.

IDC, are you sure that TJMax is 90C for these chips? I always thought for AMD chips it was around 70C? And also that link I provided earlier in the thread says 70C, although I don't know if that is TJMax.

 

[Idontcare]

My CPU will start throttling once it hits 65C (I have that number set in BIOS somewhere)...a combination of settings got me to throttling rather than just shutting off, which did happen before with different BIOS settings. I can see it lowering the clock speed and voltage in ASUS Aisuite.

IDC, are you sure that TJMax is 90C for these chips? I always thought for AMD chips it was around 70C? And also that link I provided earlier in the thread says 70C, although I don't know if that is TJMax.

Coretemp claims it is 90C, that is the only reason I keep looking to 90C.

The link you gave, thanks btw for that (I did study it, numerous times in fact), is blank for the 8350 and 8320 which would imply it is not the same as bulldozer...but only AMD knows what it is and apparently they aim to keep it secret

At any rate I can attest to the fact that the processor runs along just fine without throttling all the way up to 87C, and then it just shuts down.

 

[0___________0]

Coretemp claims it is 90C, that is the only reason I keep looking to 90C.

The link you gave, thanks btw for that (I did study it, numerous times in fact), is blank for the 8350 and 8320 which would imply it is not the same as bulldozer...but only AMD knows what it is and apparently they aim to keep it secret

At any rate I can attest to the fact that the processor runs along just fine without throttling all the way up to 87C, and then it just shuts down.

I don't think that AMD CPU's have internal thermal sensors, at least they didn't in the past; you're seeing the motherboard's sensor at 87 C, but the sensor is slightly cooler than the CPU because it's underneath it. So your chip is probably hitting 90 C.

That actually seems really high to me, I can't ever recall people running BD FX chips near those temps.

 

[thilanliyan]

Coretemp claims it is 90C, that is the only reason I keep looking to 90C.

The link you gave, thanks btw for that (I did study it, numerous times in fact), is blank for the 8350 and 8320 which would imply it is not the same as bulldozer...but only AMD knows what it is and apparently they aim to keep it secret

Yes, I put it up anyway even though they were blank because they did not switch to a different process node or anything like that for Vishera did they? I assumed that temp recommendations would be similar (maybe not exact) between the Bulldozer and Vishera...

EDIT: This is something I just found at THG:
http://www.tomshardware.com/reviews/fx-8350-vishera-review,3328-2.html
"The FX-8350 wanted to go even faster, but the key here is a voltage setting low enough that you avoid hitting 70 degrees Celsius. At that point, the thermal monitor starts cycling cores to throttle down (evidenced in the image above), keeping the chip from getting any hotter and negatively impacting performance."

THG is experiencing throttling at 70C also (this value can be set in BIOS I believe...and at default it was 70C for me, I put it down to 65C). IDC, are you using an updated version of Coretemp?

 

[Martimus]

Is this normal behavior for AMD chips when they hit TJmax? They don't throttle, they just flat out shut down?

This has been my experience, but I haven't overclocked an AMD chip since I had a S939 4200+ X2. That chip would shut down when it got too hot.

 

[Pilum]

Is this normal behavior for AMD chips when they hit TJmax? They don't throttle, they just flat out shut down?

It shouldn't be, if you check the Family 15h BKDG, chapter 2.11. If properly setup, the CPU should trigger PROCHOT and throttle. If it hits THERMTRIP, it'll shut down. I don't know if this is really the responsibility of the BIOS or if the OS will re-program the appropriate MSRs.

You might test the behavior under Linux. If it properly throttles, it's a Windows bug. You could also check the values of the MSR in Linux or use HWDIRECT in Windows. But you'll need a real Linux install, this won't work in a VM.

Hm. Just checked the Intel Docs. They state explicitly that it is a requirement for the BIOS to enable either TM1 or TM2 (Thermal Monitor), and that OS and applications should leave the settings alone. The AMD BKDG doesn't state this requirement. Have you contacted your motherboards manufacturers tech support for clarification? That's really their job to know about this, especially considering it's a high-end board.

It's really sad that AMD systems still suffer from such stupid problems. Oh, when will they ever learn...

BTW, Intel CPUs shut down at 20°C above Tjmax if throttling fails to keep temperatures in check. That's 125°C for IVB.

 

[Idontcare]

Yes, I put it up anyway even though they were blank because they did not switch to a different process node or anything like that for Vishera did they? I assumed that temp recommendations would be similar (maybe not exact) between the Bulldozer and Vishera...

EDIT: This is something I just found at THG:
http://www.tomshardware.com/reviews/fx-8350-vishera-review,3328-2.html
"The FX-8350 wanted to go even faster, but the key here is a voltage setting low enough that you avoid hitting 70 degrees Celsius. At that point, the thermal monitor starts cycling cores to throttle down (evidenced in the image above), keeping the chip from getting any hotter and negatively impacting performance."

THG is experiencing throttling at 70C also (this value can be set in BIOS I believe...and at default it was 70C for me, I put it down to 65C). IDC, are you using an updated version of Coretemp?

Interesting. So I wonder where in the BIOS I need to enable the throttling. I certainly don't want to be needlessly damaging my chip.

I suppose I better ask ASUS.

Hey, another noob question here, is it common to not be able to run ram at T1 command rate with AMD rigs?

I've got this 4x4GB GSkill ram, the exact same sticks I ran at T1 rate on my Intel Asus mobo, and I can run the ram at the same speed, latency and voltage on the AMD mobo but I cannot enable T1 without the mobo locking up on reboot.

Are AMD platforms not compatible with T1 command rate, or are they more sensitive to the timings and so I will need to bump up the voltage or some such?

 

[Idontcare]

Anybody here fold? I'm very curious what kind of PPD is FX-8350 pulling and at what power consumption. Because you see I'm debating whether to drop one into my secondary rig which was built just for this reason - to play with FX chips. But if the power consumption is anywhere near what LinX pulls then I'll have to scratch that idea and keep my X4 980 at 4GHz/2.8GHz.

I don't fold but if there is some form of bench you'd like me to run just point me to it and I'll download and run the bench for you.

I'll have to talk to Anand about writing something up for the main site looking at intraspecific variability in popular components.

Sounds like an excellent idea, not many buyers get a chance to find out about those tradeoffs before they make their purchase decision.