Observations with an FX-8350

[AtenRa]

Sorry if this is off-topic, but I have a question: did you need to update the BIOS of your motherboard for it to recognize your CPU?

I ask because we got the ASUS R2.0 boards at work and I'm kinda hesitant about recommending them to people doing new builds with Vishera because there's a chance they won't POST.

Just installed the CPU and worked. No BIOS update needed.

 

[inf64]

That's pretty good analysis IDC,it looks like VID variation is what is producing all those different power draw numbers in various reviews(a lot of reviewers use p95 these days,not as taxing as linpack but close).

If you are lucky enough than this CPU should be a great OCer,albeit with huge power draw. I look forward to your other results,good work.

 

[guskline]

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!

 

[Idontcare]

That's pretty good analysis IDC,it looks like VID variation is what is producing all those different power draw numbers in various reviews(a lot of reviewers use p95 these days,not as taxing as linpack but close).

If you are lucky enough than this CPU should be a great OCer,albeit with huge power draw. I look forward to your other results,good work.

Two things struck me when I was delving into this - the first is exactly as you say, spot on, the natural VID variation across samples is likely playing a large role in the variation of power-consumption numbers across the various reviews, and second was that even without the VID variation, the variation in "stock" LLC settings from mobo to mobo (and even from bios rev to bios rev for a given mobo make & model) is driving a Vdroop variability from review to review.

For example when I reduced my LLC setting from "High" to "Med", that alone reduced the Vcore at load by some 0.04V and power by nearly 30W...all because Vdroop increased.

Less expensive mobos are probably going to have higher Vdroop, it costs money to install all the components needed to stabilize voltages. My crosshair V formula-Z cost me $230, that is $30 more expensive than the FX-8350 I bought to go in it! But consider what I'd get if I had bought an $80 AM3+ combo board? I bet Vdroop is rather large with that tier of mobos, and power-consumption will be quite lower as Vdroop is effectively a way to unwittingly undervolt your chip...and undervolting will always lower power-consumption and operating temperatures.

And as we saw at the beginning of this thread, if people are solely relying on software reporting their Vcore after Vdroop is factored in then (versus physically measuring it) then they are probably blind to just how big (or small) their actual Vdroop is.

My power consumption climbed because my motherboard was doing a good job eliminating the Vdroop and driving the loading voltage to the spec'ed VID. If I had a less expensive board then I probably would have never seen the power consumption get as high as it did at "stock" settings.

And that might explain why I can considerably undervolt my FX-8350 from its stock VID. Maybe AMD spec'ed the stock VID so high because they are trying to build in overhead in terms of Vcore such that their chips will still be stable once the reality of Vdroop on cheaper AM3+ mobos is factored in...all while knowing the lower Vcore will result in the FX-8350 coming in under the TDP spec.

And it only becomes a problem when an enthusiast plugs the chip into a mobo that has expensive LLC components that can eliminate Vdroop at stock, driving Vcore to VID and pushing power usage well above TDP. For those few individuals who do that, AMD is probably thinking those individuals don't care about power usage at stock settings anyways. A "no harm no foul" scenario.

Because lets be real, I didn't buy a $230 mobo for my $200 cpu just to run that cpu at stock Who would?

 

[Ferzerp]

This shows us nothing we didn't already know, but it does put numbers to it for this specific LinX app. Shows the CMT-tax of loading 2 threads per module versus 1 thread per module for the first four threads with LinX.

Actually, that graph is extremely telling in regards to the claims that a module is a full 2 cores. I don't believe I've ever seen that particular set of data before.

 

[Idontcare]

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!

Thanks for the kind words guskline, and I agree with your assessment of the hardware choices and paths :thumbsup:

I like my FX-8350. It idles about 10W lower than my 2600k or 3770k. And at a platform level CoO (cost of ownership) my FX-8350 + mobo set me back a mere $430, whereas the Intel CPU + mobo set me back nearly $700.

Now obviously my choice in mobos drives that price gap more than anything, but that is what these unlocked CPUs are made for - people who like to OC and tinker.

Yes performance is higher with the Intel, and power usage is lower, but the price differential is substantial as well and the performance/dollar favors the FX-8350 IMO.

From my perspective it is all a wash in the end, financial wise. Some days I like eating a burger at Red Robin, other days I fancy myself some Fuddruckers. In neither case do I hunger for a whopper or big mac.

I put the FX-8350 and the 2600k/3770k in the same league - Red Robin and Fuddruckers - not different leagues like Red Robbin vs McDonalds.

And naturally that would place the LGA2011 extreme cpus in the "pure kobe beef" class of burgers, a league of their own in price and quality.

Just installed the CPU and worked. No BIOS update needed.

FWIW the same was true of my Crosshair V formula-z mobo. It came with a surprisingly outdated bios too, Rev 0704. It still booted just fine with the FX8350. And then I updated to the recent bios rev 1101 (some 4 bios revs above the shipping bios).

Actually, that graph is extremely telling in regards to the claims that a module is a full 2 cores. I don't believe I've ever seen that particular set of data before.

I know AtenRa has done some CMT efficiency tests that are captured in his blog, but yeah I don't know of too many threads around here (or pro reviews for that matter) that delve into it.

The "CMT tax" is something I am keenly interested in exploring. I've got more data to show on this, pulling it together right now. The cores are real, but the IPC suffers tremendously when the module is fully loaded. I can see why AMD chose to focus on this weakness of the microarchitecture with the steamroller update, just too bad the update had to wait until steamroller before it gets implemented.

 

[Gigantopithecus]

Two things struck me when I was delving into this - the first is exactly as you say, spot on, the natural VID variation across samples is likely playing a large role in the variation of power-consumption numbers across the various reviews, and second was that even without the VID variation, the variation in "stock" LLC settings from mobo to mobo (and even from bios rev to bios rev for a given mobo make & model) is driving a Vdroop variability from review to review.

Since my 'day job' is studying how variation between humans can inform our understanding of human evolution, I've wondered for years about the parameters of variation in computer components - and how this variation manifests itself in real world usage.

It's crystal clear from this thread (which is a great, informative read) that variation between multiple parts can have enormous and very real consequences for power usage. Over the lifetime of systems that are being used to do work, the discrepancies we've seen in this thread amount to meaningful real world cost differences.

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

 

[SPBHM]

I like my FX-8350. It idles about 10W lower than my 2600k or 3770k. And at a platform level CoO (cost of ownership) my FX-8350 + mobo set me back a mere $430, whereas the Intel CPU + mobo set me back nearly $700.

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...

another test showing the voltage effect

if you look at how high the 8320 is even with significantly lower clock, it's clear that there is a lot of variation on these CPUs, and undervolt works quite well,

 

[Vesku]

This reinforces my opinion that if I ever upgrade from 1090T to 8350 I'm not going to set it for maximum safe voltage.

 

[Puppies04]

Ok to sum up IDC's data: Load in Linx results in 281W. Idle is 87W. He has some peripherals attached so they do contribute a bit to both numbers(board contributes the most and ram since they are stressed somewhat during the test). PSU eff. is 89% according to OP. So we subtract 30W(11%) from 281W and we end up with 251W. Subtract 87W-11%(77W) in idle : 251-77=174W. How much can be the power drawn by motherboard,ram and other stuff connected? According to above,in order for CPU to be in its 125W rating, the other stuff(board,ram,drives,fans,graphics card) have to draw ~50W which is kinda high IMO but possible. I know each stick of memory draws ~4W approx. If OP is running 2 stick then ram constitutes 8W or power draw(for 4 stick it's 16W, approx.) Graphics card can go into low 15W when power saving is engaged. That leaves ~26W for motherboard and drives. I have found out via quick google search that motherboard usually draws from 25 to 40W(I suppose high end boards draw more). So it's possible that IDC's CPU actually draws 125W or a bit more(130-135W) in this one benchmark such as FMA optimized linpack while the total system power at killawatt is 281W.

Your maths assumes that an 8350 at idle uses 0w, Or am I misquoting you?

 

[ShintaiDK]

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.

 

[Idontcare]

Actually, that graph is extremely telling in regards to the claims that a module is a full 2 cores. I don't believe I've ever seen that particular set of data before.

Here are the Cinebench results:

Cinebench scales well with Piledriver, but the CMT tax is right near the expected 20%.

Comparing the M1/M2 test versus the M1 C0 test, the thread scaling is a nearly perfect 2:1. M1/M2 = 1 thread per module

Compare that to the M1-M4 test with four threads, thread scaling is again a nearly perfect 2:1. M1-M4 = 1 thread per module

Likewise if we look at the C0/C1 test versus the M1 C0 test, the CMT tax is ~18%. Pretty close the 0.8x scaling that AMD mentioned would be the trade-off with their specific implementation of CMT on bulldozer. C0/C1 = 2 threads per module

The CMT tax is consistent, extending the calculation from 2 threads to 4 threads the C0-C3 test shows the same 0.82x scaling of that produced by the M1-M4 test. C0-C3 = 2 threads per module

It is also interesting to note the impact of the CMT tax on operating temperature and power consumption. Operating temperatures go up when ganging the threads together onto modules despite the power overall consumption decreasing.

This is expected if we consider thermal density of the sources of that power dissipation. When threads are on the same module that increases the thermal density within the module despite the overall power consumption decreasing (commensurate with performance decreasing).

When we have 1 thread per module the under-utilized portions of the module are acting as a form of dark-silicon, distributing the heat such that the overall peak operating temperature is reduced despite the higher overall power consumption (and higher performance).

Again, nothing we didn't already expect to see happening with a CMT processor but this puts specific numbers to the example for us to speak to.

 

[Phynaz]

The next time my wife calls me a Geek I'm going to have her read your threads.

:awe:

 

[guskline]

The next time my wife calls me a Geek I'm going to have her read your threads.

:awe:

Funny! Actually, I was going to nominate Idontcare to take over AMD's R&D. Maybe it would thrive!

 

[SlowSpyder]

Funny! Actually, I was going to nominate Idontcare to take over AMD's R&D. Maybe it would thrive!

Having more smart people might help with R&D, but I feel bad for the marketing team today.

The new AMD 8350 200 watt TDP processor...

Their slogan could be: AMD... Hey, it could be worse.

 

[Idontcare]

Having more smart people might help with R&D, but I feel bad for the marketing team today.

The new AMD 8350 200 watt TDP processor...

Their slogan could be: AMD... Hey, it could be worse.

LOL, they need more R&D money, not more people per se at this point in time.

They could do the ASML thing and petition would-be future customers to take an equity stake in AMD for the purpose of bolstering the R&D of said future products.

I can see it now - AMD takes to kickstarter to find funding to finish off steamroller development

The 200W TDP isn't really AMD's fault here, it is Intel's fault for creating the Intel Burn Test. If it weren't for IBT, and LinX by extension, then my FX-8350 would probably not exceed 125W power usage.

 

[ShintaiDK]

LOL, they need more R&D money, not more people per se at this point in time.

They could do the ASML thing and petition would-be future customers to take an equity stake in AMD for the purpose of bolstering the R&D of said future products.

I can see it now - AMD takes to kickstarter to find funding to finish off steamroller development

The 200W TDP isn't really AMD's fault here, it is Intel's fault for creating the Intel Burn Test. If it weren't for IBT, and LinX by extension, then my FX-8350 would probably not exceed 125W power usage.

Oh the irony!

http://shintai.ambition.cz/amd.pdf

 

[mrmt]

LOL, they need more R&D money, not more people per se at this point in time.

How can they field a bleeding edge product when they are tied to the third best foundry out there?

 

[inf64]

Your maths assumes that an 8350 at idle uses 0w, Or am I misquoting you?

Close to that,to be exact 3.6W :
http://www.hardware.fr/articles/880-4/consommation-efficacite-energetique.html (Click ATX12V link at the bottom of the chart)

@ IDC

That are some good point wrt cheaper boards . It's logical that cheaper designs would yield higher vdroop and lower power draw (which in that case is a welcome thing due to weaker VRMs etc.). Higher end boards on the other hand are meant for OCers(primarily) and they couldn't care less about power draw since they OC anyway. In their case vdroop needs to be as little as possible and that's why high end boards like yours perform like that.

 

[Ferzerp]

My setup is a piggy....

114W Idle
224W peak linpack
720W peak linpack + furmark
630W furmark alone

 

[Greenlepricon]

My setup is a piggy....

114W Idle
224W peak linpack
720W peak linpack + furmark
630W furmark alone

Holy cow! You could turn off your heater this winter and just run this! Well let everyone know how your system does performance wise. That's a lot of power if you aren't getting something amazing out of it.

 

[Ferzerp]

It's actually not bad. 114W idle covers the motherboard, 16GB of ram, 2 WD black drives spinning, 6 case or radiator fans, 2 sealed liquid cooling pumps at their idle speeds, 1 SSD, idle proc, 2 idle gpus.

224W (which only happened for a second, and I'm not sure it wasn't a GPU clock up) covers linpack so 110 for an overvolted, OC'ed IB. It sat at 190-191W for all the other active time in linpack which would mean that there is only a 75-76W difference in idle and full load which I don't really think could happen, but maybe (1.224V), but the 220 value was only a tiny amount of time.

Add another 500W for two reasonably OC'ed 680's running furmark. Though they're power limited at most OCs in furmark.

So yeah, that's about what one should expect for that. These days, it's only on 4 or 5 hours a week. Obviously, typical game load is much lower.

My HTPC/Media center server idles at 39-41W, increases to 47.5-49W for serving 1 cable channel to a remote xbox. Add in displaying the same tuner signal on its local screen and you're between 49-53W, kick in a second tuner and you're up to 57.5W peak

 

[Maximilian]

Having more smart people might help with R&D, but I feel bad for the marketing team today.

The new AMD 8350 200 watt TDP processor...

Their slogan could be: AMD... Hey, it could be worse.

I loled

AMD... At least its cheap!

 

[Greenlepricon]

It's actually not bad. 114W idle covers the motherboard, 16GB of ram, 2 WD black drives spinning, 6 case or radiator fans, 2 sealed liquid cooling pumps at their idle speeds, 1 SSD, idle proc, 2 idle gpus.

224W (which only happened for a second, and I'm not sure it wasn't a GPU clock up) covers linpack so 110 for an overvolted, OC'ed IB. It sat at 190-191W for all the other active time in linpack which would mean that there is only a 75-76W difference in idle and full load which I don't really think could happen, but maybe (1.224V), but the 220 value was only a tiny amount of time.

Add another 500W for two reasonably OC'ed 680's running furmark. Though they're power limited at most OCs in furmark.

So yeah, that's about what one should expect for that. These days, it's only on 4 or 5 hours a week. Obviously, typical game load is much lower.

My HTPC/Media center server idles at 39-41W, increases to 47.5-49W for serving 1 cable channel to a remote xbox. Add in displaying the same tuner signal on its local screen and you're between 49-53W, kick in a second tuner and you're up to 57.5W peak

Ah I was wondering what gpu's you had but that does add up nicely. Nice system though!

 

[inf64]

@IDC
Check out this link .