FX-8320 overclocking (probably obvious)

[teflon6678]

Hi guys,

It's about time I move on from my ageing Core 2 Quad, and my feeling is that when AMD react to Haswell, I'm probably going to hop on with their system and a FX-8320 or FX-8350. They're cheaper, for one thing, the socket will have more support so I'm not staring at another motherboard replacement, and the extra physical cores seem to be good for on-the-fly encoding of gaming video and streaming, where that isn't giving the CPU the most even of workloads

I just have a few questions, which I haven't been able to satisfy through my own research, and was wondering if you all could help. Forgive me for not reading through page after page of forum threads, but it gets a bit much, sometimes!

- I've seen that the TDP for the FX-8350 in particular might be getting broken even at stock clocks, and was wondering if this had been diffinitively proven, or if mitigating circumstances with motherboard etc. was the reasoning behind this?

- With that in mind, how regular is it to be able to undervolt the CPU, and actually lower the power consumption at stock clocks?

- With an 8320, I know they do overclock well enough to 4Ghz. Is Turbo also something which can be modified, so that I could get an exact equivalent to an 8350, and save my pennies? Would I then be right to expect 8350 levels of power consumption?

- I'm asking lots about power, because I have a fairly recent 500W PSU, which is serving me well. With a stock 8350 and a mildly OC Radeon 7850, this should all be plenty under 500W when gaming, shouldn't it?

- This is all coming to around £250 for me. Is there any particularly good argument why I shouldn't go down this path at this price point? As far as I'm aware, an i5 would be either much more expensive and/or not have the streaming/recording quite as well covered.

Thanks in advance, and again, apologies for being a bit dim with all of this!

 

[bononos]

Yes it is a problem. Search for Idontcare's threads and look for his 8350 one. (With any luck, he'll probably pop in to post the link).
Idontcare's testing with LinX showed that a stock 8350 can eat up ~200W just for the cpu. There are other reviews showing that this is indeed the case and MSI have basically thrown up their hands with their motherboard throttling workaround and told interested customers to direct their enquiries to AMD as to why the cpu exceeds the spec.

 

[Vic Vega]

Did Anand mention this in his review?

 

[inf64]

Yes it is a problem. Search for Idontcare's threads and look for his 8350 one. (With any luck, he'll probably pop in to post the link).
Idontcare's testing with LinX showed that a stock 8350 can eat up ~200W just for the cpu. There are other reviews showing that this is indeed the case and MSI have basically thrown up their hands with their motherboard throttling workaround and told interested customers to direct their enquiries to AMD as to why the cpu exceeds the spec.

This is a good post on this subject:

This is getting ridiculous.

Unless there is a major node overhaul (which reduces the power consumption by 30-50%) there is absolutely no way AMD can release a 8-core FX clocked to 5.0GHz. Nothing indicates that there would be a new node version coming and releasing one would not make any sense anyway. There is no point in using the limited resources to anything else but on the Gen. 3.

A nasty thing to say but using the limited resources on Piledriver would be like force feeding hay to a dead horse.

Currently the fastest 8-core FX is clocked to 4.0GHz.
At this frequency the 125W TDP envelope is already fully used in well optimized multithreaded applications (such as rendering or video encoding), there is no headroom really.

A proper binning of the dies can make around 200MHz of difference on current Piledriver dies.
The additional scalability comes purely from the lower leakage properties of the binned parts.

The current Piledriver cores do not scale well beyond 4.6GHz, which seem to be the critical frequency really.
They can be pushed beyond that but the increase in the power consumption & emitted thermal power can no longer be justified by the resulting performance increase. Beyond this point the performance-power curve is anything but linear.

We might see a 8-core FX (Piledriver) clocked to 4.3 - 4.5GHz (base frequency) at some point, but anything higher than that is very unlikely.

Even if AMD could produce a 8-core FX clocked to 5.0GHz, there would not be any motherboards for the chip.
A 5.0GHz 8-core Piledriver FX consumes 180W - 200W of power during Cinebench R11.5 for example (DCR Pmax, from CPU VRM).

In some applications the power consumption can be even 30% higher than that so the motherboard design has to have enough of headroom to accomodate the difference + the design margin. The motherboard would need to be able to deliver 235W - 260W of constant power. After the average conversion (VRM efficiency) loss the total VRM input power would be 294W - 325W (24.5 - 27A from the 12V rail).

The power dissipation of the CPU VRM alone would be 36W - 46W.

There are only couple of motherboards which can temporarily manage such of power requirements.
Even these motherboards cannot support such power requirements reliably for extended periods of time.

It's not just the CPU and that's why IDC measured this kind of power draw. The surrounding circuit(it's efficiency) plays a big role in measured power draw.

 

[bononos]

This is a good post on this subject:


It's not just the CPU and that's why IDC measured this kind of power draw. The surrounding circuit(it's efficiency) plays a big role in measured power draw.

Yes I know that IDC measured the total power draw but the figure was amazing. Close to 300W just for LinX at stock speeds. There is no avoiding the fact that something is wrong with BD/PD.

 

[Yuriman]

Food for thought, a lot of the power circuitry is moving on-die with Haswell so there's a good chance motherboard prices will drop significantly. If it doesn't, chalk it up to greed.

 

[pyjujiop]

It sucks down some juice, that's for sure. I just ran Cinebench 11.5 on my 8350, hard-set at 4.2 GHz, and the total power draw for the system rose from 191W to as much as 369W during the test. There's a lot of hardware in this rig, but the 8350 had to have been pulling almost all of that extra power. I rarely pulled that when I was running an overclocked, overvolted X6 Thuban and a 3870x2 video card.

AMD needs to do more than just change to hard flip-flops to get the power consumption down.

 

[Headfoot]

have you priced a similar i5 3570k set up? I'm not familiar with UK prices so I don't know what that will cost. If you want to overclock you can get to 4.3-4.4 with a Coolermaster Hyper 212 and a cheap motherboard almost guaranteed with the i5. It's hitting 4.5-4.7 that will require the high end stuff. Even overclocked the i5 will draw about the same or less power at the wall than the 8320 at stock given total motherboard and CPU consumption fully loaded. Most games don't use more than 2.5-3 cores so the 4th will easily be able to handle streaming, you don't need "8" cores for that (not even really 8 cores, 4 modules with CMT). I understand if it's a lot more money it might not be worth it, but make sure you compare apples-to-apples first

 

[teflon6678]

Thanks guys. Definitely food for thought. It does sound like the 8320 and 8350 aren't coming out of the factory with the highest level of quality. I guess I'll have to root around a bit more.

@Vic - Anand's review seems to have grabbed a perfect 125W model. It idles at 74W and is at load with 195W...

@Yuriman - Doubt prices will drop. For one thing, they have to cover that USB3 hardware bug, but greed is always going to win.

@Headfoot - Yeah, I've been browsing around for a while. A 3570k set up would be around £80 more than what I'm looking at with an 8320. They throw a lot more features at AMDs for less cost, with USB 3 headers, SATA 3 etc. etc.
I was thinking FX-8320 (£130) and AsRock 970 extreme4 (£75), which jumps to 3570k (£180) and an AsRock z77 extreme4 (£105), with an i5. (That's just a rough equivalent) It's doable for me, but might take a while longer...

 

[Idontcare]

It does sound like the 8320 and 8350 aren't coming out of the factory with the highest level of quality.

It is a huge chip, twice as large as an Ivy Bridge. That makes it cost a lot more money, then consider GloFo isn't a charity, they want profits selling the wafers to AMD too. That makes it even more expensive. Yields won't be good because the chip is large and you are looking at buying the 8-core model which can't benefit from die-harvesting like the 6 and 4 core models.

And then to top it off they have to sell them on the cheap because the performance is not good compared to the competition.

So what is AMD, a financially besieged AMD at that, to do with their binning of 8-core Piledrivers? Basically they have to open up the tolerances really wide, let all kinds of craptastic silicon get through the validators and binning, and try to save a few pennies by selling as much junk silicon as possible.

There are no max temperature spec and no max voltage spec for these Piledriver FX chips, so AMD can basically bin anything and everything that walks with all 8-cores functioning by dialing in the voltage as needed to hit the clockspeed targets.

Ordinarily that would be a problem in terms of staying below a TDP spec value but as we learned from MSI's engineers, AMD isn't interested in adhering to their own TDP spec.

So it has very much become a "silicon lottery" in more ways than it was traditionally.

This is pretty standard stuff though, companies under financial duress will start cutting back on quality almost right off the bat. They also start cutting corners when it comes to accounting regularities and so on. Which is why investors get psychologically skittish when they see smoke (signs of poor financials at a company) because they rightly assume it is the tip of the iceberg.

I could name names, but I'd be foolish to do so, when TI was still in the foundry business (we did foundry work for decades down to 65nm) we had a customer who was on their last legs, really backed into a corner by a combination of inept management and savvy competitors.

These guys were desperate, how desperate? You know how chips are not round but the wafers are round? So the edge die always have their corners cut off and don't have proper die-seals in place to protect against environmental deterioration after die-cut?

Well this company was so hard-up to lower their chip costs that they contracted with us to sell them those semi-functional edge die missing all the corners. They knew they'd last maybe 6 months because of the missing die-seals, and they shipped the chips inside devices even though they were literally missing entire circuits.

Quality was crap, but the company was about to go bankrupt, no one at the company cared if customers got products that simply didn't function. If the returns started coming back then they'd deal with it then, if they still had enough cash to service the warranty claims that is.

It was pretty bad, but for a short while (few months) we had a particular product line that was technically yielding more than 100% because we were selling more die per wafer than there were technically die per wafer because you aren't supposed to count edge-die without corners as sellable dies

I'm not saying AMD is selling FX chips with their corners missing, I am saying that when the chips are down and a company is backed up against the wall then quality always goes out the window. Always.

 

[teflon6678]

That's a crazy, crazy story. I hope AMD's console gigs can give them a platform to get back into the game.

That's part of why I'm still tempted, in spite of all this, to go with an AMD set up... Play the role of the greater fool.

So, though I'll give this more thought before I take the plunge in either direction, it all hangs on the total power consumption I could reasonably expect.

Putting a kind of worst case scenario into a PSU calculator, with an 8350 "overclocked" to give 200W, a 7870 to account for my 7850's OC, 2 sticks of DDR3, 2 SATA drives, an SSD, a couple USB things, 3 140mm fans. 100% utilisation, 10% ageing and 100% CPU utilisation plops out with a recommended wattage just a shade under 500W... Which suits my PSU exactly!!

Does that sound like a fair assessment? I'm really just trying to figure out if an 8350 would still be plausible, or if I'd be running too close to the line?

Of course, feel free to tell me I'm being silly for even considering the AMDs still.

 

[blckgrffn]

You're silly.

I think you'll likely be fine on the PSU so long as you aren't doing something like distributed computing. It's pretty unlikely even things like games are going to hit your GPU and CPU for concurrent 100% utilization. Either your video card or CPU will be bottlenecking the other.

The 8350 is a decent match for your GPU though, and even though it is inefficient at load and not as speedy as an Intel chip could be, it still idles nicely and packs a pretty modern punch. You probably would be hard pressed to tell the difference between it and an i5 in real world use with that GPU.

*shrug* I still recommend AMD for people with the right needs/budget. If it works for you - it works for you.

 

[Headfoot]

You should take a look at your PSU and let us know exactly what model it is. Once you start getting close to the rated number of the PSU you have to look at make sure that the number is for "continuous usage," and not for "peak power." Your power supply may be able to handle a lower wattage on a 24/7 basis. This is all usually on the label on the side of the power supply, otherwise we can google the particular model.

 

[teflon6678]

This one's an oddity. I bought an "own brand" PSU from a local store called Novatech. However, I'd done a bit of backtracking, just to be sure it wasn't total junk, and found that their stuff was made by Sirfa, effectively being rebranded OCZ parts.

The store page is here:
http://www.novatech.co.uk/products/components/powersupplies/nov-psg500.html

A few forum threads which back that up:
http://forums.bit-tech.net/showthread.php?t=195579&page=2
http://www.jonnyguru.com/forums/showthread.php?t=7519&page=3

a review of the equivalent OCZ 500W:
http://www.bit-tech.net/hardware/psus/2009/04/02/ocz-modxstream-pro-500w-psu-review/1

and finally, the part you've all been waiting for, a photo of the side:

 

[VirtualLarry]

No UL listing == unsafe junk PSU. Get rid of it and get a decent PSU.

 

[Headfoot]

Excellent, thanks for doing the legwork. Looks like a reasonably honest 500w supply to me. One consideration is that this supply has 2 separate 12v rails which generally (in practice) don't respond as well as a single, high amperage rail would near its rated maximum. Still, I think you'll be fine regarding power with either set up

 

[USER8000]

This one's an oddity. I bought an "own brand" PSU from a local store called Novatech. However, I'd done a bit of backtracking, just to be sure it wasn't total junk, and found that their stuff was made by Sirfa, effectively being rebranded OCZ parts.

The store page is here:
http://www.novatech.co.uk/products/components/powersupplies/nov-psg500.html

A few forum threads which back that up:
http://forums.bit-tech.net/showthread.php?t=195579&page=2
http://www.jonnyguru.com/forums/showthread.php?t=7519&page=3

a review of the equivalent OCZ 500W:
http://www.bit-tech.net/hardware/psus/2009/04/02/ocz-modxstream-pro-500w-psu-review/1

and finally, the part you've all been waiting for, a photo of the side:

They seem similar to the OCZ ModXStream and StealthXStream units if you look at the internal pictures of the units.

The 500W is 80+ certified:

http://www.plugloadsolutions.com/psu_reports/NOVATECH_POWER%20STATION%20%28GAMING%29%20500_ECOS%20638.1_500W_Report.pdf

The PSU should be fine. Here are some figures under gaming loads for the FX8350 with a GTX680 and a HD7970:

http://hexus.net/tech/reviews/cpu/46985-amd-fx-8350/?page=6
http://www.legitreviews.com/article/2055/13/

Here are figures with an HD7950 and a HD5870 or GTX680(Anandtech does not clarify this) under a video encoding load:

http://techreport.com/review/23750/amd-fx-8350-processor-reviewed/4
http://www.anandtech.com/show/6396/the-vishera-review-amd-fx8350-fx8320-fx6300-and-fx4300-tested/6

All figures are at the wall.

The OP might also want to look at an FX6300 or the upcoming FX6350. The former is fine in a range of decent 970 motherboards too like the ASUS M5A97 EVO and Gigabyte GA-970A-UD3 if you want to get a decent enough overclock out of it.

The ASUS Crosshair V Formula is not also the most efficient AMD motherboard either and since the OP has a lower end graphics card,power consumption should be less anyway.

The higher end cards add anywhere from around 40W to 90W additional power consumption at the PCI-E power connectors,which should be somewhat higher at the wall:

http://www.techpowerup.com/reviews/Club_3D/HD_7850_RoyalQueen/26.html

Even with a HD7850 running at a 140MHZ overlock,the higher end cards add around 30W to 75W more:

http://www.techpowerup.com/reviews/Powercolor/HD_7850_PCS_Plus/26.html

Regarding the MSI motherboards,they are known to have had VRM failures with not only a number of their AMD motherboards,but even some of their Intel ones. Go onto overclock.net and you can see a few examples.

I even knew someone who had a high end P55 one with a Core i5 750 which just crapped out when running the CPU with a mild overclock,taking out the CPU too. He was none too pleased.

They do have a habit of underspeccing stuff at times.

 

[bononos]

......
These guys were desperate, how desperate? You know how chips are not round but the wafers are round? So the edge die always have their corners cut off and don't have proper die-seals in place to protect against environmental deterioration after die-cut?

Well this company was so hard-up to lower their chip costs that they contracted with us to sell them those semi-functional edge die missing all the corners. They knew they'd last maybe 6 months because of the missing die-seals, and they shipped the chips inside devices even though they were literally missing entire circuits.

.......

Didn't know such a thing could even contemplated let alone had been done before.

Its just more confirmation that AMD is going down in the short-medium time frame.

 

[teflon6678]

@Larry - I know the risks, but feel that the research I did in the run up does properly track the PSUs back to a reputable source. It'll do for now, anyway!

@Headfoot - I'll have to note that down for next time I need to go PSU hunting.

@USER8000 - I can't reply to your DM, because I don't have 25 posts, but cheers for the pointers and all the ideas in your post. The 6300 is still on my radar, and we'll see what the 6350 brings.

Thanks everyone.

 

[Playerdvlkng89]

Iam using an Amd FX-8320 with a CROSSHAIR V Formula-Z and a ASUS HD 7970
and i have Antec 1000w PSU Model TPQ-1000.
Iam still trying to OC with an Noctua NH-D14 cooler and at 4.2 GHz at 1.356 i get 43-44 degres celcius anything i should think about?

Thx

 

[jaqie]

just for a bit of historical data: my thuban x6 has a tdp listed in cpu-z as 161w while amd claimed it was 125w. It really ate the power.

This doesn't seem to be a new thing for AMD to do.

I still like their products when they hold water, but the newest stuff is a sieve.

 

[jacktesterson]

I'm running an overclocked 7850 (1050/1325 - not a massive OC) and a 8320 @ 4.7 GHz 24/7 off a 550w Power Supply - Also OCZ, although its a little better quality. (OCZ ZT 550w)

 

[SlowSpyder]

just for a bit of historical data: my thuban x6 has a tdp listed in cpu-z as 161w while amd claimed it was 125w. It really ate the power.

This doesn't seem to be a new thing for AMD to do.

I still like their products when they hold water, but the newest stuff is a sieve.

Do you have a screenshot of that by chance? I thought CPU-Z was more or less just a database that looked at what model CPU you had and displayed information based on that. I don't think it actually calculated a TDP or anything like that..?



Does anyone have a link to the discussion with the MSI engineers? I'd like to read it.

 

[jaqie]

Do you have a screenshot of that by chance? I thought CPU-Z was more or less just a database that looked at what model CPU you had and displayed information based on that. I don't think it actually calculated a TDP or anything like that..?

This is a 960T (zosma) 3GHz black edition x4, 95w advertised tdp. unlock two cores, it becomes a thuban 3GHz x6, 125w tdp advertised, but obviously not accurate.

I have had the max TDP read different based on different overclocks with prime95.

 

[guskline]

I have both a FX 8350 and FX 8320. I use Asus Sabertooth 990FX mbs (Rev1 and Rev2 respectively) with them. The 8350 runs at 4.6Ghz (21 x219) at vcore 1.46 cooled by a Corsair H100 AIO and the 8320 runs at 4.3Ghz (21.5 x 300) at vcore 1.43 cooled by a Thermaltake Water 2.0 Pro AIO.

At max Intel burn or Prime95 temps never exceed 59C and for regular gaming are in the 40s.

They do suck power but give you decent performance for the price.

My flagship is an Intel 3770k. see my sig below.