Observations with an FX-8350

[SPBHM]

Sure why not, just convince xsplit developers to use it, piece of cake.

last year when I was testing xsplit there was already some users interested on this possibility on their forum, I think it would be great,

Nvidia seems to be using their hardware encoding part of the GPU for their streaming with the "shield" thing, also WiDi uses quick sync I think,

but yes, it's irrelevant when there is no software support,


anyway, I haven't watched the entire video but some of the results posted on that page are really "strange", even if you exclude the FX...

but well, logic would tell me that in games using 4 cores, having HT or additional cores for the video encoding would work well, but it's not always that simple, and honestly, looking at some of the results I think their testing methodology needs some work or clarification...

 

[sequoia464]

I'm not completely on board with their conclusion, but this does show how Piledriver isn't bad if you don't freak over powerdraw. ;p It brings up live streaming of games which seems to have gotten somewhat popular. I'm personally not into it but to each their own.

http://www.youtube.com/watch?v=eu8Sekdb-IE

The test systems - stock results only:

http://teksyndicate.com/videos/amd-...s-3820-gaming-and-xsplit-streaming-benchmarks

Thanks for this, if you happen to see the productivity video please post a link as well.

 

[Fox5]

In the video review he benchmarks with xsplit live streaming, a perfect multitasking scenario for the nature of the chip, xsplit uses x264 for encoding the stream and it clearly shows that the 8 integer cores of the FX carry more workload and higher fps than Intels quads and ht.

It sounded like only crysis was benchmarked with xsplit to show that AMD could pull ahead. The other benchmarks seemed to be just normal game benchmarks.

 

[Idontcare]

I'm still optimizing my 8350's OCs with the stock HSF. The good news to report there is that my previously observed Vmin's (minimum voltages needed at various clockspeeds for LinX stability) were way too high for one very specific reason - not enough air cooling was being provided to the motherboard's passive heatsinks that are atop the vrms and chipset chips.

I added two strategically located case fans to blow air more directly onto the surface of the mobo and now my 4GHz Vcore has dropped from ~1.32 to a much lower 1.244V. This enabled me to hit 4.3GHz with 1.347V, LinX stable.

I would have never expected the minimum Vcore could be lowered so dramatically on the basis of just getting the mobo cooled.

So my advice to anyone who is OC'ing these 8350's: before you settle on a given Vcore for your OC at any given clockspeed, adds ridiculous amounts of air cooling to the mobo just to see if it makes a materially significant difference to your OC like it does to mine :thumbsup:

 

[lau808]

new line chart/graph please

 

[lau808]

i love your threads idc, you are very detailed in every aspect. even your posts in other threads as well. you are by far my favorite poster great job keep up the good work, or play or whatever lol

 

[Idontcare]

:$ thanks for the kind words lau808 :$

new line chart/graph please

Coming right up! I've got one more data point to nail down (Vmin for 2GHz) and then I'll post up the goods :thumbsup:

 

[Ajay]

i love your threads idc, you are very detailed in every aspect. even your posts in other threads as well. you are by far my favorite poster great job keep up the good work, or play or whatever lol

I agree with you 100%. He's an engineer through and through. Even in his hobby, he lives up to the high standards that all engineers should have.

 

[sequoia464]

So my advice to anyone who is OC'ing these 8350's: before you settle on a given Vcore for your OC at any given clockspeed, adds ridiculous amounts of air cooling to the mobo just to see if it makes a materially significant difference to your OC like it does to mine :thumbsup:

Thanks for the heads-up, I'm just using an 8320 with what now appears to be way too much voltage - 1.416 for 4300MHz. (on a Sabertooth)

I'll set up a couple of fans this weekend and see if I can bring that Vcore down some.

 

[Greenlepricon]

I might ramp up the cooling on my rig too. I thought it was fine but I didn't know that it would make that big a difference. Nice work!

 

[Idontcare]

This is extra cooling from just pointing a case fan directly towards the passive VRM heatsink that is on the mobo.

I feel like 4.5GHz is probably within striking range with the stock HSF and at something well below the 1.5V I thought I was going to have to use. Gonna try tomorrow

 

[Gikaseixas]

Great info IDC, gonna try this later today

 

[bgt]

Good find IDC :thumbsup:

 

[jvroig]

I feel like 4.5GHz is probably within striking range with the stock HSF and at something well below the 1.5V I thought I was going to have to use. Gonna try tomorrow

Incredible discovery :thumbsup:

You've shown us the scaling, can you also show us how it affected power consumption? That's a pretty significant voltage reduction, I would expect over 30W at max load, if Deneb and Thuban voltage-power scaling is an indicator.


What's really amazing to me is it is still all at stock. Great work, carry on! :salute:

 

[Idontcare]

Incredible discovery :thumbsup:

You've shown us the scaling, can you also show us how it affected power consumption? That's a pretty significant voltage reduction, I would expect over 30W at max load, if Deneb and Thuban voltage-power scaling is an indicator.


What's really amazing to me is it is still all at stock. Great work, carry on! :salute:

I expected the power savings to be higher, like yourself, but they come in at about half your expectations - at 4GHz it was a 14W reduction at the wall (278W -> 264W) and a 4°C reduction for the CPU (55°C -> 51°C, 19.1°C ambient).

That said, 14W is a significant percentage when reduced to the CPU's contribution alone - that is probably around a 10% reduction in the CPU's power consumption.

What impresses me is just how well the stock HSF performs. To replace it with a 3rd party HSF costs a bundle but at best is going to net another 10% performance boost from OC'ing (maybe another 400MHz?). At that point it really is a questionable investment to OC on anything other than the stock HSF.

 

[Rvenger]

I expected the power savings to be higher, like yourself, but they come in at about half your expectations - at 4GHz it was a 14W reduction at the wall (278W -> 264W) and a 4°C reduction for the CPU (55°C -> 51°C, 19.1°C ambient).

That said, 14W is a significant percentage when reduced to the CPU's contribution alone - that is probably around a 10% reduction in the CPU's power consumption.

What impresses me is just how well the stock HSF performs. To replace it with a 3rd party HSF costs a bundle but at best is going to net another 10% performance boost from OC'ing (maybe another 400MHz?). At that point it really is a questionable investment to OC on anything other than the stock HSF.

If the CPU was better binned it would probably be worth it.

 

[Homeles]

I expected the power savings to be higher, like yourself, but they come in at about half your expectations - at 4GHz it was a 14W reduction at the wall (278W -> 264W) and a 4°C reduction for the CPU (55°C -> 51°C, 19.1°C ambient).

That said, 14W is a significant percentage when reduced to the CPU's contribution alone - that is probably around a 10% reduction in the CPU's power consumption.

What impresses me is just how well the stock HSF performs. To replace it with a 3rd party HSF costs a bundle but at best is going to net another 10% performance boost from OC'ing (maybe another 400MHz?). At that point it really is a questionable investment to OC on anything other than the stock HSF.

The thing about coolers is that they're transferable to future machines. So the investment can certainly be worth it over long periods of time.

 

[Idontcare]

For me, gaming is mostly an after-thought when it comes to what I do with my computers. And even then what little gaming I do, it is usually with legacy games that aren't very demanding. (my most modern game is Dragon Age: Origins)

When it comes to compute-intensive tasks, I use three very specific apps: TMPGEnc (transcoding DVDs), Gaussian (computational chemistry), and Metatrader (programming foreign currency trading algorithms).

When it comes to transcoding DVDs there is no shortage of apps out there. I specifically use TMPGEnc because it gives me the kind of IQ (image quality) that I want at the smallest bit-rate (smallest filesize) that I need so I can cram as many movies and episodes onto a dual-layer DVD disk as possible (or hard-drive on my HTPC).

When transcoding with the goal of reducing file-size while maintaining IQ, eliminating video noise is absolutely critical. Pixel noise takes up a lot of your available bandwidth because the compression algorithms will attempt to replicate the video noise as much as possible, leaving less bits per second to allocate towards attempting to faithfully replicate the rest of the image.

To this end I tend to remove both spatial and temporal video noise, which makes the transcode process time-intensive in its own right, in addition to then attempting to compress the images further while maintaining good image quality.

To benchmark the transcoding performance with TMPGEnc VMW 5 (Ver 5.3.1.85), I ripped a portion of one of my daughter's little mermaid dvd's which contained an 8 minute short movie clip of Mermbabies (a movie clip from the 1938). In its original format on the DVD the clip consumed 599MB, I transcoded the movie clip (applying spatial and temporal video de-noise filters) and compressed the bandwidth to 3.7Mbps which results in a final file size of 231MB...reducing its footprint to ~38% the original size without sacrificing the image quality (as far as my layman's eye can tell).

Everything about this particular transcode job is very much a realistic proxy for how I transcode most of my children's animated movies and episodes. Using the trial version of TMPGEnc VMW 5 (Ver 5.3.1.85) I benchmarked the transcoding job time for Mermbabies at different clockspeeds and compiled the following results:

This really shocked me. Based on the general consensus on the web - that bulldozer/piledriver may suck at single-threaded apps but it rulez when it comes to multi-threaded apps (particularly transcoding apps) - I have every expectation that the FX-8350 would dominate this specific application without question.

And yet it comes up way short.

A note regarding hardware configurations - the 3770k and fx8350 rigs are identically configured (same DDR3-1866 ram, GTX460 vcard, OCZ V3 ssd, Win7 x64 ult, etc), whereas the Q6600 rig is outdated with WinXP, 4GB DDR2-734, 160GB spindle hdd and passive cooled ATI vcard from 2006.

That said, at 4.5GHz the FX-8350 barely bests the 3770K clocked at 3GHz

Oh well then, so I won't be using my FX-8350 as my main transcode computer. I've got two other dedicated workstations I can farm it out for (gaussian and metatrader)...so let's see which it is going to be best at. (next post)

 

[Homeles]

Dat Ivy Bridge scaling.

At least the AMD crowd can be happy that they're beating the venerable Core 2 Quad, although now that I've read your systems setups, maybe not.

 

[guskline]

Once again, great work IDC. First, I'm glad you could use your 8350 with a stock HSF. I knew I wanted to push mine and had some gift card $$ to buy the Corsair H100. I also like your find about cooling the vrm. I use a CM 932 Advanced case with an exhaust fan behind the VRM, fans blowing down on it (the push/ pull fans from the H100) and the huge side case fan drwing cool air in.

Sorry to see the poor performance relative to the 3770k on the transcode job but let's not kid ourselves, the 3770k is a superior, and more expensive, cpu.

I really enjoy your posts. Accurate and without the Intel vs AMD confrontational language.

Have you tried to run OCCP and/or AMD Overdrive stablity test with your 8350 set at 4.5Ghz?

 

[jvroig]

t least the AMD crowd can be happy that they're beating the venerable Core 2 Quad, although now that I've read your systems setups, maybe not.

It's beating the Core2Quad using twice the core count and faster memory, so most likely not. Even then, it only gets ~33% more FPS.

 

[Idontcare]

Dat Ivy Bridge scaling.

At least the AMD crowd can be happy that they're beating the venerable Core 2 Quad, although now that I've read your systems setups, maybe not.

It's beating the Core2Quad using twice the core count and faster memory, so most likely not. Even then, it only gets ~33% more FPS.

It blows the Q6600 out of the water. At 4GHz the FX8350 generates 40% more FPS than a Q6600 scaled to 4GHz, but comparing stock-to-stock the 8350 is a solid 107% faster than the Q6600.

Sorry to see the poor performance relative to the 3770k on the transcode job but let's not kid ourselves, the 3770k is a superior, and more expensive, cpu.

That fact cannot be overlooked. This is a $200 CPU being compared to a CPU that costs $330 and a CPU that cost me ~$270 so many years ago.

It is not so surprising in terms of price/performance where the FX8350 lands on the chart, what I was more surprised about is that I was (perhaps naively) expecting this to be one of those rare corner-case niche application for which the FX8350 would rise up and really trounce the 3770k.

I fully admit this was an unfair expectation on my part. Hope got the better of me here. But as you rightly point out, there is a reason it is priced at $200 and not $330 :thumbsup:

 

[inf64]

Maybe it's just that app you are using? Have you tried some other similar software?

 

[jvroig]

It blows the Q6600 out of the water. At 4GHz the FX8350 generates 40% more FPS than a Q6600 scaled to 4GHz, but comparing stock-to-stock the 8350 is a solid 107% faster than the Q6600.

Is it 40%? I was looking at the last dot of the Q6600, (3.5?) from there it looks like 40FPS vs 30FPS, so on a clock-to-clock basis it's 33%, although it (Q6600) does seem to be scaling poorer as the clock increases.

Out of the box, of course, hands down the FX8350 performs better. This is not news, because AMD has been beating the Q6600 in out-of-the-box performance since higher-clocked Denebs (Q6600 is clocked pretty low), then Thubans, then Bulldozer. I therefore assumed Homeles' comment was in some context other than out-of-the-box, although I admittedly may have settled unfairly on an IPC / per-core context that was unwarrated, which led me to conclude it's still not a major ass-whooping architecturally against the Core2Quad, since it has 100% more cores, but only manages 33% more performance clock to clock.

 

[jvroig]

Maybe it's just that app you are using? Have you tried some other similar software?

It is exactly just his app. Idontcare is drawing no other conclusion than "for this specific app I use, it is not as good as I had hoped".

He is not making the conclusion it is no good for transcoding/encoding in general. He made that rather clear.