Intel Silvermont(Next-gen OOE Atom)

If you run linux most of the time H264 decoding hardware won't work :p, unless you shop for the right hardware and use a very recent and fresh distro and/or proprietary driver.
That's where a 22nm Atom with Intel graphics is good, I think. (using open source driver)

With a tablet/phone 22nm Atom with PowerVR, you'd be not safe again. Thanskully Bay Trail (used in Venue 8 Pro) runs Intel graphics.
I've had to check the codename word soup : so Pentium and Celeron are Baytrail-D, Atom Z3680 to Z3770 are Baytrail-T ; the stuff with PowerVR is Merrifield and Mooresfield, and maybe are another subsidized failure to get on the phone market.
 
Last edited by a moderator:
But the Quicksync encoding is garbage. moreso that the one on Haswell
You shouldn't quote old things like that, cause it's no longer true anyway :). All the newly launched Silvermont Celerons / Pentiums seem to have Quicksync enabled (those with a launch date of Q1/14 in the ark.intel.com database), though I don't know if that would include the desktop (J prefix) versions (latest parts there were launched Q4/13).
 
Would Quicksync be worth using on my 3770K? Been doing a lot of encoding lately and it takes quite a while in SUPER to encode a 45min episode.

Which brings me to my next off topic question: what is the best free encoding software? I've always used SUPER but I've never done much encoding until now. It seems slooooow considering the hardware it's running on.

I also have a GTX670. Could that be used for encoding videos?
 
Would Quicksync be worth using on my 3770K? Been doing a lot of encoding lately and it takes quite a while in SUPER to encode a 45min episode.

It's very fast but quality isn't the best.
You'd have to compare the video quality between some custom software encodings and Quicksync to see if it's worth it by yourself.


Which brings me to my next off topic question: what is the best free encoding software?

Opinions may differ, but I personally use Handbrake (the one mentioned in Anandtech's article). It works with Quicksync.




I also have a GTX670. Could that be used for encoding videos?

Yes, but I don't know the freeware options for CUDA encoding.
At least for transcoding, between quicksync (best speed) and software encoding (best quality), CUDA and AMD APP stand in an awkward middle where they're not much faster than software encoding but the quality isn't up to par.
GPGPU does help quite a bit in video creation/editing, but not quite in the encoding/transcoding part.
 
Excellent, informative post, TT. Thank you! (And I don't even faff with video en/transcoding. :))
 
Done quite a bit of testing with my T100 since I got it and always thought the GPU performance was quite close to the original Core i3/i5/i7-series processors. Never had much of a chance to look in to it, but had a bit of a quiet week and was scrapping my old Clarkdale based HTPC the other day and thought I'd take a look at how they stacked up...

This is in NO way a fair fight, on paper the Clarkdale chip has far more graphics resources and a much greater power envelope to work in, but the little Atom does shockingly well in Unigine Heaven and is able to match the Core i5 almost frame for frame in a pretty intensive test, whilst also running in DX11 mode as opposed to DX9. It gets beaten in CPU performance, but even then not by the margin you might expect!

 
Handbrake, yeah I've heard of that. Thanks!

Since I don't use the IGP for anything currently, what will I have to do for Quicksync to work? I think on the Z77 chipset it's possible to use the IGP for Quicksync even if you're not using it for video, but the details escape me.

BTW I am not a quality whore as long as it looks decent.
 
This is in NO way a fair fight, on paper the Clarkdale chip has far more graphics resources and a much greater power envelope to work in, but the little Atom does shockingly well in Unigine Heaven and is able to match the Core i5 almost frame for frame in a pretty intensive test, whilst also running in DX11 mode as opposed to DX9.
Well this isn't THAT unfair. Clarkdale is Gen5 graphics, which wasn't terribly efficient, and while it has 12 EUs instead of the 4 of Bay Trail, these are not really comparable. Even the small Gen6 graphics with 6 EUs (HD2000) would easily beat the Gen5 graphics with 12 EUs - and Gen7 are quite a bit faster still. I wonder actually how it would compare in terms of transistor count (Gen5 may have had more EUs, but as said they aren't as powerful, it still had the dreaded shared mathbox, required threads for interpolation setup / clipping etc.). In terms of samplers / render backends they might also be quite close in theory. I believe Gen7 should also be way more power efficient than Gen5 (aside from any process technology advantage) which should help obviously with the lower tdp of BayTrail (though with the upcoming Gen8 still being quite a bit better there).
 
Hate to be a bother but could a mod please move the Encoding/Quicksync talk to a new thread? I hate to spam this one but we've already kind of dug in.
 
Holy crap Handbrake is sooooo much faster than SUPER. It will complete a 45min episode in more than half the time, not even using Quicksync!

Edit: Scratch that, it's more like 1/4 the time mkay. Avg FPS=~182 compared to SUPER HQ which was around 50fps. :oops:

Yeeps, I can do 4 episodes in the time it took to do one, and the quality is better too.
 
Last edited by a moderator:
This is in NO way a fair fight, on paper the Clarkdale chip has far more graphics resources and a much greater power envelope to work in, but the little Atom does shockingly well in Unigine Heaven and is able to match the Core i5 almost frame for frame in a pretty intensive test, whilst also running in DX11 mode as opposed to DX9. It gets beaten in CPU performance, but even then not by the margin you might expect!

In terms of max flops Bay Trail has the advantage

i5-660: 12 EUs, 4 Flops/EU x 900MHz = 35.2GFlops
Z3740: 4 EUs, 8 Flops/EU x FMA x 667MHz = 42.668GFlops

Z3740's GPU would have to be running at 550MHz to be equal to the max i5-660. I guess that's reasonable.
 
In practice, AIDA64 gpgpu benchmark rates Baytrail-GFX w/4 CUs at around 16 GFLOPS on a Celeron J1900. On Iris Pro, I'm getting a little above 800. :)
 
Back
Top