Return-Path: Received: from mail-ob0-f169.google.com ([209.85.214.169] verified) by media-motion.tv (CommuniGate Pro SMTP 4.2.10) with ESMTP-TLS id 4734732 for AE-List@media-motion.tv; Mon, 04 Jun 2012 20:55:45 +0200 Received: by obbwd18 with SMTP id wd18so7947389obb.28 for ; Mon, 04 Jun 2012 11:58:07 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:in-reply-to:references:from:date:message-id:subject:to :content-type; bh=lu322WBD81U8vMoBxFMtjLnzAEmFlZDJZdNdc9BihrA=; b=vcDbvPpYwec8SoHD9wThH9EvIuoiO5T+bnMk4SoxWSVl4XpGgiLigMblqCvvuAgT8G 5lvOxl8gyNAnU21htUhlstR98WeT2BOfP4itKWoey0L/QouXIHO7aBhzZnZQbl9CauCr Ei8aaVBrk4oIT6IohRLDWeICGLWvxoHn0k2eu7e8UTuQVMv/r25o1fp1TiLyyQnsg4OF ZoUZbOeQu/dGghLxo+KVh6HwzBFCqPWx1FiBnBEjy8Wl3gMGRNrLdSc0rwFAZdoxPNCq HsVPQ2NNWxQg1tSBXooq8i6G5cQd2k7PppO7E0MxlrkWvZ46rad95XSa2c0LSYGyPGn8 CUCg== Received: by 10.182.77.170 with SMTP id t10mr1107661obw.70.1338836287263; Mon, 04 Jun 2012 11:58:07 -0700 (PDT) MIME-Version: 1.0 Received: by 10.182.80.167 with HTTP; Mon, 4 Jun 2012 11:57:27 -0700 (PDT) In-Reply-To: References: From: Teddy Gage Date: Mon, 4 Jun 2012 14:57:27 -0400 Message-ID: Subject: Re: [AE] [OT] Benchmarking AE CUDA on nVidia GTX 580 vs. GTX 680 - SHOCKING RESULTS To: After Effects Mail List Content-Type: multipart/alternative; boundary=f46d044517c150928604c1aa1ded --f46d044517c150928604c1aa1ded Content-Type: text/plain; charset=ISO-8859-1 No, it's because they fundamentally changed the architecture of the gtx 680, my tests confirmed this on multiple machines. The 680 is much more power efficient and is optimized for openGL and directX performance, whereas the 580 has better compute performance even though it has fewer CUDA cores. I'm not sure of the technical reasons why, but it makes sense to extrapolate that two of the fastest GPUS (ie the 590) on one card would be faster than two slower cards (ie the 690). I would definitely go for a 3GB 580 or the 6 GB 590 if it's in your price range. Just about the best card for CUDA that's (well) under $4,000 On Mon, Jun 4, 2012 at 2:47 PM, Andrew Embury wrote: > Why is the 590 beating everyone else out? > > Could it be just a software support issue? > > If that's the case, should I wait for the 680 to catch up or just splurge > on the 590? > > Thanks guys for the time and patience. > > Cheers. > > - Andrew > > > > On Mon, Jun 4, 2012 at 2:34 PM, rendernyc wrote: > >> we did a little bit of testing on a raytrace scene with the q4000, GTX285 >> 480 490 580 590 &680 >> >> heres a chart with some results so far >> https://twitter.com/rendernyc/status/208421894792298496/photo/1/large >> >> >> On Mon, Jun 4, 2012 at 1:38 PM, Andrew Embury wrote: >> >>> Hey Teddy, >>> >>> Anything more to report? I'm looking at building my new PC for the sole >>> purpose of AE/ C4D work and I'm curious to know if you have anything more >>> to report between the GTX 680 and the GTX 580. >>> >>> Thank you ever so much, >>> >>> - Andrew >>> >>> >>> On Sat, May 26, 2012 at 2:53 PM, Todd Kopriva wrote: >>> >>>> I'd be curious to see a test with an animated environment map, since >>>> that will tax the VRAM and the memory bus to the same.**** >>>> >>>> ** ** >>>> >>>> ** ** >>>> >>>> *From:* After Effects Mail List [mailto:AE-List@media-motion.tv] *On >>>> Behalf Of *Teddy Gage >>>> *Sent:* 26May2012 11:50 >>>> *To:* After Effects Mail List >>>> *Subject:* [AE] [OT] Benchmarking AE CUDA on nVidia GTX 580 vs. GTX >>>> 680 - SHOCKING RESULTS**** >>>> >>>> ** ** >>>> >>>> Well, shocking if you care about this sort of thing. So after some >>>> struggles getting the GTX 680 to work with AE CS6 11.0.1 I finally got it >>>> working. How would the 3 GB VRAM GTX 580, with 500 CUDA cores stack up to >>>> the brand new 2 GB VRAM GTX 680 with 1,500 CUDA cores? >>>> >>>> Would it be worth upgrading if you already owned a 580? Would the extra >>>> 1 GB VRAM make a difference for the older card? >>>> >>>> Well I came up with a benchmark (228K) available HEREthat maxes out the GPU and tests your CUDA processing ability. You will >>>> need about 900 MB local space for the output and the new 11.0.1 patch >>>> (probably). >>>> >>>> Now a lot of figures are at play here but with the project using 100% >>>> GPU and 25% CPU I think it's a decent bench for comparing graphics cards. >>>> Here are the results: >>>> >>>> GTX 680 (2GB) = 6 min. 11 sec to render >>>> GTX 680 (2GB) (overclocked) = 5 min. 52 s >>>> GTX 580 (3GB) = 5 min. 42 s >>>> >>>> So the GTX 580 with 3GB VRAM is faster. Now it's hard to say whether >>>> that's because the architecture is more compute-friendly, or the extra GB >>>> of VRAM makes that much of a difference. >>>> >>>> Considering I got the 580 for about $415 shipped used on eBay, I'd say >>>> for now nobody needs to rush out and buy a 680. It's performance is nearly >>>> as good, and great if you are focusing on games, but not for purely compute >>>> / cuda / mercury in CS6 >>>> >>>> I would love to hear some results on a 4GB 680, a 590 or a 690, let me >>>> know >>>> >>>> TG >>>> >>>> -- >>>> Animator & Editor >>>> www.teddygage.com >>>> Brooklyn**** >>>> >>> >>> >> >> >> -- >> danny princz >> >> exposedideas.com >> > > -- Animator & Editor www.teddygage.com Brooklyn --f46d044517c150928604c1aa1ded Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable No, it's because they fundamentally changed the architecture of the gtx= 680, my tests confirmed this on multiple machines. The 680 is much more po= wer efficient and is optimized for openGL and directX performance, whereas = the 580 has better compute performance even though it has fewer CUDA cores.= I'm not sure of the technical reasons why, but it makes sense to extra= polate that two of the fastest GPUS (ie the 590) on one card would be faste= r than two slower cards (ie the 690).

I would definitely go for a 3GB 580 or the 6 GB 590 if it's in your= price range. Just about the best card for CUDA that's (well) under $4,= 000

On Mon, Jun 4, 2012 at 2:47 PM, Andre= w Embury <aembury@gmail.com> wrote:
Why is the 590 beating everyone else out?
Could it be just a software support issue?

If that's the ca= se, should I wait for the 680 to catch up or just splurge on the 590?

Thanks guys for the time and patience.

Cheers.

- Andrew<= /font>



On Mon, Jun 4, 2012 at 2:34 PM, rendernyc <rendernyc@gm= ail.com> wrote:
we did a little bit of testing on a raytrace= scene with the=A0q4000, GTX285 480 490 580 590 &680

heres a chart with some results so far
https://twitter.com/rendernyc/status/20842189= 4792298496/photo/1/large


On Mon, Jun 4, 2012 at 1:= 38 PM, Andrew Embury <aembury@gmail.com> wrote:
Hey Teddy,

Anything more to report? I'm looking at building my n= ew PC for the sole purpose of AE/ C4D work and I'm curious to know if y= ou have anything more to report between the GTX 680 and the GTX 580.

Thank you ever so much,

- Andrew


On Sat, May 26, 2012 at 2:53 PM, Todd Kopriva <= kopriva@adobe.com> wrote:

I'd be cu= rious to see a test with an animated environment map, since that will tax t= he VRAM and the memory bus to the same.

=A0<= /p>

=A0

From:= After Effects Mail List [mailto:AE-List@media-motion.tv] On Behalf Of= Teddy Gage
Sent: 26May2012 11:50
To: After Effects Mail List
Su= bject: [AE] [OT] Benchmarking AE CUDA on nVidia GTX 580 vs. GTX 680 - S= HOCKING RESULTS

=A0

=A0=A0=A0=A0 Well, shocking if you care about this s= ort of thing. So after some struggles getting the GTX 680 to work with AE C= S6 11.0.1 I finally got it working. How would the 3 GB VRAM GTX 580, with 5= 00 CUDA cores stack up to the brand new 2 GB VRAM GTX 680 with 1,500 CUDA c= ores?

Would it be worth upgrading if you already owned a 580? Would the extra= 1 GB VRAM make a difference for the older card?

Well I came up with= a benchmark (228K) available HERE that maxes out the GPU and tests your CUDA p= rocessing ability. You will need about 900 MB local space for the output an= d the new 11.0.1 patch (probably).

Now a lot of figures are at play here but with the project using 100% G= PU and 25% CPU I think it's a decent bench for comparing graphics cards= . Here are the results:

GTX 680 (2GB) =3D 6 min. 11 sec to render GTX 680 (2GB) (overclocked) =3D 5 min. 52 s
GTX 580 (3GB) =3D 5 min. 42 = s

So the GTX 580 with 3GB VRAM is faster. Now it's hard to say w= hether that's because the architecture is more compute-friendly, or the= extra GB of VRAM makes that much of a difference.

Considering I got the 580 for about $415 shipped used on eBay, I'd = say for now nobody needs to rush out and buy a 680. It's performance is= nearly as good, and great if you are focusing on games, but not for purely= compute / cuda / mercury in CS6

I would love to hear some results on a 4GB 680, a 590 or a 690, let me = know

TG

--
Animator &= amp; Editor
www.t= eddygage.com
Brooklyn





<= font color=3D"#888888">--
danny princz

exposedideas.com



--
Animator & Editor
www.teddygage.com
Brooklyn

--f46d044517c150928604c1aa1ded--