obobskivich wrote:
I've always wondered about that with SSE3, since it was largely a Prescott update (and AMD's implementation, at least on K8, is not as "complete," because they dropped all of the Prescott-specific stuff).
obobskivich wrote:
With GPU decoding, my P4 has no trouble with 1080p Flash video. My dual socket 604 machine can do that on the CPUs, but it puts both of them at 100%. I do agree with the rest of what you're saying based on those two machines as well. 😀
The one application I've used that uses SSE3 (or at least acknowledges its presence) is the Dolphin emulator. It always displayed D3D/SSE3 when I ran it on my Opteron 185 years ago. Dolphin has always been ahead of the game in that regard. Much like x264, Dolphin can tap into SSE4 on Nehalem CPUs and AVX on Sandy Bridge and later processors. I haven't tried Dolphin on Core 2 yet. I wonder if it can make use of the S-SSE3 instructions on those CPUs.
Another interesting case is Google's VP9, the codec Youtube streams to Chrome. I've always thought that VP9 was at least partially decoded in hardware, but it is not. It's all done on the CPU, and indeed, low power CPUs like AMD A4 and Intel Silvermont just don't have the horsepower to tackle 1080p and/or 60fps VP9 playback under Chrome. These devices can, however, handle 1080p/60fps playback under IE11, since Youtube streams bog standard H.264 to IE11's HTML5 player.
Anyway, I *believe* VP9 uses SSE4 at the very least because the Core i series CPUs are far more efficient at decoding it than previous CPUs. I "benchmarked" a few of my CPUs by playing back VP9 encoded YouTube video using Chrome's HTML5 player, and the results were quite interesting.
Opteron 185 (2C/2T, 3GHz OC, SSE3, GTX 560)
720p: 40% average CPU utilization
1080p: 85%
720p/60: 95% and still completely smooth
1080p/60: 100%, playing back at roughly half the frame rate.
Core 2 Quad Q6700 (4C/4T, 3.33GHz OC, S-SSE3, GTX 560)
720p: 20%
1080p: 45%
720p/60: 35% (Not sure why 720p/60 uses less CPU time than 1080/30 in this case, but it is what it is)
1080p/60: 92%
Core i7 4930K (6C/12T, 4.5GHz OC, SSE4 & AVX, GTX 970)
720p: 0-1%
1080p: 1-2%
720p/60: 1-2%
1080p/60: 2-3%
The very low CPU usage on the i7 is what made me initially believe that VP9 was hardware accelerated. However, GPU-Z shows zero load on the GPU's Video Engine during playback, and according to everything I've read, no GPU currently offloads VP9. I remember SSE giving software MPEG-2 a huge boost back in the day, perhaps SSE4 is doing the same thing with modern codecs.
obobskivich wrote:
Also don't install/use the nVidia firewall (and/or if the board has a non-nVidia NIC, use that). 😵 (see here for a brief summary: https://en.wikipedia.org/wiki/NForce4#Flaws)
IME I've had no problems with my Xpress 200 CF (when the board still worked) and 4GB of RAM (it will run 2T DDR333, like any other 939 board with all four banks populated), but my A8N-SLI flat out refuses to recognize more than around 3300MB of memory (doesn't matter how much chocolate I promise to share with it). The single VIA 939 board (Abit AV8, with VIA K8T800Pro) I have has been fantastic thus far, but I've had it maybe a week, and its only done some prelim testing - I haven't lived with it for any length of time, nor have I tried 4GB of RAM in it.
Interesting. I too have an A8N-SLI tucked away in storage, and that board has no problem recognizing 4GB of RAM. It is running one of the newer BIOS revisions though. Perhaps the ATI and VIA boards that I used also had BIOS updates enabling the use of 4GB of RAM.
94 MHz NEC VR4300 | SGI Reality CoPro | 8MB RDRAM | Each game gets its own SSD - nooice!