@Archean
I would beg to differ on a couple of (or three) points.
1. APU (or CPU w/IGP) is not likely to threaten the gaming sector for the foreseeable future. A boon for the netbook, laptop, HTPC and “light” or “casual” gaming sectors, but GPU’s still require shader pipelines for higher resolutions (1080p is fast becoming an industry standard), which leads to higher power consumption, higher transistor density and larger die. Another point to note: Intel’s record with graphics drivers.
2. Workstation marketshare
Figures from the previous link
1.3m shipped ws boards x .875 nvidia market share = 1,137,500 SKU’s.
Quadro price range ~ $120 (QuadroFX 370-budget) -$5000 (Quadro 6000-professional), with an average selling price that exceeds most (if not all-barring triple/quad card setups- desktop graphics solutions), Not including the Quadro Plex 7000 @ $14,500 (yet to launch) or module based solutions such as this
EVGA Quadro Plex 1000.
Point to note: Whereas CUDA is now widely adopted, AMD are reliant upon OpenCL uptake (slow), Linux and OpenGL (less than stellar driver performance, a seeming lack of documentation with the Stream SDK, and long-standing bugged core apps such as
clBuildProgram() (as example), whereas the ability to port CUDA to OpenCL seems to reasonably well established and relatively troublefree. I’m not sure about the worldwide situation, but in this country, any person undergoing medical imaging (CT, MRI, X-ray radiography etc.), auditory mapping etc. , likely has their results compiled and graphically rendered using nvidia hardware and CUDA (or in some cases Red Hat Linux) based programming.
3. HPC. If you believe that parallel computing is not in the future then GPGPU could, I suppose be consigned to the margins. It would seem that
DARPA and a few (new)
supercomputer builders would beg to differ
Note: SKU costs: Tesla C2050 ($2500), C2070 ($3999), S2050 ($12995), S2070 ($18995) - not bad markup for what are essentially GTX465/470 specced GF100 GPU's.
Until nvidia get a significant sales win (as opposed to design win) with Tegra in the smartphone market-or the SoC’s achieve a wider audience, then I definitely wouldn’t see nvidia leaving the compute + software arena. They (in my opinion) be squeezed out of entry level desktop graphics –but only when a significant percentage of OEM systems ship with an APU/CPU+IGP, but a shift in strategy or an evolution in computing doesn’t neccessitate the death of a company (Remember when IBM sold desktop PC’s?).
If you limit the scope of proceedings to desktop gaming (since video encode/playback should be well within the remit of Sandy Bridge/AMD Fusion) then the future could be somewhat more grim if the scenario you envision comes to pass. Like it or not, nvidia’s TWIMTBP is still the pre-eminent game development program. If nvidia exit desktop gaming cards then the program exits too. There is no way in hell that AMD’s Gaming Evolved fills the gap- in fact I wouldn’t be overly surprised if the program gets quietly buried
like it’s predecessor so that every new game is playable on the all new-all singing-all dancing APU (read console port)…but requires just that teeny, tiny bit more graphics horsepower than Intel graphics can supply for good gameplay.
Software has never been AMD's (and ATI's) strong suit. I most certainly wouldn't see them making a strong funding committment to what is in effect a niche market (mainstream/enthusiast PC gaming). History would tend to back that view up I think.
@ruzveh
Thanks for the laugh. It always make me smile when someone invokes
Poe's Law and parodies the non-tech savvy. Keep up the good work.