By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Soleron on 14 January 2013
Edit Thread
Pemalite said:
BlueFalcon said:

...


...

Soleron said:
@Permalite

ARM chips will not work out for them, because they aren't customising the core so they'll be selling the same or worse as what everyone else with more experience is selling.

I dimiss GPU compute out of hand. I believe it will never be relevant no matter its theoretical strenghts. HSA is the latest in a series of failed AMD initiatives along these lines -> Close To Metal, Torrenza, the R6xx tesselator.

ARM chips can work out, Seamicro has excellent relations to allot of players in the server industry so they already have a foot in the door.
If you are going with a very slow but energy efficient architecture like ARM, then it doesn't matter how each core performs, what does matter is how dense you can make the server, if you want single-threaded performance then go Opteron or Xeon, ARM after-all is in the same league as the Intel Atom/AMD Brazos, so you can't be to worried about the performance in the first place.

As for the R600. It wasn't AMD's first chip to have Tessellation, Look back at the Radeon 9000 and 8000 series with it's Truform, you could even set up The Elder Scrolls: Morrowind (Direct X 7/8 game) to take advantage of it.
Besides, it was a first step and it's probably in part thanks to AMD collaborating with nVidia and Microsoft on it becoming a feature in Direct X 11.
Now we need those consoles to catch up so Console Ports will feature Tessellation more heavily and not look so crap.

disolitude said:


...


That is true. However, DDR3 1600mhz memory is the sweet spot in terms of price/performance, it doesn't make sense for someone to spend-up big on the fastest DDR3 Ram in a cheap Trinity notebook.

It actually gets me excited about the move to DDR4 soon and what kind of tangible increases it may bring on the IGP front. :)

BlueFalcon said:

...

With Intel's QuickSync (Provided it's not a socket 2011 processor) and AMD/nVidia GPU's, encoding can be moved off the CPU.

- Seamicro's tech is nothing to do with CPU silicon. It doesn't work for high performance servers. Calxeda's fabric does and they have shipping ARM server products; AMD does not.

- Woah. You just argued that AMD was fine in servers because ARM, but then you say ARM isn't suited for single-thread performance. So you must conclude that Intel cleans up on the low thread count servers, which is in fact most of the market (look at 2-socket vs 4-socket stats).

- One game AMD paid to have it included in is exactly my point. For a tech to have succeeded I want to see multiple vendors use it with no payment by the maker. Paying people to make stuff is the entire history of GPGPU up to now, see Nvidia and everything.

And btw no, that's not why it was a feature in DX11. It was in DX11 because IT WAS IN DX10, but Nvidia couldn't do it so MS pushed it out for them screwing over AMD.

- You're saying an extra $20 on RAM 1600 -> 1866 isn't worth a 10% graphics improvement? Because I'd sure pay for that.

- It's not DDR4 that will improve on-die graphics (look at the slides, DDR4 starts at the same performance and worse power consumption than DDR3). It's on-package memory that Intel will introduce in 2013 (Crystalwell) and that AMD isn't even close to releasing.

- QuickSync is awesome,  but it is anti-GPU compute. GPU compute says, why not use a big hot expensive vector processor to encode your videos? And QuickSync says, but if we add a tiny extra piece of dedicated silicon to the CPU it can beat the GPU's performance, do it in 1/10 the power, and have higher video quality. So isn't the implied future small dedicated SoC logic rather than GPU compute?

Programming for QS: "Encode my video"
Programming for GPUs: worse than the Cell