- Joined
- May 13, 2008
- Messages
- 762 (0.13/day)
| System Name | HTPC whhaaaat? |
|---|---|
| Processor | 2600k @ 4500mhz |
| Motherboard | Asus Maximus IV gene-z gen3 |
| Cooling | Noctua NH-C14 |
| Memory | Gskill Ripjaw 2x4gb |
| Video Card(s) | EVGA 1080 FTW @ 2037/11016 |
| Storage | 2x512GB MX100/1x Agility 3 128gb ssds, Seagate 3TB HDD |
| Display(s) | Vizio P 65'' 4k tv |
| Case | Lian Li pc-c50b |
| Audio Device(s) | Denon 3311 |
| Power Supply | Corsair 620HX |
Except all indications seem to point to late q3 to q4, regardless of what tsmc says. Be that a year from the launch of Hawaii, six months or so from this onslaught of refresh parts, the fact amd publicly stated they were taping out chips this quarter, that nvidia said the rest of maxwell was coming later this year, and that the public cost per transistor maps (that has been floating around for years) all pretty much point to exactly that. To be literal to their respective histories on the ramp/price-per-xtor it would suggest late q3 for amd and late q4 for nvidia, but both may end up around the same time.
On top of that, what you're asking for doesn't make sense. Maxwell is clearly based around 6smm and 2mb cache blocks. 750ti is essentially half of what a 8800gt was in a lower market...1 less smm released earlier on a larger process to compensate for the power increase (20% in power from logic + whatever clockspeed increases 20nm gives with such more logic in the same power envelope; probably around 1200mhz to 750ti's ~1150 for a complete difference of around 20-25%) while simultaneously working as class-leading <75w part. What you're asking for is essentially a part that is less than 2x gm107, which really doesn't make sense on 28nm short of something really weird like a low-clocked 9smm part with 24 ROPs/3mb cache on a 192-bit bus, which I doubt we see.
On the flipside, what do you expect of a higher-up part on 28nm? If you're expecting die savings compared to gk104, I think you will be sadly mistaken. The arch again is catered towards the higher logic use of 20nm (ex: the extra cache) while using less power for similar to slightly higher core clocks (rather than scaling to higher clocks with more voltage and less logic like most processes before it) and conceivably smaller/slower mem controllers to keep power down, which will become more-so apparent as they move to larger designs. In short, a 12smm part (the only one that makes sense versus gk104) would be faster per clock than gk104, probably larger (depending on cache + more units vs mem controller size offset), but proportionally use less power because of design efficiency/slower memory speed/bus. While I suppose it's possible they could keep such a design under 225w on 28nm, and it would be good for us, it would probably be bad for nvidia's bottom line (a wholly new design for what amounts to a small improvement).
That all said, when it comes to this 280...the price is indeed a shame compared to old product prices, but also reflective of the situation: the yields are good and the cost savings are already passed on to 280x (at least at msrp). The price difference is pretty much directly proportional to real avg shader perf per clock and memory speed differences....meaning at least these will probably clock decently and similar to 280x (as opposed to 7950 which was power/clock limited) so it should be able to wiggle a decent spot between 760 and 770 if priced accordingly.
Last edited:
that is always the trend and it will stay like that. no matter green or red cards.