NVIDIA GF100 & Friends speculation
- Thread starter Arty
- Start date
FenderBender
Newcomer
If the GTX480 has only 480 cores and not 512, how are the unused cores disabled? By frying on-chip fuses after packaging, by laser cutting of chip surface metal before packaging, by firmware, by board level changes?
When a company yield-manages its chips like this, but doesn't release a top-end "all cores" part, what happens to those chips that actually do have a full working complement of cores? Are they saved for later potential parts (GTX485?) or are they just hobbled and sent off anyway as lower end parts? I'm sure it depends on the financials and technology scheduling and forecast. but has there been a similar case of a GPU never released at full-core in the past?
I know IBM Cell had many similar problems (which is why the PS3 doesn't use all of the SPEs on chip.) Am I right that after so many years of process optimization, the PS3 Cell chips are likely super-low defect and therefore every chip has an SPE completely wasted?
When a company yield-manages its chips like this, but doesn't release a top-end "all cores" part, what happens to those chips that actually do have a full working complement of cores? Are they saved for later potential parts (GTX485?) or are they just hobbled and sent off anyway as lower end parts? I'm sure it depends on the financials and technology scheduling and forecast. but has there been a similar case of a GPU never released at full-core in the past?
I know IBM Cell had many similar problems (which is why the PS3 doesn't use all of the SPEs on chip.) Am I right that after so many years of process optimization, the PS3 Cell chips are likely super-low defect and therefore every chip has an SPE completely wasted?
AFAIK, once upon a time, laser cutting was used to trim ultra-precise (super expensive) opamps. I haven't heard it ever being used to disable digital features.
Fuses are most likely.
Sorry for this off-topic..
After so many years IBM has ported cell to a newer manufacturing process a couple of times, so they propably are not super-low defect, but just much smaller and low-defect.
After so many years IBM has ported cell to a newer manufacturing process a couple of times, so they propably are not super-low defect, but just much smaller and low-defect.
I will fall over laughing if the launch is delayed and turns out to be April 6th, I truly will. 
It was already mentioned - some Quadros and Teslas might have all cores enabled, hence there's probably no fully working chips left anyway. (Normally, you'd think Quadros and Teslas are tiny quantities compared to the normal ones, but if the sub-10k cards is true, then that's probably not the case.)
I'm still wondering though what effect this has on performance if any due to the asymmetric partitioning of the chip (4gpc with 4x32 cores in full configuration) and how that's going to work.
I was never expecting reviews on the 26th, sometime later that week, the cards were to go out to reviewers on the 26th.
For something like this you may also find that, because you are putting resource into making the new chip anyway, you may also choose to put a little engineering and layout resource into shuffling off that additional SPU from the smaller designs (if it is only targetted to the PS3 configuration). It all becomes the tradeoff at what point you waste more die area from the redundancy vs loosing full die. This only really becomes a discussion / issue with you have a longer process shrink roadmap for the part (which consoles often do).
No, when did Nvidia state there would be no reviews on the 26th? See Razor's post above. That's a new revelation AFAIK.
Bouncing Zabaglione Bros.
Legend
If there's only a few of those, Nvidia is obviously going to save them for the $3000+ professional market. Even in tiny quantities, the increased margin makes it much more worthwhile than fusing down a fully working chip and selling it as a consumer GTX480.
Even if it just gives a few key guys a head start on HPC installation work for when B1 gets taped out, it's probably worth it.
In the general "nvidia is always right and never lies" kinda hymn that some seem to chant, you could interpret it as "you expect a launch date? you'll know more on March 27th"
Let's just say that the press was told directly by nvidia that they would get their cards well before launch (i.e. two weeks, which is now one week.)
Kind of Ironic how this whole "new media" thing failed as they haven't given us a new "Fermi fun fact" since February 22nd (insert feigned outrage here!)
Last edited by a moderator:
This is from the facebook account:
And from Baxtor:
Andrew Fear at GDC: http://nvidia.fullviewmedia.com/gdc2010/02-andrew-fear.html (16:40)
Last edited by a moderator:
"I guess you respecting something like NDA and launch dates right ? Sure, soon on March 27th in Europe."
You think Igor wrote the box text for the iGame GTX480?
PS. I assume he's trying to say NDAs runs out and the cards will launch on the 27th in Europe (twittering rots the brain).
You think Igor wrote the box text for the iGame GTX480?
PS. I assume he's trying to say NDAs runs out and the cards will launch on the 27th in Europe (twittering rots the brain).
Wait 'till you get your greedy hands on the review guide.. we'll start by saying that you can see the difference in objective and subjective reviews without opening your browser tabs. They will use everything they can to leverage the performance "difference" in PhysX games and we might(is this a hint?) even see them asking reviewers to please benchmark their own demos on AMD hardware (!)
Similar threads
- Deleted member 2197
- Architecture and Products
