PSU-failure
Newcomer
True...
... without any architectural overhaul.
Considering the move to VLIW4 and the transistor/power budgets, AMD have probably improved the arch elsewhere.
AMD: R9xx Speculation
- Thread starter Lukfi
- Start date
... without any architectural overhaul.
Considering the move to VLIW4 and the transistor/power budgets, AMD have probably improved the arch elsewhere.
DavidGraham
Veteran
We've been over this before , RV670 is a special case , ATI manufacturing tools and experience were not as efficient at that time , thus RV670 came with a bigger die than needed ,then ATI gained more experience and improved it's tools , and made RV770 at the right size .
aaand, that is:
Tell me this is not key for power-regulation :S
How about dual BIOS Switch.
Like this one.
aaand, that is:
Tell me this is not key for power-regulation :S
No it isn't,
hint: http://www.bit-tech.net/hardware/graphics/2010/09/17/msi-n480gtx-lightning-preview/2
@no-X and DavidGraham:
I was also referring to the different layout of the SIMD engines starting with RV770. The TMUs went into the engines, which is a fairly big change. It keeps the texture data much more local and enables a far easier scaling to a higher SIMD engine count. And it does so while reducing the necessary die size for it (because you don't need to route the texel data across half the chip, the individual TEX-L1s can have less bandwidth than the unified one for RV670). The R600/RV670 design wouldn't have scaled easily to 10 SIMDs and not within 256 mm².
And the argument, that RV770 used a one time gain from improved layout tools is basically at the same level as Cayman can use a one time gain from the VLIW5 => VLIW4 change, which is supposed to increase the area efficiency by ~10%. And RV770 actually increased the transistor density less than that (only ~7.5%)
I was also referring to the different layout of the SIMD engines starting with RV770. The TMUs went into the engines, which is a fairly big change. It keeps the texture data much more local and enables a far easier scaling to a higher SIMD engine count. And it does so while reducing the necessary die size for it (because you don't need to route the texel data across half the chip, the individual TEX-L1s can have less bandwidth than the unified one for RV670). The R600/RV670 design wouldn't have scaled easily to 10 SIMDs and not within 256 mm².
And the argument, that RV770 used a one time gain from improved layout tools is basically at the same level as Cayman can use a one time gain from the VLIW5 => VLIW4 change, which is supposed to increase the area efficiency by ~10%. And RV770 actually increased the transistor density less than that (only ~7.5%)
NathansFortune
Regular
Indeed. Looks like we're in for the long haul now.
This is the biggest piece of luck Nvidia ever got, it gives them a massive reprieve and a lot of time to perfect Fermi beyond the GF100b. I fully expect the GTX680 to come in August of next year on the 40nm process node and be a massively optimised with high clocks, better density, better thermals and all round better performer. It also gives Nv time to do a decent dual chip card on 40nm something they would not otherwise be able to do.
there's more switches there.
the over(and under)-voltage part is regulated by AMD's overdrive
A1xLLcqAgt0qc2RyMz0y
Veteran
OK what am I missing here!!!Indeed. Looks like we're in for the long haul now.
This is the biggest piece of luck Nvidia ever got, it gives them a massive reprieve and a lot of time to perfect Fermi beyond the GF100b. I fully expect the GTX680 to come in August of next year on the 40nm process node and be a massively optimised with high clocks, better density, better thermals and all round better performer. It also gives Nv time to do a decent dual chip card on 40nm something they would not otherwise be able to do.
What has happened that gives Nvidia a massive reprieve?
ditto me on this query....what gives?
A1xLLcqAgt0qc2RyMz0y
Veteran
Radeon HD 6970 runs at 880MHz and has only 1536 ALU streaming processors not the 1920 that has been widely quoted in this forum.
True or False?
http://www.fudzilla.com/graphics/item/21155-radeon-hd-6970-runs-at-880mhz
True or False?
http://www.fudzilla.com/graphics/item/21155-radeon-hd-6970-runs-at-880mhz
Not false, Fuad conveniently leaves out the low TDP at which this is achieved, let's say it's not even close to the 300W the NV propaganda machine promoted.
Talk to me like I'm a third grader please...based on what new information?
DavidGraham
Veteran
While true , we still don't know the validity of that claim , whether it applies to specific work loads or not , or whether the change have introduced other inefficiencies elsewhere ..And the argument, that RV770 used a one time gain from improved layout tools is basically at the same level as Cayman can use a one time gain from the VLIW5 => VLIW4 change, which is supposed to increase the area efficiency by ~10%. And RV770 actually increased the transistor density less than that (only ~7.5%)
A1xLLcqAgt0qc2RyMz0y
Veteran
Performance is what counts at the high end and if the 6970 isn't up to the performance level of the GTX570 who cares that it burns less watts.
DavidGraham
Veteran
You could have at least had the courtesy to choose 15.3 and 890/100 in my riddle !
Anyhow even with only 1536 ALUs , HD 6970 could very well match or slightly beat GTX 580 .
Who said, that HD6970 isn't able to outperform GTX570?
Well Neliz, ATI own slides stated it was >225W (8pin+6pin). While it might not be close to 300W, its not also far from GTX580 244W does it? To keep the same perf/watt of Cypress, they really have to perform great then!
Similar threads
