Blazkowicz
Legend
DDR3 on the C2070 used to make sense but right from horse's mouth it will use gddr5
http://www.nvidia.com/object/product_tesla_C2050_C2070_us.html
http://www.nvidia.com/object/product_tesla_C2050_C2070_us.html
NVIDIA GF100 & Friends speculation
- Thread starter Arty
- Start date
But it would be interesting to know if the reason for the later launch is GPU related. If C2050 would make up for the majority of the sales and only differ in memory size, then it would hardly be GPU related to hold back the C2070. (unless they would come over 225W TDP with the added memory)
If C2070 is the more popular version, this could be seen as a sign for NV lacking suitable GPUs in numbers.
Last edited by a moderator:
I honestly can't understand why we have to waste bandwidth on a low margin market like the Teslas. But again:
Meaning both can come with frequencies of up to ~1.4GHz.
The latter sentence is for the Q3 release variant and as I said pick your poison whether you personally want it to be a chip or ram supply constraint. Given the data above the first scenario doesn't make sense to me.
http://www.nvidia.com/docs/IO/43395/BD-04983-001_v01.pdf
PSU-failure
Newcomer
Even if we consider 5 watt for an SODIMM, based on their promises to lower power consumption by 30%, that's for 16 2Gb chips.
GDDR5 presumably draws a little more power, but with just 24 2Gb chips it's probably not over 20 watt.
aaronspink
Veteran
Depending on grade and static power, you are between 2.5-3W per GDDR5 chip according to datasheets. So yes, 24 chips at full power is WELL over 20 watts.
With 24 chips and 384bit memory controler it shouldnt hapen that u will have more than 12 chips at full power at a given time no
Why do people whine about names before a product is launched? They can call is A$$ sandwich 10kXYZ if they want. I would not recommend it, but it is hardly a renaming scandal when you haven't launched or decided on a name to begin with. It actually makes sense due to the other actual renaming to put the fermi/dx11 based stuff in the 400 range. That should be a good thing from the consumers perspective.
aaronspink
Veteran
With 24 chips and 384bit memory controler it shouldnt hapen that u will have more than 12 chips at full power at a given time no
No, GDDR5 isn't multidrop. 24 chips would require a 12x2 T configuration for the CMD/ADDR and x16 data interface.
I think its more of the fact that the 8800 became the 9800 became the gts 250 and there are many other parts that got the same treatment.
Yeah, but the datasheet has IDD data for clamshell mode, and it is quite a bit lower. Not quite half more like 2/3, dropping power per chip to about 1.5W or so.
agree.
I think it also means that some die shrinks will happen to the G200 GTX lines (250, 260, 275, 285) from 55nm to 40nm. Those will likely be named 300 series.
The "die shrink" should have happened long time ago if not for the fact that the 40nm capacity was so limited and Nvidia didn't want to cannibalize their FERMI production. Since now TSM claims that the 40nm is at full production, Nvidia should be able to rebrand these old 55nm parts.
Once FERMI rolls out fully in a quarter and Nvidia gets enough salvaged parts, they will introduce GTX 460, GTS 450 (or even mixed with native ones at that time)
I think the GTX 4xx monikor also indicates they intend to keep the 4xx range for a while. So rather than rename *again* which the enthusiast community is sensitive about, they will keep the same name for the refresh product as well by adding a 5 on the end.
Ninjaprime
Regular
Can anyone translate better? I gather that the GTX 480 is about 20% faster than the 5870 and it can't beat the 5970?
Hope its more than 20% . Ati can most likely get that with a simple clock jump of 200-300mhz and faster ram most likely. Heck drivers can even close some of that gap.
Also it sounds like the 5970 is going to be much easier to find than the 480 . Though I have dual chip or sli/crossfire setups .
Also it sounds like the 5970 is going to be much easier to find than the 480 . Though I have dual chip or sli/crossfire setups .
Bear in mind that as the GF100 is a new architecture and hasn't even been released yet, nVidia is more likely to gain from driver improvements prior to its release than ATI. This is a big, big reason to not bother much with pre-release performance 'leaks': even if they're genuine, it's not a finished product just yet.
Highly unlikely. They're both chips that run on the same process, and have the same basic manufacturing constraints. The 5970's chips may be smaller, but the board has two of them.
Of course but still i wouldn't expect drivers to suddenly make the parts 50% faster. I think if its 20% faster than ati who also has an ew architecture and its first dx 11 part it will stay close to 20% through out their lives. A refresh from ati that closes that 20% gap will be possible and then even if there is more performance to find in an nvidia part your still looking at a much bigger and more expensive chip to create with more expensive pcb and more ram .
I'm going by the number of parts the link states 5-8,000 of the faster cards will be nothing. I'm sure that ati will easily be able to create more 5970s than that. The bigger problem for nvidia is if the costs are close to each other as others have claimed.
Personaly the 5870 at $300 or less is the sweet spot for me. If nvidia can offer a fermi with that performance at that price point or if the higher end fermis come out knocking the 5870 down to that price I will be very greatfull to nvidia. Though I may try and hold out till 32/28nm northern islands come out as there isn't much i want to play that taxes my video card at the moment.
You call that a simple clock jump? Its only really possible to conceive of such a clock jump if they have a significantly better process available. TSMC doesn't have anything more to give them so that leaves GloFo. I have no idea whether GloFo has anything on the way for them and even if they do, whether it will be good enough to net them a 20-30% clock jump. I don't consider an HD 5890 review edition to be a viable SKU either.
Which is why I think the report his highly inaccurate. That doesn't sound like anything remotely like a launch. I'd frankly be exceedingly surprised if nVidia launched with so few.
Personally I'm going to wait until the family is filled out, or the refresh comes. My 8800 GTX is still quite good, and I don't want something even more power hungry at the moment. Plus I've been quite annoyed with my new ATI laptop, so I think I'll stay away from ATI at least for the time being.
Ok 200-300Mhz might be a bit too much. But I'd say something around a 150Mhz increase to 1Ghz (+18%) and a ram clock around 1.45Ghz (+20%) should be doable with current chips for a roughly 20% increase. A bit more voltage, some chip binning is all that would be required. Sure might need new board layout with better vrm and TDP might increase quite a bit (requiring better cooling), but still should be massively cheaper to build than any card with a GF100. I doubt a GTX480 will be available in huge quantities neither, if nothing else price will probably prevent this...
Similar threads
- Deleted member 2197
- Architecture and Products
