NVIDIA Fermi: Architecture discussion

You keep saying this but what's the better alternative you have in mind?
Like I said earlier, alternatives need not be better.

Your original 2.9 number was correct :D
Not if you use 5900. :D

Sure, but where is Cypress ECC support ? Where is AMD's tool for Visual Studio based development on the GPU ?

They simply do not exist and are features directly aimed at the HPC market.
You are making it sound like those features are the only ones that made OR sign up for Fermi.

You guys are making too big of deal of this OR fallout. We do not know what were OR's internal processes that led to sign up and cancellations of the project.
 
You are making it sound like those features are the only ones that made OR sign up for Fermi.

You guys are making too big of deal of this OR fallout. We do not know what were OR's internal processes that led to sign up and cancellations of the project.

No big deal at all. You are the one assuming that that "article" is even true :)

As for the features I mentioned, they most likely aren't the sole reasons for the OR deal, yes, but they definitely are part of the set.
 
Fun Fact of the Week: GF100 graphics cards will provide hardware support for GPU overvoltaging for extreme overclocking!

*yawn* but I guess they had no choice but to follow AMD to avoid losing their street cred.
 
Fun Fact of the Week: GF100 graphics cards will provide hardware support for GPU overvoltaging for extreme overclocking!

*yawn* but I guess they had no choice but to follow AMD to avoid losing their street cred.
Or clocking them slower to stay within PCI-E spec.
 
You are comparing a card to a 4 board enclosure. It isn't unreasonable that the enclosure uses the extra 150W or so for multiple PCIe switches, fans, blinkenlights, and other things, not to mention PSU efficiencies if read at the wall.

-Charlie

Yeah, I found the max power figure:

 
Or clocking them slower to stay within PCI-E spec.
Screw it, I doubt any buyers except corporate care about that sticker ... and they can just buy the versions which do stay within the limits.

Neither the consumers nor the system builders who use these devices need hand holding by PCI-SIG in designing a system which can handle the load. Screwing about on the economical limit of PSUs and cooling is not important for the machines in which these cards go ... you can afford to just overdimension it a bit.
 
Last edited by a moderator:
I'd guess OEM versions would disable overvolting with their card's BIOS by default.
That way they are covered if someone decides to fry the card, MB or PSU.
 
I'd guess OEM versions would disable overvolting with their card's BIOS by default.
That way they are covered if someone decides to fry the card, MB or PSU.

Why would they? They don't do so for any other component with user adjustable voltages - RAM, chipset, CPU etc. In any case isn't this the big hyped feature of the 5970?
 
Screw it, I doubt any buyers except corporate care about that sticker ... and they can just buy the versions which do stay within the limits.

Neither the consumers nor the system builders who use these devices need hand holding by PCI-SIG in designing a system which can handle the load. Screwing about on the economical limit of PSUs and cooling is not important for the machines in which these cards go ... you can afford to just overdimension it a bit.

I would imagine that since PCI-SIG controls the specifications for PCI-E that they would have some say-so regarding the licensing of PCI-E devices.
 
Fun Fact of the Week: GF100 graphics cards will provide hardware support for GPU overvoltaging for extreme overclocking!

*yawn* but I guess they had no choice but to follow AMD to avoid losing their street cred.



Ginger Plastic Fermi is on pricey pre-order !
 
Creig, I don't think so ... from the bylaws ruling under which condition the cross-licensing deals work :
ARTICLE 15. INTELLECTUAL PROPERTY LICENSING POLICY

SECTION 15.1 DEFINITIONS

The following definitions shall apply to this Article 15:

(a) "Compliant Portion" means only those specific portions of products (hardware, software or combinations thereof) that: (i) implement and are compliant with all relevant portions of a Specification, and (ii) are within the bounds of the Scope.

<snip>

(d) "Scope" means the protocols, electrical signaling characteristics, mechanical requirements for connectors, cards and cabling, register models, data structures and verbs software interface solely to the extent disclosed with particularity in a Specification where the sole purpose of such disclosure is to enable products to interoperate, interconnect or communicate as defined within a Specification. Notwithstanding the foregoing, the Scope shall not include (i) any enabling technologies that may be necessary to make or use any product or portion thereof that complies with a Specification, but are not themselves expressly set forth in a Specification (e.g., semiconductor manufacturing technology, compiler technology, object oriented technology, basic operating system technology); (ii) the implementation of other published specifications not developed by or for the Corporation but referred to in the body of a Specification; or (iii) application programming interfaces, device drivers, device driver models, peripheral control languages, and peripheral devices, except for the portions of peripheral devices that are required by an interconnect that is compliant with a Specification.

<snip>

When the Member or its Affiliate makes a Contribution to a Specification of the Corporation, including revisions thereto, or when the Corporation adopts and approves for release a Specification after providing notice as set forth in Section 15.2, above, the Member and its Affiliates hereby agree to grant to other Members and their Affiliates under reasonable terms and conditions that are demonstrably free of any unfair discrimination, a nonexclusive, nontransferable, worldwide license under its Necessary Claims to allow such Members to make, have made, use, import, offer to sell, lease and sell and otherwise distribute Compliant Portions, provided that such agreement to license shall not extend to any part or function of a product in which a Compliant Portion is incorporated that is not itself part of the Compliant Portion. Each Member agrees that they will not transfer, and have not transferred, patents having Necessary Claims for the purpose of circumventing this Section 15.3.
Power use of a product outside of the "compliant portion" is completely irrelevant to the patent license AFAICS. Only through the trademark licensing can they force any other kind of compliance.
 
Last edited by a moderator:
Why would they? They don't do so for any other component with user adjustable voltages - RAM, chipset, CPU etc. In any case isn't this the big hyped feature of the 5970?
I'm talking about a OEM system. Most have a very limited user feature BIOS for changing anything.

The OEM user might be able to flash the card's BIOS to accept voltage adjustment, but I doubt it would be as simple as download the tweak software and go.
 
How is that a power problem? Sticking CPUs in there instead will still be worse overall for a given performance target. Basically you're saying that they cancelled it because instead of being X times more efficient than CPUs it's only Y times more efficient. Oh noes!!!

See the problem with that theory?

No, not really, who said they were going to stick CPUs in there? I don't know where it goes from here.

-Charlie
 
Running a Cypress system would be an interesting experiment. I'd wonder if even a laboratory's HPC programmers would have the stomach to hack the assembly to get decent compilation.

The payoffs are potentially very good, though. Cypress can do a little over 2 TFLOPs single-precision matrix multiply and 500 GFLOPs DP with a bit of hacking.
 
Back
Top