NVIDIA has issued an official statement on recent
ATi’s claim about NVIDIA
hardware isn’t SM3.0 compliant :

The current debate over WHQL waivers and SM3 compliance forces us to take a
public stance regarding these issues. Microsoft is the ultimate judge of
pass/fail for a particular hardware/software combination. NVIDIA hardware passes
WHQL as judged by Microsoft, and fully supports all the features of SM3. Other
hardware also passes WHQL for SM3, while not necessarily fully supporting all of
the SM3 features.

When a vendor fails to implement a required feature (such as Vertex Texture),
typically Microsoft would either fail to certify the hardware/driver, or issue a
waiver for the lack of the feature. We would not expect tests to be changed or
the interpretation of the DirectX specification to be changed.

A failure in a WHQL DCT test is by itself not an indication of hardware or
driver flaws. There can be several causes for this, including test errors, and
hardware mismatches with refrast (the Microsoft software reference rasterizer).
A mismatch with refrast can be due to the hardware producing superior, inferior,
or simply different images than refrast.

WHQL failures and errata by themselves are not necessarily an indication of
customer experience with any particular hardware/software combination. Microsoft
is the sole arbiter of WHQL passage, and further questions on this topic should
be directed to them.

20102005 ATI SLIDE NVIDIA Rebukes ATis Claim On SM 3.0

NVIDIA has issued an official statement on recent
ATi’s claim about NVIDIA
hardware isn’t SM3.0 compliant :

The current debate over WHQL waivers and SM3 compliance forces us to take a
public stance regarding these issues. Microsoft is the ultimate judge of
pass/fail for a particular hardware/software combination. NVIDIA hardware passes
WHQL as judged by Microsoft, and fully supports all the features of SM3. Other
hardware also passes WHQL for SM3, while not necessarily fully supporting all of
the SM3 features.

When a vendor fails to implement a required feature (such as Vertex Texture),
typically Microsoft would either fail to certify the hardware/driver, or issue a
waiver for the lack of the feature. We would not expect tests to be changed or
the interpretation of the DirectX specification to be changed.

A failure in a WHQL DCT test is by itself not an indication of hardware or
driver flaws. There can be several causes for this, including test errors, and
hardware mismatches with refrast (the Microsoft software reference rasterizer).
A mismatch with refrast can be due to the hardware producing superior, inferior,
or simply different images than refrast.

WHQL failures and errata by themselves are not necessarily an indication of
customer experience with any particular hardware/software combination. Microsoft
is the sole arbiter of WHQL passage, and further questions on this topic should
be directed to them.