ads

Sunday, March 20, 2016

Specifications of future NVIDIA Pascal? Sorry but it's all a fake

In less than one month NVIDIA will hold its regular GPU Technology Conference, or GTC, an event organized to meet the industry engaged in the development of solutions for the GPU market with a focus on the professional context. Here it is likely that NVIDIA will perform public demonstrations of the first next-generation architecture based cards Pascal, the market expected in the second half of the year.

No wonder that appear online rumors about what will be the technical specifications of the upcoming NVIDIA GPU, as at present lacking solid confirmations and of course any kind of official information from NVIDIA without charge.

Yesterday appeared online an image containing what should be the next generation NVIDIA video card technical specifications of 3 architecture-based Pascal, intended for the most powerful desktop systems. We refer to the models listed with the names GeForce X80, X80 and GeForce GeForce X80Ti Titan. The document is marked as Confidential NVIDIA already but just to observe the way in which the GeForce brand has been written (Geforce, with tiny F) to understand that it is a fake written from scratch by someone with fictional data. Here below the slide, published by Guru3D site, using it to do some 'teaching that helps to understand why this information is false simply by using common sense.


It is indicated two different GPU the Pascal family, GP104 and GP100 models, and this logic seems to be the only information that may have some truth. Moreover, even here just a little 'common sense: GP104 is the GPU that will take over from GM204, the Maxwell family model used in the GeForce GTX 980 cards, while GP100 has as correspondent between Maxwell cards the GM200 model used in GeForce GTX 980Ti and GeForce GTX Titan-X.

Everything else is invented out of whole cloth, with some apparent inconsistencies from as indicated by the memory technology. For the Titan top card of the range it is reported using HBM2 memory, while with the same GPU for GeForce x80Ti the choice fell on GDDR5 memory. HBM and GDDR5 memory technologies are very different between them and it is not conceivable, for matters of cost and complexity, that the same chip GP100 intact to the own internal memory controller, or two separate memory controller, compatible with standard GDDR5 and with that HBM2 .

That said, if GP100 complement to the own internal controller GDDR5 this will be compatible with the GDDR5X standard, able to ensure a doubling of the bandwidth at the same clock frequency. Again it is possible that NVIDIA can not propose top video cards with GDDR5 memory range when it will be available GDDR5X solutions that precisely with the most powerful GPUs are the perfect match.

Needless therefore talk about the number stream processors, or cores CUDA, GPU integrated in each, the number of texture units or of ROPS or even the amplitude of the combined memory controller. The table shows invented data, without even special care, and it is good to forget using this news as an aid to understand when an online information is appropriate and when it is based on nothing. We will see in a few weeks if the NVIDIA GTC will anticipate some of the information on the first GPU of the Pascal family; our opinion is that the public demo will be performed but no data Technical trapelerà, at least officially.

No comments:

Post a Comment

Apple Vision Pro: Day One

It’s Friday, February 2, 2024. Today is the day. You’ve been eyeing the Vision Pro since Tim Cook stepped onstage with the product at last y...