While CPU efficiency sometimes grows when it comes to proportion factors with every successive technology, GPU efficiency will increase have a tendency to return within the type of whole-number multipliers. There’s a number of causes for this, however the largest one is solely that GPU workloads are a massively-parallel activity, and it is simple to speed up them by merely piling on extra useful models in your parallel processor.
Responding to a tweet from over every week in the past, frequent GPU leakster kopite7kimi said that the efficiency of NVIDIA’s Ada Lovelace AD104 processor is roughly equal to that of its extant Ampere GA102 processor. To put that in perspective for you, GA102 is the processor used within the GeForce RTX 3080, RTX 3080 Ti, RTX 3090, and RTX 3090 Ti. It is the biggest Ampere chip that will get used for GeForce merchandise.
Meanwhile, AD104 is the third-largest Ada Lovelace processor, behind AD103 and AD102. Going by the leaked data we have seen beforehand, it is going to be the GPU that powers the GeForce RTX 4070 and its related merchandise. WIth that in thoughts, kopite7kimi is actually implying that the GeForce RTX 4070 (or maybe an RTX 4070 Ti or such) shall be as quick as a card based mostly on GA102.
We actually do not know what particular graphics playing cards kopite7kimi is evaluating, although. As famous above, these GPUs get used (in additional and additional cut-down kinds) for a wide range of graphics card SKUs. It’s potential that the fully-enabled AD104 solely matches as much as the unique GeForce RTX 3080 10GB card. That’s nonetheless based mostly on GA102, so we’re again to my earlier assertion concerning the breadth of efficiency on this single technology of GPUs.
Unfortunately, on condition that NVIDIA and its companions appear detest to launch a brand new mannequin of GPU any time quickly, we’ll most likely have not less than 4 or 5 months to attend earlier than we are able to lay fingers on Lady Lovelace and discern the actual story. Of course, we’ll let our great readers know as quickly as we do.