[ad_1]
For a few days now, on social networks such as Twitter and in some forums, even in some publications that jumped at the news, the hypothesis has been circulating that AMD marketed the first architecture-based video cards RDNA 3 – Radeon RX 7900 XTX and 7900 XT – with a Incomplete or defective GPU.
The company, consulted by Tom’s Hardware USAhas categorically denied this thesis. “Like previous generations of hardware, RDNA 3 supported shader pre-fetching such as indicated here. The code object of speculationscheck one experimental function which was not intended for inclusion in these products and will not be enabled in this generation. It is common practice in the industry to include experimental features to allow them to be explored and refined for adoption in a future generation of products,” an AMD spokesperson explained.
But how did we come to doubt the goodness of the new GPUs? All born from a tweet of the Kepler_L2 account (visible below) which, pointing to a part of the Mesa3D driver code, signaled how the Navi 31 GPU did not have a working hardware shader pre-fetch. In particular, this would have been true for the GPU with revision A0 aboard the first batch of Radeon RX 7900s.
According to the enthusiast and leaker, the resolution of the alleged problem would have required a revision of the chip, ready only in several months, condemning AMD to market thousands of substantially malfunctioning or underperforming cards. Hypothesis that seemed strange to us right away and for this reason we have not written about it, waiting for clarifications.
The issue of revision, or stepping, of the GPU is also decidedly interesting.According to Tom’s Hadware, even if the GPUs on board the new RX 7900 were indeed “A0 stepping”, it would not mean that the chip is incomplete in any part or in the way it works.
“AMD didn’t respond to our questions about whether or not to use A0 silicon for the first wave of RDNA 3 GPUs, but industry sources tell us the company uses A0 silicon. In fact, we’re told that AMD shipped cards with revision A0 silicon for nearly all of the 6000 series and most of the 5000 series. This fact is not indicative of an unfinished product. The goal of all design teams is to arrive at a ready-made, working and marketable silicon design. NVIDIA, for example, often ships cards with A0″ stepping silicon.
Chip reviews are used to fix bugs (errata) and refine other aspects throughout the commercialization period of a chip. Usually you start right from A0, then move on to A1, A2 and so on. After which, at a certain point, we move on to “B” and if necessary we continue: for Sapphire Rapids, Intel would have faced 12 steppings.
Generally, if a bug is not serious you tend to fix it by updating firmware, drivers and other software tricks. Other times, however, we proceed precisely with a hardware stepping, but it does not necessarily mean that the product will gain performance or functionality previously turned off. “The goal of any design team remains the same: to deliver silicon that, early on, meets the design goals to market the product. In that respect, using A0 silicon is considered a home run,” concludes Tom’s Hardware.
In short, for now one would say “a lot of talk about nothing”, rather it will be interesting to see – but this applies to every generation of product and for any company involved in the graphics sector – if AMD will be able to squeeze the new RDNA 3 architecture through driver updates, obtaining further performances compared to those seen at the debut.
.
[ad_2]
Source link
