The AMD RDNA3 GPUs became available for purchase a few days ago and these GPUs did face some controversy. The performance of the RX 7900 XTX, which was supposed to be an RTX 4090 competitor, was just marginally better than the RTX 4080. This raised some eyebrows in the community. Kepler on Twitter claimed that Navi31 does not have working shader prefetch hardware. Today, thanks to Tom’s Hardware, AMD has denied all such accusations.
Kepler through some code from the Mesa3D drivers claimed that the A0 silicon revision for Navi31 lacks shader pre-fetching hardware. AMD has responded by saying that the code in question refers to an experimental function included in Navi31. Team red had no plans of including this feature with the final release of Navi31. In addition, AMD states that including such features gives developers a wide variety of tuning options for better performance.
“Like previous hardware generations, shader pre-fetching is supported on RDNA 3 as per https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/gallium/drivers/radeonsi/si_state_draw.cpp#L586. The code in question controls an experimental function which was not targeted for inclusion in these products and will not be enabled in this generation of product. This is a common industry practice to include experimental features to enable exploration and tuning for deployment in a future product generation.” — AMD Spokesperson to Tom’s Hardware.
Either way, the shader pre-fetching hardware with RDNA3 works completely fine. What this means is that despite not meeting the architectural expectations of 3GHz+, Navi31 still has some room for more performance. Through a series of tweets, All The Watts highlights that the A0 and A1 silicon use the same VBIOS. Furthermore, the performance of Navi31 will improve through firmware and driver updates.
Microprocessors go through a number of revisions to improve performance/eliminate problems etcetera. The A0 silicon in our case is the first revision (for most cases). Further revisions are codenamed A1, A2 and the list goes on.
Tom’s Hardware says that AMD did not respond to their query regarding the A0 silicon. In any case, as per All The Watts, Navi31 actually uses the A1 revision but due to the same firmware, most softwares detect it as A0. What’s interesting is that even Navi21 used the A0 revision.
All things considered, AMD has to improve the performance of Navi31 at all costs, be it with new drivers or new revisions. In some games, the RX 7900 XTX can easily outpace the RTX 4090. However, over a wide range of games, the RX 7900 XTX loses its lead. This variation in performance has led to AMD receiving some fierce and harsh criticism. Tell us your views regarding RDNA3 in the comments.
Source: Tom’s Hardware