Rumors Claim AMD Ships Unfinished 3 Navi 31 A0 RDNA GPU Silicon in Radeon RX 7900 Series

AMD recently launched its flagship RDNA 3 offerings based on Navi 31 GPUs, the Radeon RX 7900 series. While AMD pitted the Navi 31 against NVIDIA’s RTX 4080, the final performance was slightly disappointing and wasn’t the best. huge disruption that everyone expected. Now rumors suggest that the reason behind such lackluster performance may be due to an unfinished GPU silicon shipped on flagship cards.

AMD allegedly shipped unfinished RDNA 3 ‘Navi 31’ GPU Silicon with Radeon RX 7900 series graphics cards

As discovered by Kepler_L2, it seems that the early RDNA 3 silicon had a broken shader prefetch HW. This was featured in three chips, GFX1100 (Navi 31), GFX1102 (Navi 33) and GFX1103 (the APU range consisting of Phoenix chips). Now based on the latest GitHub posting. Kepler also suggests that Navi 32 GPUs are actually based on the “GFX1103” IP, while Navi 33 chips feature the “GFX1102” IP. The same issues are seen on all other chips besides Navi 32, which will be featured in several mainstream discrete GPUs across desktop and mobile platforms by early 2023.

According to Kepler, this is something that can’t be fixed or revised in a few weeks and will take several months if AMD is going to fix it. It would be a big blow if the new silicon with a fix came out a few months later because most gamers would have already bought the first revision of the Radeon RX 7900 series with the unfinished silicon. What’s likely is that AMD may prepare an update a year later to address these issues, but for now, the Radeon RX 7900 series may have to depend heavily on driver-level optimizations to address the unfinished GPU nature of their best. silicon RDNA 3.

But that’s not all, other major features like the VOPD instructions present on RDNA 3 GPUs claimed to offer a big performance boost, but in reality only managed to deliver a 4% improvement over RDNA 2 in ray tracing titles. The RDNA 3 silicon was designed to handle two Wave32 instructions for twice the floating point performance using 64 multi-precision, multi-purpose ALUs implemented on 2 SIMD32 units. Speaking with an AMD rep, the folks over at HardwareTimes were able to confirm that AMD is currently fine-tuning performance here and expects optimizations along the way:

Wave64 can natively access the new ALUs for 2x execution speed to unlock performance when running dense ALU code. For Wave32 mode, the compiler does localized reordering and instruction packing in VOPD encoding. An RT test scene using VOPD encodings delivered about a 4% increase in frames per second while removing the ALU bottleneck. We expect to see more improvements as the compiler matures with more optimizations for mapping code sequences to VOPD encodings. And with advances in the use of AI, RT, and compute-based rendering techniques for more realistic rendering, we expect to see ALU-constrained code making more and more use of these new ALUs.

AMD Representative via HardwareTimes

Reviewers who also managed to get a first look at various custom AMD Radeon RX 7900 series graphics cards also reported inconsistent and fluctuating clock speeds on the cards. AMD also promised up to +54% improvement in power efficiency, but it was nowhere to be seen, and NVIDIA’s Ada Lovelace architecture blew away RDNA 3 GPUs in all power efficiency metrics on review day.

Just a few days ago, NVIDIA threw shade at AMD and Intel for using their consumers as guinea pigs to BETA test their graphics drivers for them. Now AMD shipping an unfinished silicon also looks like a beta test for its RDNA 3 GPUs that are being shipped to gamers around the world.

Below is a breakdown of all AMD GPU architectures released by Vega:

GRAPHIC ARCHITECTURE GPU CODE NAME ALTERNATIVE NAME ARCHITECTURE PRODUCT
GFX900 VEGA 10 Google Display Network 5.0 RX VEGA / Radeon Pro
GFX902 CROW The Crow’s Crest / Picasso Google Display Network 5.0 Ryzen 2000/3000(G/GE)
GFX904 VEGA 12 Google Display Network 5.0 Vega Pro 20 (MAC)
GFX906 VEGA 20 Google Display Network 5.0 Radeon VII / Radeon Pro VII
GFX908 ARTHUR CDNA 1 Instinct MI100
GFX90A Aldebaran CDNA 2 Instinct MI200
GFX909 CROW2 Google Display Network 5.0 TB
GFX909 RENOIR Google Display Network 5.0 Ryzen 4000 (H/U/G)
GFX1010 SHIPS 10 RDNA 1 RX5700/5600(M/XT)
GFX1011 SHIPS 12 RDNA 1 PRO 5600M (MAC)
GFX1012 SHIPS 14 RDNA 1 RX 5500 (M/XT)
GFX1030 SHIPS 21 Big Ships / Sienna Cichlid RDNA 2 RX 6900/6800 series
GFX1031 SHIPS 22 Sea flounder RDNA 2 RX 6700 series
GFX1032 SHIPS 23 Dimgrey Cavefish RDNA 2 RX 6600 series
GFX1033 SHIPS 24 Beige goby RDNA 2 RX 6500/6400 series
GFX1033 Van Gogh Aerith RDNA 2 Steam bridge
GFX1036 Raffaello RDNA 2 Ryzen 7000?
GFX1040 VAN GOGH LITE / Mendocino Green sardine RDNA 2 Ryzen 7000?
GFX1100 Ships 31 Plum bonito RDNA 3 Radeon RX 7900
GFX1101 Ships 32 Wheat nose RDNA 3 Radeon RX 7800
GFX1102 Ships 33 Bright pink bonefish RDNA 3 Radeon RX 7700?
GFX1103 Ships 3X Phoenix Point RDNA 3 Ryzen 7000 APU?
GFX1200 Navi4X RDNA 4 Radeon RX 8000?

The days ahead for RDNA 3 and especially the Navi 31 GPU silicon based Radeon RX 7900 series graphics cards will be interesting to see. We’ve been told by extreme overclockers that they can’t stabilize their cards at higher clock speeds as the silicon is severely limited by power limitations and it is utterly meaningless to run LN2 OCs with the current BIOS.

What do you think of the launch of AMD’s Radeon RX 7900 series?

Share this story

Facebook

Chirping

Leave a Reply

Your email address will not be published. Required fields are marked *