An all-new architecture debuts within AMD’s new GPU family, marketed under the 5000-series nomenclature as a nod to their 50th Anniversary, but the majority of key info will be divulged at E3 next month.
An all-new architecture debuts within AMD’s new GPU family, marketed under the 5000-series nomenclature as a nod to their 50th Anniversary, but the majority of key info will be divulged at E3 next month.
AMD‘s keynote in Monday’s early hours saw major announcements for the consumer CPU segment, particularly the Ryzen 3000-series performance SKU’s pricing and and launch date, and workstation/datacentre products. But during Dr. Su’s presentation also tantalised with a taste of Navi, AMD’s upcoming GPU architecture which will got up against NVIDIA’s dominant RTX 20-series to win hearts, minds and performance comparisons in the latter half of this year.
While not the first piece of news revealed, the most significant for the sake of this brief discussion is that the upcoming range of GPUs will be known as the Radeon RX 5000-series. It had been widely assumed that AMD would leap on the ‘3000-series’ for GPU naming, aligning with their CPU range and gazumping NVIDIA, who already marked out their own intentions in 2018. While there will be no naming conflicts this time around (and no urgent need for NVIDIA to exercise a trademark on the 3080 filed earlier this year), it still makes you wish that both organisations would stay in their lane when it comes to naming SKUs.
At least one of the GPUs in new series will be the Radeon RX 5700. This gives very little away in terms of the rest of the series – there’s plenty of space above and below ‘5700’ to fit in other more mainstream and higher performance SKUs, but does hint that a higher performance SKU does exist. That being said, we learned very little about Navi from an architectural standpoint. Just a handful of key aspects to the design were plainly laid out:
– It retires Graphics Core Next in favour of RDNA (AKA RadeonDNA), the codename for a new series of AMD GPU microarchitectures and instruction sets which will span years of releases,
– as a consequence the Compute Unit has been redesigned from the ground up,
– its cache has been redesigned to incorporate a new cache hierarchy,
– the rendering pipeline has been streamlined,
– it will be manufactured on 7nm,
– It’s explicitly not a chiplet design,
– and it will be the first graphics card family to support PCIe 4.0.
– as a consequence the Compute Unit has been redesigned from the ground up,
– its cache has been redesigned to incorporate a new cache hierarchy,
– the rendering pipeline has been streamlined,
– it will be manufactured on 7nm,
– It’s explicitly not a chiplet design,
– and it will be the first graphics card family to support PCIe 4.0.
The first product announced to utilise Navi architecture was in fact Sony’s next-generation console, but as a semi-custom APU it wouldn’t necessarily match the full featureset and performance capabilities of a discrete Radeon Navi GPU.
Dr. Su’s comments indicated that Vega would remain as part of AMD’s product portfolio in the workstation and enterprise market for some time to come, and by implication Navi is architecturally aligned towards consumer graphics (i.e. gaming). This could mean that RDNA is a significant departure from General Purpose GPU Compute, a performance aspect with which GCN excelled. It has long been supposed that by keeping the GPU architecture oriented towards Compute subsequent GPU generations couldn’t adapt to new innovations specific to gaming, and resulted in GPUs with huge TFLOP capability that weren’t as competitive on a in-game framerate and perf/watt basis.
A brief demo compared the RX 5700 with NVIDIA’s RTX 2070 in Strange Brigade, showing frame rates for AMD’s card that were approx. 10% higher than the competition despite early drivers. However we’d caution readers against extrapolating too much from these figures, it is just one isolated test under controlled conditions. It remains to be seen whether Navi exists in a form that competes against the RTX 2080 or 2080 Ti (and indeed their own Vega-based Radeon VII).
A further demo utilising an unreleased 3DMark tool that assesses PCIe bandwidth showed RX 5000-series comprehensively beating the RTX 2080 Ti (by as much as 66%), but it’s doubtful that actual in-game frame rate gains due to higher PCIe bandwidth will come even close to one tenth of that in this generation.
If all this is a bit too surface-level for you, and you crave more of a deep dive into Navi’s design and all the upcoming graphics card models, you’ll be relieved to know that the wait for answers will not be long. On June 10th during E3 AMD will be hosting another conference and livestream dedicated to gaming technologies with a focus on Navi, and the first cards equipped with Navi GPus will be available in July.
