This site may earn affiliate commissions from the links on this folio. Terms of use.

Radeon-Vega

A piffling over two years agone, AMD announced that it would restructure its high-end GPU development squad. This new unit, dubbed the Radeon Technologies Group, would be led by senior vice president Raja Koduri. While RTG was never an independent spin-off or subsidiary, information technology enjoyed a off-white degree of autonomy and concentrated AMD driver evolution, programmer relations, GPU design, and GPU architectures all nether the aforementioned unit of measurement. This was a deviation from the previous system, in which AMD has substantially unified its APU, CPU, and GPU teams. RTG, we were told, would exist the vehicle AMD needed to deliver a best-in-class part that could compete with annihilation Nvidia had to offer. Now, Raja has announced he'south leaving AMD (he had been on sabbatical for the past 40 days).

RajaAMD

Hexus obtained a copy of the memo Koduri sent to other members of the RTG group. In information technology, he thank you his co-workers for their difficult work and dedication to a difficult two-yr period and specifically mentions both Lisa Su and Mark Papermaster for the trust they extended to him in forming and backing RTG in the first place.

An Unsurprising Divergence

On the 1 mitt, this news is no surprise; we've heard rumors Raja might be leaving for several weeks at present. But the other reason information technology's not surprising is because both Polaris and Vega have struggled to friction match Nvidia's ain GPUs. Now, keep in mind, the lead time on GPU designs is quite long. AMD was working on Polaris long before they announced it in late 2022; Vega's design would've probable been underway by then as well. AMD's RX 400 series was solid and reasonably well positioned, if non a knockout, but the RX 580 only managed to increase its operation by burning a great deal more power. Boosting the RX 580'south clocks brought it more in line with the GTX 1060, simply ability efficiency remained elusive.

Vega 56 and Vega 64 take a similar problem. By the time the cards were ready for market, Nvidia's Pascal had been out for 15 months. In ordinary circumstances, one might reasonably expect AMD to (pardon the pun) piledrive Nvidia. Just that's non what happened. Instead, the Vega 56 established itself equally a somewhat faster GPU compared with the GTX 1070, while Vega 64 lags a bit behind the GTX 1080.

99thfps (1)

Data by Tech Written report

This is certainly improve than having no high-cease GPU at all, and Raja deserves credit for delivering a function that does compete reasonably well, given that a price drop is all AMD would really need to sell a much better operation-per-dollar ratio than Nvidia's college-end cards. AMD took a beating for years over its erratic frame timing compared with Nvidia, but Vega reviews accept shown that this problem is largely in the past. In some cases, AMD cards are really smoother. Simply once again, expectations were extremely high for Vega, and the GPU didn't deliver the knockout dial that investors and enthusiasts expected.

We've long speculated that HBM2's difficult ramp may have played a part in this, but ultimately it doesn't affair. Vega may be a contender, but it wasn't a knock-out. It's an architecture AMD tin can build on (and is building on, given the recent deal with Intel), merely it didn't vault AMD back into the commuter'southward seat.