The result is an overpriced, underperforming POS that sucks down so much power it's an embarrassment to both AMD and TSMC. 7nm delivers more speed at the same power draw, or uses less power at the same clockspeed. AMD seem to have taken that and delivered a zero-new-features Vega64 with 1/16th of the die disabled yet consumes more power than a Vega64 and barely manages to run faster than factory-overclocked versions.
Funny, mine seems to be pretty competitive with the 2080 and is even better under water. A Vega56/64 isn't even close.
Yes, the VII matches a 2080 under certain conditions, but for god's sake don't pretend that it's competitive across the board!
When running a non-RTX, non-DLSS, DX12-specific game engine, at 4K or higher resolutions, such that the additional HMB2 bandwidth is an advantage
, a Radeon VII matches a 2080. Unless the game is an Nvidia TWIMTBP title using black-box Nvidia code that cripples AMD performance.
In a more professional test across 21 currently popular modern titles performed by an independent review site, The VII is between a 2070 and 2080 in performance, rarely matching or beating a 2080. In TWIMTBP DX11 titles it can barely compete with a 2070, and although I find the TWIMTBP program abhorrent, a large proportion of games use it, so it's very relevant to the discussion:https://www.techpowerup.com/reviews/AMD/Radeon_VII/28.htmlAt best, a Radeon VII matches a 2080 at 1080p, marginally winning at 4K Ultra. These scenarios are few and far between, and in one case (BF V) is unable to run with RTX or DLSS.
At worst, a Radeon VII struggles to come close to even a stock-clocked 2070, losing in several tests to even the GTX 2060 and 1070!
In all cases, there are two question marks hanging over the RX Fury
, uh, the RX Vega 64 minus 4
, I mean the Radeon Instinct MI50
- Nvidia's TWIMTBP program sways a much larger number of titles in Nvidia's favour. Nvidia simply dumps more money, time, and effort into this than AMD.
- The VII has no ray-tracing, no DLSS, no CUDA support, and inferior MRS/VRS support. None of those things are essential but at $699 for a card, I don't want to be short-changed on features; DLSS and VRS have the potential to be game-changers as screen resolutions continue to climb.
Vega 56 is capable of outperforming Vega 64 stock, and with water can be pushed another 10% higher still. The VII is potentially
much faster than even that, because it has more memory bandwidth, but a plethora of independent reviews put it at 15% faster than a Vega 64 when averaged across a large sample size of games and tests. Of course you can always cherry-pick a specific test or scenario where the VII does really well, but that's only relevant if your usage exactly matches that one specific game or workload. The rest of us buy a GPU to run everything as well as possible - including as of yet unrealeased titles on engines yet to be developed. Either way, If you're prepared to have a large, noisy, power-hungry card in your PC, you will find that an overclocked Vega 56 gets you most of the way - maybe 90% of the way - towards a Radeon VII for for just one-third of the price. That's a real kick in the balls for the VII, IMO, which is why I called it stupid.
Radeon VII is a rebadged workstation compute card with workstation compute RAM that AMD hastily wrote a gaming driver for because they had nothing new to show as a response to the RTX launch. That is all there is to it. If you watch launch-day reviews of the thing you can see that all of the highest-profile reviewers had issues with the rushed drivers, lockups, or inconsistent performance on the Radeon VII: Jayzee 2c, Linus (LTT), Steve Burke (GamersNexus) Richard Leadbetter (DigitalFoundry) all saying that it's a rush job that brings nothing new to the table and does not force Nvidia's hand WRT the insane price hikes of the RTX series over previous generations.