AMD’s RX Vega, the gaming version of AMD’s long-awaited high-end GPU architecture, finally saw its release this month. While RX Vega 56, the smaller brother of the full Vega GPU, offered great price-to-performance compared to NVIDIA’s price-equivalent GTX 1070, the story is different with RX Vega 64 when comparing it to its price-equivalent, the GTX 1080.
Pricing
In a previous article comparing the Vega 56 against the GTX 1070 and GTX 1080, no official statement from AMD was available regarding pricing of Vega GPUs. Now, AMD has stated multiple times that the company is doing its best to keep the RX Vega cards at the suggested retail price originally advertised: $399 for the RX 56 and $499 for the RX 64.
However, with the launch of RX Vega 56, the pricing problem has been reinforced even further, with close to no cards being available at the suggested prices. The issue cannot be blamed solely on AMD, as they can only suggest the prices retailers sell the cards at, but manufacturing partners (ASUS, Gigabyte, MSI, etc) and major distributors can easily set their own prices because they know there is high demand and the cards will sell either way.
The Radeon Packs that were introduced with the launch of RX Vega only make matters more complicated, which is why there is still no concrete conclusion to the pricing situation with RX Vega cards. Due to the fact that nobody knows when the prices will stabilize, we will compare the RX Vega 64 to both the GTX 1080 ($499 MSRP) and the GTX 1080 Ti ($699 MSRP).
Gaming performance
Steve from HardwareUnboxed and Techspot compiled a huge 32-game benchmark that we recommend for a deeper dive into the performance of all RX Vega cards. A few games will be used as an example of the general performance of the reference RX Vega 64 (worst case scenario: stock settings) and the liquid-cooled RX Vega 64 (best case scenario: close to maximum overclock).
It is important to note that even though the liquid-cooled version of RX Vega 64 does come at a price of $699, a highly overclocked and well-tuned aftermarket Vega 64 will most likely perform similarly for a comparable price to the GTX 1080 (when the prices actually stabilize).
DOOM, AMD’s best case scenario, shows what the company’s hardware is actually capable of. The RX 580 is 17% faster than the GTX 1060, its direct competitor, but the RX Vega 64 is at the same level of performance as the GTX 1080. This means either the game isn’t optimized for Vega yet, or the new architecture is that much different from Polaris. The performance of the top-of-the-line overclocked RX Vega 64 would be around 8% faster than a GTX 1080. Source: Techspot.com
GTA V, still one of the most popular games today, shows Vega’s worst case scenario. Even the liquid-cooled Vega 64 falls noticeably behind the reference GTX 1070. Driver optimizations are clearly needed for GTA V, but no one knows when or if these optimizations will appear. Source: Techspot.com
PUBG, the most popular game on Steam right now, is also an example of very poor optimization for the Vega architecture. The top-of-the-line liquid-cooled RX Vega 64 is merely on par with the reference GTX 1070. Source: Techspot.com
Titanfall 2, a game that more often favors AMD graphics cards, shows the state of RX Vega’s drivers. Even though the RX 580 is slightly faster than its direct competitor, the GTX 1060, stock RX Vega 64 falls behind the reference GTX 1080. The liquid-cooled Vega 64 can barely match an overclocked GTX 1080 from MSI. Source: Techspot.com
Total War: Warhammer running in DX12 mode is an example of good driver optimization for Vega. Just like the RX 580 vs GTX 1060, the reference version of Vega 64 is able to outperform an overclocked GTX 1080 from MSI, with the highly overclocked liquid-cooled version being almost 15% faster than the competition. Source: Techspot.com
Power consumption
By far the biggest gripe of Vega is not only it being very late to the market, but also its power consumption.
Overclock3D manually overclocked their RX Vega 64 as high as it would go and the total power consumption of the Vega system was actually comparable to a system with two GTX 1080 Ti cards in SLI. A single RX Vega 64 running in the pre-determined Turbo mode consumes a similar amount of power to an overclocked R9 390X or two highly overclocked ASUS Strix GTX 1080s in SLI. Even though it is possible to under-volt RX Vega cards, very few media outlets actually tested that with RX Vega 64. Source: overclock3d.net
Conclusion
Vega was hyped as being AMD’s best GPU architecture yet (as it happens with all new microarchitectures). Unfortunately, the Radeon Technology Group heavily underdelivered this time. Even though Vega 10 (the fully unlocked Vega 64 GPU) is of comparable performance to NVIDIA’s GP104 GPU (the fully unlocked GTX 1080 chip) in terms of gaming, it is still a year late to the market and is a much larger chip, making it more expensive to produce.
Judging by Vega 10’s size (484 mm²) it should be competing with the GP102 (471 mm²), the chip in Titan X Pascal and GTX 1080 Ti. And Vega does beat out GP102 in certain compute benchmarks, but in gaming it can barely compete with the much smaller GP104 (314 mm²). Even then, judging by the benchmarks presented by many media outlets, we can confirm that the drivers for RX Vega are still nowhere near where they should be. In some games where the RX 580 is able to beat the GTX 1060, Vega 64 falls behind a non-overclocked GTX 1070. In other games, overclocked Vega 64 can actually step on the toes of the GTX 1080 Ti. The drivers look to be a mess and it is not known when and if they will improve.
In conclusion, RX Vega 64 may be a good buy only if it’s cheaper than the GTX 1080 and is paired with a FreeSync monitor. Otherwise, accounting for the extreme power consumption, poor driver optimizations in many games and current lack of aftermarket cards and competitive pricing, we currently recommend an aftermarket GTX 1080 over the RX Vega 64. If the price of the AMD card is comparable to the GTX 1080 Ti, then there is absolutely no reason to get AMD. The 1080 Ti card is ~30% faster in all scenarios and consumes less power.
Sources: