AMD Radeon RX Vega 56 Performance Overview and Review 2019

[AlTech]

Introduction

From June 2014 to September 2017 the graphics card in my system was the GTX 770, which I've talked about a fair amount here and here.

 

In August 2017 I wanted to buy the new Wolfenstein 2: The New Colossus. It turned out that the GTX 770's measely 2GB of VRAM in a Vulkan title and the lack of optimizations due to being quite old at that point meant that performance was absolutely unplayable and within 30 minutes I applied to have the game refunded using Steam Refunds. Ultimately, this experience finally convinced me I needed to buy a new GPU as previously I had wanted to buy one but couldn't justify it.

Later on in September 2017 due to me leaving Hong Kong to go to Uni in the UK I ended up selling most of my PC parts and ultimately decided in January 2018 that I wanted to buy a new PC.

 

I was determined not to buy Nvidia because of my awful experiences with Nvidia drivers with my 770 in the last few months of using it. I wanted an AMD card. Unfortunately for me the mining craze was still very much alive in early 2018 and after careful consideration I decided to buy an RX 480 8GB at a ludicrous asking price of £320. It was actually a really nice card other than the excessive heat and noise due to being a blower reference card. I put up with the noise and heat for a while but in August 2018 I decided I wanted to sell my reference 480 and buy a much better 480. I ended up with a seller refurbished Sapphire Nitro+ RX 480 8GB and bought it for £199 on eBay. I was able to sell my reference RX 480 8GB but at substantial loss and I came to terms with the fact that I'd never get that money back.

 

I really liked my Sapphire Nitro+ RX 480 8GB. It was the perfect RX 480 8GB. But gaming had moved on and some video games were pushing the envelope of graphics too hard and also the increasing number of unoptimized PC titles meant that in some games my 480 delivered a mediocre experience. In December 2018 I decided I wanted to buy a new graphics card. Either the upcoming rumored at the time Navi based RX 3080 with the rumored performance of an RX Vega 56 or buying a Vega 56.

 

In January 2019 I made the decision to purchase an RX Vega 56 after realizing that any potential GPUs being shown off at CES would be many months away at best. By a stroke of luck I was unintentionally correct as AMD announced the Radeon VII, a $699 RTX 2080 competitor, instead of the Navi series.

 

Before I sold my Nitro+ 480 8GB I wanted to benchmark it to compare it to my current RX Vega 56 so I do have comparisons to it in most games and synthetic benchmarks I've chosen. In several synthetic tests I have results from my GTX 770 as well for a 3 way comparison.

 

 

Testing notes

 

The test system is my main PC with an R7 1700X Stock, MSI X370 Gaming Pro Carbon Motherboard, 16GB DDR4-3200 (2x 8GB), Corsair RM850i PSU, and Corsair H100i V2 CPU cooler.

 

The performance numbers for my RX 480 are from my Sapphire Nitro+ RX 480 8GB with AMD Radeon Software Adrenaline Edition 18.11.2 . My RX Vega 56 card is an Asus ROG Strix triple fan that I bought from @ChewToy! and is running AMD Radeon Software Adrenaline 2019 Edition 19.1.2 although at the time of posting the latest version is 19.2.1 . I didn't switch to the new version when it came out as I had already done a large amount of testing before and wanted to be consistent across all the games and tests.

 

Between changing GPUs the drivers were freshly removed using DDU, display driver uninstaller (a 3rd party program which is good for removing gpu drivers), with a new version downloaded from AMD's website and installed.

 

Both were on Windows 10 Version 1809 X64 with no real difference between them in the software side. During testing I closed all applications that weren't required for benchmarking.

 

Performance information in most games was recorded by FRAPS except in games using DX12 or Vulkan. In games using DX12 or Vulkan, the built in benchmarking tool was used which provided performance information.

 

Some games were tested using Virtual Super Resolution. It's basically AMD's version of allowing upscaling to higher than native resolutions. In those cases I have specified if it has been used and what resolution it was tested at. I did also decide to test some synthetics as well.

 

I may add more game benchmark results at a later date if I have time.

 

List of games tested (Year it was released) and notes:

• Ashes of the Singularity: Escalation (2016)
• Battlefield 4 (2013) - Actual gameplay of first story checkpoint
• F1 2015 (2014)
• For Honor (2016)
• Grid Autosport (2014)
• GTA V PC (2015) - Actual gameplay of first story mission
• Metro 2033 Redux (2014)
• Metro: Last Light Redux (2014)
• Middle Earth: Shadow Of Mordor (2014)
• Tomb Raider (2013)
• Rise of The Tomb Raider (2015)
• Watch_Dogs 2 (2016) - Actual gameplay - I have found some issues with the benchmark results so I'll be re-doing the Vega 56 benchmarking for it.

 

List of Synthetic Benchmarks tested

• 3DMark Firestrike
• 3DMark Firestrike Extreme
• 3DMark Timespy
• 3DMark Timespy Extreme
• Folding at Home Benchmark

 

Benchmark charts

 

Gaming Benchmarks

Spoiler

Notes: The "Crazy" preset is the highest preset and "Extreme" is the preset below it.

I tested VSR 1440p Extreme just to see how playable it would be at 1440p if some settings were turned down.

 

Despite being a 2013 title, Battlefield 4 is still taxing to to this day at 4K. This uses the first checkpoint for benchmarking starting from Dunn in the sinking car to getting Irish into the safe house safely.

 

F1 2015 is a much easier title to run by comparison and the Vega 56 is able to deliver 4K 60fps whereas the RX 480 8GB is good for 4K 30fps with a tiny bit of headroom.

 

Things are much the same with Grid Autosport although the RX 480 and Vega 56 deliver higher framerates with Vega 56 offering almost a rock solid 60fps 99.9% of the time.

 

With For Honor since I didn't have any RX 480 benchmark results, which is my bad because I did own For Honor at the time but I didn't think to benchmark it until later, I've chosen to show performance at 1440p maximum settings and 4K High settings to showcase how dropping the settings can make the framerate playable.

 

For GTA V PC I decided to test the first mission in the Story as the built in benchmarking tool provides unrealistic results and the first mission provides and adeequate range of shooting, driving, switching characters and in game rendered cutscenes.

 

Once again back to using built in benchmarks. Metro 2033 Redux does okay on the RX 480 although the 0.1% and 1% lows are not great. The Vega 56 is able to deliver 4K 60fps on average but doesn't seem to be able to hold it consistently and as a result double buffered V-Sync or FreeSync with Low Framerate Compensation seems preferable as a solution to this issue.

 

Moving on to the remastered Metro Last Light,

 

 

 

 

Synthetic Benchmarks

Spoiler

 

 

 

 

 

I'll re-add Watch Dogs 2 results once I'm satisfied the re-tested benchmark results are indicative of actual gameplay performance.

 

Conclusion

Overall then is the Vega 56 a good upgrade from someone around an RX 480 level of performance? I would say hell yeah. It's a major improvement in performance and the only thing holding it back is prices although prices have gotten a lot better recently. If you can get one near MSRP or buy one used for cheaper then it's definitely worth it. I'll be doing more testing on my Vega 56 in the future. Sorry I rushed this a bit.

 

I'm going to update this with the synthetic performance and more benchmarks later so check back on this thread in the future for updates.

 

Definitely am thrilled with my purchase and I would buy a Vega 56 again if given the same situation.

 

Thanks for checking this out .

[Firewrath9]

Good review. What sofware did you use for making the charts? I upgraded from a 670 to a 1070, it was night and day. Almost 4x performance upgrade. I upgrade every 4gens to a full system rebuild.

 

[AlTech]

Just now, Firewrath9 said:

Good review. What sofware did you use for making the charts?

Excel although LibreOffice Calc can also make graphs.

Just now, Firewrath9 said:

I upgraded from a 670 to a 1070, it was night and day. Almost 4x performance upgrade. I upgrade every 4gens to a full system rebuild.

 

Nice.

[Firewrath9]

On 2/9/2019 at 9:49 PM, AluminiumTech said:

Excel although LibreOffice Calc can also make graphs.

Nice.

Thanks, pls dont laugh but i use google sheets lmao

(i use word tho)

[xAcid9]

Man if i'm not broke i probably have Vega56 in my hand right now. They cost like $220-$250 2nd hand with factory warranty which is cheaper than a brand new 1060 6GB. It's probably a small upgrade or simply a sidegrade but V56 Nitro+ is just too sexy to pass.

[ch3w2oy]

16 hours ago, xAcid9 said:

Man if i'm not broke i probably have Vega56 in my hand right now. They cost like $220-$250 2nd hand with factory warranty which is cheaper than a brand new 1060 6GB. It's probably a small upgrade or simply a sidegrade but V56 Nitro+ is just too sexy to pass.

Not all the Nitro+ have the RGB.. But they really are great cards..

[Morgan MLGman]

Nice. I'm looking forward to when you add FireStrike GPU scores so I can compare not only my 1080Ti, but also my previous cards (1080, R9 Fury, R9 290X) and see how they stack up with the V56 once overclocked.

Have you overclocked the card? Are those charts that you did at stock clock speeds?

[AlTech]

10 hours ago, Morgan MLGman said:

Nice. I'm looking forward to when you add FireStrike GPU scores so I can compare not only my 1080Ti, but also my previous cards (1080, R9 Fury, R9 290X) and see how they stack up with the V56 once overclocked.

Have you overclocked the card? Are those charts that you did at stock clock speeds?

The clockspeeds are without me ocing it but the card is already factory overclocked to 1592MHz although sometimes it stays around 1400MHz

[AlTech]

On 2/11/2019 at 1:32 PM, Morgan MLGman said:

Nice. I'm looking forward to when you add FireStrike GPU scores so I can compare not only my 1080Ti, but also my previous cards (1080, R9 Fury, R9 290X) and see how they stack up with the V56 once overclocked.

Have you overclocked the card? Are those charts that you did at stock clock speeds?

Sorry for taking so long to get this done. I should have the redone synthetics shortly.

 

Edit: Synthetics are done and I've added them to the original post .

[GamerDude]

The VEGA series got a bad rep price-wise due to the mining craze, grab a used VEGA56/64 (I myself had grab 2x Gigabyte VEGA64 at pretty decent price) and they do very nicely in games. My dual VEGA64 can do Metro Redux series (no CF support for Exodus) at 3440x1440, max ingame + SSAA at 2x and still net the performance of a single RTX2080 Ti (and they were cheaper than an RTX2080 Ti by a fair margin). I had an RTX2080 Ti before going dual VEGA64's (in the same rig) and Vega II to boot......yes, something of an AMD fanboy I'll admit to that.

[BuckGup]

Power table mod??

[Princess Luna]

Here in Brazil second hand Vega 64 (unfortunately blower coolers) are cheaper than brand new RTX 2060 cards... can't deny it makes me think of them more but with FreeSync finally being among nVidia features I don't know if it's worth it the heat + noise.