AMD Radeon VII Review: The highend contender

From the center stage of CES last month, AMD gave us the outlines of two major products. One was about Ryzen gen 2 and we expected as much; the other, however, caught many of us off-guard. Not to be out-done by peers from CPU division, RTG (Radeon Technologies Group) announced Radeon VII, world's first 7nm consumer graphics card with 16GB HBM2 video memory. Fast forward a month and the Radeon VII is here with a price tag of $700 (~₹60K). 

AMD Radeon VII and its packaging, click to enlarge

Considering AMD's focus on the mainstream of late, not many anticipated it would be gunning for high-end with its next GPU. But that's exactly what Radeon VII is supposed to do - to take the fight to the likes of Nvidia GeForce RTX 2080. And that's easier said than done as Nvidia has asserted its dominance over the high-end ever since the launch of its RTX 2000 series of graphics cards. Powered by Turing architecture, these GPUs not only routinely outperform AMD's Vega and Polaris based products but also features Real-time Ray Tracing. In our review of GeForce RTX 2060, we saw Nvidia's mid-range card to nibble at the heels of RX Vega64 indicating how desperately AMD needs to update its aging line-up.

Specs, Techs and Features

Along cometh Radeon VII, directly supplanting RX Vega64 as the fastest Radeon in the market. Built on cutting-edge TSMC 7nm process and an improved Vega 20 GPU, AMD's latest and greatest graphics card comes with 3840 shader cores and a staggering 16GB of 2nd gen High Bandwidth Memory (HBM)! Let's have a look -

Graphics card	Radeon VII	RX Vega64	GeForce RTX 2080 FE
GPU	Vega 20	Vega 10	TU104 (Turing)
Process	7nm TSMC	14nm GloFo	12nm TSMC
Shader cores	3840	4096	2944
Base clock	1400MHz	1247MHz	1515 MHz
Boost clock	1750MHz	1546MHz	1800 MHz
Memory clock	1000MHz	953MHz	1750 MHz
Memory Bus	4096-bit	2048-bit	256-bit
Memory Bandwidth	1 TB/s	484GB/s	448GB/s
V-RAM	16GB HBM2	8GB HBM2	8GB GDDR6
ROPs	64	64	64
Texture Units	240	256	184
Power(TDP)	300w	295w	225w
Price	₹ 60000	₹ 50000	₹ 70000

Essentially, Vega 20 is a die-shrunk Vega 10 with faster clocks and wider memory interface. At 331mm2, Vega 20 is smaller than Vega 10's 495mm2 die. But thanks to the move to a smaller 7nm process node, AMD has been able to increase the transistor count on Vega 20 to 13.2 billion compared to its predecessor's 12.5 billion. As a result, Radeon VII enjoys higher engine clocks. With a 1400MHZ/1750MHz base/boost configuration, the new Radeon has a 200MHz clock-speed advantage over RX Vega64 which also levels the playing field with competing RTX cards.

The biggest highlight of Radeon VII is its memory subsystem which sports 16GB HBM2 along a 4096bit wide interface. AMD first introduced High Bandwidth Memory with its Fury-X graphics cards since when the memory standard has matured quite a bit, although it's still costly compared to conventional GDDR ones. The advantage of HBM is its much wider memory bus or interface which enables significantly higher memory throughput to feed the ever increasing number of shader cores.

GPU die comparison, click to enlarge

Leveraging TSMC's advanced 7nm node, AMD has managed to double the amount of video memory  adding 2 more HBM2 stacks on the Vega 20's silicon interposer. The theoretical total bandwidth of Radeon VII is a stupefying 1TB/s! This also means that the shader cores and ROPs (Render Output Units) will be getting twice as much bandwidth.

Architecturally, Vega 20 is pretty similar to Vega 10 as it's a 4 Shader Engine design with each housing 16 Compute Units (CU) along with geometry processing engine and DSBR (Draw-Stream Binning Rasterizer) as a front-end. Each CU has 64 Stream processors and 4 texture units inside it.

Logical diagram of Vega 20 GPU, click to enlarge

An interesting observation would be the striking similarity between Radeon VII and Radeon Instinct MI50 - AMD's datacenter/HPC oriented professional solution. Both are driven by the same Vega 20 GPU, only the one inside Radeon VII misses a few CUs resulting in a total of 3840 Shader Cores and 240TMUs (Texture Mapping Unit). AMD has also limited some of Instinct's general purpose computing abilities which is understandable with Radeon VII being a consumer graphics card aimed more towards gaming. Still, Radeon VII delivers a very respectable 3.5 TFLOPS (TeraFLOPs) of double precision FP64 computing, something that's unmatched within this price bracket.

All the other features of Vega 10 like Primitive Shaders, HBCC (High Bandwidth Cache Controller), DSBR and rapid packed math are still present on Vega 20 along with some minor refinements of the Asynchronous Compute Engine (ACE), video encoding and addition of a few new instructions and data types.

Oh, and you also get FreeSync - AMD's alternative to Nvdia's proprietary G-Sync; both of which are Adaptive Sync technologies employed to provide you with smooth tearing-free gaming experience. The one big difference being price and as the name suggests, AMD's implementation is cheaper, giving rise to many more affordable FreeSync monitors than there are G-Sync ones. Recently, Nvidia has expressed its willingness to embrace AMD's free standard which bodes well for the future of FreeSync. 

The card

Radeon VII is launched in its reference designs made exclusively by AMD much in the same vein of Nvidia's Founders Editions. There will be partner cards with custom designs (and probably with custom clocks too) later. The reference card sports a clean industrial look, clad in steel-gray metal plates. It's an inconspicuously big and heavy card but no bigger than RTX 2080 FE. The construction is solid and components used are of high quality.

AMD's reference design, click to enlarge

AMD has made use of a vapor-chamber design with three axial fans blowing out the heat. Most reviewers found this cooling solution to work well with Radeon VII's 300w thermal envelop. Both Tom's Hardware and Techpowerup found average load temperature during gaming is well within 75*C which is acceptable from a card of this level.

Acoustic is a whole other story though! while at idle or mild load the fans are whisper quiet, under peak gaming load they start to get real noisy real quick indicating aggressive fan curves which some reviewers found to be annoying.  Also these fans lack the "long idle" halt feature, found in some of AMD's older generation reference designs where the fans stop spinning when there isn't any load for extended period of time. Regardless, it seems that this time around the red camp has traded off acoustics for performance. 

Thermal sensors, click to enlarge

In an effort to increase the sustenance of boost-clock and reduce throttling AMD has also refined Vega 20's power monitoring mechanism. It now have 64 integrated thermal sensors, doubling Vega 10's 32 sensors, covering a wider area across the GPU die resulting into more accurate measurements of temperature. The junction temperature as AMD calls it gives the GPU even more granular control over clock-speed, voltage and fan curves. 

Next-up is power efficiency which has not been AMD's strong suite lately, at least on the GPU side of things. On paper Radeon VII's board power is 300w and again the card generally adheres to the rating except a few scenarios where stress test like FurMark is involved. Most reviews show Radeon VII to draw less power than Vega 64, despite the former's clock-speed advantage and larger HBM2. So, AMD has definitely made some improvements there, partly thanks to TSMC's 7nm process. That's not enough though in the face of rivaling Turing cards' stellar efficiency. Radeon VII's main competition, the GeForce RTX 2080 is rated at 225w and on average consumes less energy to deliver slightly better (as you'll see) gaming performance. Anandtech's readings show the difference to be around 40w and PCworld's observation echoes that. This may not seem like a huge disparity if you prefer to ignore the fact that Nvidia's GPUs are still on 12nm!

Performance, value and verdict

AMD appointed Radeon VII with the sole task of dethroning RTX 2080 at the high-end of things. After going through a considerable amount of benchmarks published by many review sites, we can safely conclude that it falls a little short in that endeavor. That's not to say Radeon VII isn't fast, it is in fact the fastest Radeon AMD ever launched. Endowed with higher clock-speed and more than ample memory bandwidth, Radeon VII can hold its own against RTX 2080, especially in higher resolutions like QHD and 4K. In some latest AAA tittles like Far Cry 5 and Battlefield 5, it outperforms the RTX 2080 but trails overall.

We aren't able to provide you with in-house benchmark data of Radeon VII for the simple reason of not having one. Instead we present you with something better. Not relying on a single reference or single game, we've painstakingly compiled an indicative performance chart from numerous reliable data points across the web. The performance data will give you a clear picture of how well Radeon VII does against GeForce RTX 2080 on average. We've also included the results vs. Vega 64 to get an idea of generational increment within AMD's own ranks.

The data is derived at 4K (3840x2160 pixels) as that'd be the target resolution of most users who are interested in a graphics card of this caliber. The only exception is Techspot's result as they benched the card at QHD (2560x1440pixels). Also remember that the chart represents an average percentage of total FPS (Frames per Second) from an entire suite of game benchmarks conducted by different sites, these are not results of a singular tittle.

First up is GeForce RTX 2080 vs. Radeon VII:

Average performance percentage - higher is better, click to enlarge

Depending on the selection of games, the performance gap is between 2% (PCworld) - 10% (Techpowerup) in favor of RTX 2080. The variation is down to game optimization - an area where Nvdia always seems to have an advantage. We don't like the idea of buying a card on the merits of games being played but fact remains that there are more Nvidia optimized tittles these days than there are AMD ones.

Overall, Radeon VII has no problem churning up 60FPS in most tittles @1440p and can do so even @4K if you're willing to relax the visual settings. As the trend emerges, Radeon VII does very well in Battlefield 1, Battlefield 5, far Cry 5, Strange Brigade, Tom clancy's The division, Wolfenstein 2, F1 2018, Deus Ex: Mankind Divided and Middle-earth: Shadow of War. Games such as Ashes of Singularity: Escalation, Hitman 2, Resident Evil 2, GTA 5, Assassin's Creed Odyssey provide a somewhat middle ground while Forza Horizon 4, Destiny 2, Rise of the Tomb Raider, Final Fantasy XV, Tom Clancy's Rainbow Six Siege, Fortnite, Monster Hunter: World typically favors RTX 2080.

Another thing we noticed is when Radeon VII leads it isn't by much but when it trails the gap seems to well widen which tends to skew the overall performance graph in favor of Nvidia's offerings. Release day drivers often play a key role in this kind of scenarios and RTG definitely needs to pay more attention here.

Now let's see how AMD's latest and greatest does against its own kind. Radeon VII vs. RX Vega 64:

Average performance percentage - higher is better, click to enlarge

Free from optimization differentials, the pattern is more consistent here. Radeon VII delivers 30% higher FPS than AMD's previous flagship and it does so with fewer shaders and less power. Impressive, but at the same time it's still good old GCN (Graphics Core Next, 5th gen) which means Vega 20 has the same strengths and weakness as Vega 10 and to certain extent Fiji before that. AMD has done a commendable job of carving out the extra performance without changing the underlying architecture in any major way. Radeon VII arguably marks the limits of GCN in its current form and it's about time AMD does an overhaul of things at the core level - hopefully with Navi.

An even bigger question here is about Vega 20's 16GB frame buffer and how much of an impact are we seeing from it. Radeon VII scales very well with higher resolutions due to the massive bandwidth provided by 2nd gen HBM and that's the reason we chose to stick with 4K numbers. But even at 4K the impact isn't as significant as AMD would've like it to be. Lower the resolution and Radeon VII's performance advantage shrinks.

So what exactly is happening here? To begin with, we believe memory bandwidth or capacity has not been that critical an issue for Radeon GPUs. Rather, increasing frequency and power efficiency of its GPU design has proved to be more challenging for AMD in recent years. Also, while there is no such thing as too much V-RAM, most modern games are used to do more with less. Unfortunately, we couldn't find any benchmarks of games with high-res texture mods - something we know to gobble video memory like no other. We need further exploration into the matter but for now 16GB is more future-proofing than necessity.

This however opens up avenues beyond gaming. Thanks to its datacenter/sever DNA and AMD's lenient capping, Radeon VII packs some heavy general purpose computing punches for those who need that, making it an ideal prosumer graphics solution. High-res encoding, 3D rendering and any such bandwidth sensitive workload should nicely play into the forte of Radeon VII. The amount of V-RAM is also very crucial for content creators and game developers who can tap into its massive memory pool. But that's a different demographic altogether, one which is heavily application specific with an even greater focus on driver optimization. 

With Radeon VII, competition has finally come back to upper spectrum of PC gaming and that's something we generally welcome with great fervor. But this time the competition is rather symbolic which puts us into a bit of quandary. At ₹ 60K, this is not a cheap graphics card and price alone can be a big deterrent for average gamers but then this isn't an average gaming card. For all intent and purpose, Radeon VII is a high-end graphics card targeted  towards enthusiasts playing at 4K. That we believe is the perfect resolution which lets the card flex its HBM muscles. Anything lower and there are better alternatives value wise.

The ₹10K price advantage that Radeon VII holds over its primary competition -the RTX 2080 is all but negated by the fact that it's slower in general. Then there is the matter of Realtime Ray Tracing and DLSS. As promising as they are, it's very early for these features (or the lack thereof) to become decision making factors. Nonetheless, if you want the extra eye-candy, going green is the only way as of now.

We believe this card could have been a very alluring prospect had AMD been able to offer it at a lower price, preferably around ₹50K mark. But at its current price we can neither wholeheartedly recommend it over RTX 2080 nor can we see it put any kind of pressure on Nvidia to reconsider its pricing policy. 

With all said and done, it turns out that the best thing we like about Radeon VII is its game bundle which includes free copies of Devil May Cry 5, Tom Clancy's The Division 2 and Resident Evil 2. Great value addition but is that enough to persuade you into buying it? We leave that decision with you.

Sources and references:
AnandTech, tom's Hardware, Techpowerup, PCWorld, TechSpot, HardwareCanucks