GPUs Graphics Reviews

Small, hot-cold sister: AMD RX Vega56 with efficiency test and VR

In the end, however, this remains the most exciting question of the day, because there has been little noticeable about a wide availability and above all stable software. So it can only get better. But what's left in the end ... Disassembly and radiator details Removing the top hee cover is easy. With a small Phillips screwdriver (PH1), the six small swivels that hold this cover can be turned out. After that, there is only one... Board layout The RX Vega56, RX Vega 64 and Vega Frontier Edition have the same board, 100% identical components and differ only in the soldered package, as well as a customized firmware. The length is, consider... Ethereum Mining The latest version of Claymore's Dual Ethereum AMD/NVIDIA GPU Miner (V8.9.8) provides support for the Radeon RX Vega, so what we use it for our mining benchmark. All AMD cards run in ASM mode, which is a fe... Dirt Rally (DirectX11) This game has a built-in benchmark and therefore offers absolutely reproducible conditions for each pass While the GeForce GTX 1080 never falls into ASW mode, except for a few dripped frames, the G... Foreword to gaming benchmarks On the following pages, we let the bars and curves speak for themselves and do without filling text. In return, and at the request of many readers, we have not only the overall overviews of all maps. Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Benchmarks in WQHD (2560 x 1440 Pixels) Benchmarks in Ultra HD (3840 x 2160 pixels) Power consumption, BIOS selection and presets AMD (as on the Radeon RX Vega64) also offers two BIOS variants on the Radeon RX Vega56 via switch, whose presets differ from each other. The Power Limit as long as you don't... Clock rates and corresponding gaming performance With Witcher 3 in Ultra HD, we are now testing the worst case at maximum load. Let's just start from the default setting on delivery and set the "Balanced" mode of the BIOS and the there... Fan speeds and operating noise ("volume") As we already know, AMD is trying to prioritize GPU temperatures. It should therefore be clear to everyone that this must be at the expense of fan speeds. First, let's look at the throughs. The RX Vega56 can be convincing in some areas, but not in others. The biggest shortcomings are the launch, which took place one year too late, the almost homeopathic availability of the Vega cards as a whole and the disproportionately high...

In the end, however, this remains the most exciting question of the day, because there has been little noticeable about a wide availability and above all stable software. So it can only get better.

But what's left in the end if, for whatever reason, you don't get a sample to be on the launch in time for the early launch day? That's why we tried to generate at least some added value with an additional section on VR performance and an efficiency view, so as not to the 1001. to offer a review from the bar that no one is interested in any more.

However, the VR area around our FCAT measurements in particular shows that it is not only AMD that operates open construction sites without protection. We had to postpone our test several times because an FCAT library was faulty. Together with Nvidia, we were able to test and use the latest software exclusively. At least these values can only be read exclusively here and it already has a certain entertainment value that even Nvidia makes an extra effort to indirectly ensure an RX Vega56 article in a reasonably timely manner.

At this point, of course, we also refer again to our test of the Vega Frontier Edition and the launch article for the RX Vega64, both of which also contain the corresponding information about the technology and theory under the hood. In addition, we had already released another follow-up with the water cooling conversion ("Overclocked and boiled: AMD Radeon RX Vega64 water-cooled") that could show other facets.

What distinguishes the Radeon RX Vega56 from the RX Vega64?

The Radeon RX Vega 56 uses the same Vega-10 processor as we find in the Vega64. It is a 486 mm2 chip with 12.5 billion transistors produced in GlobalFoundrie's 14LPP process. There are four shader engines, each with its own geometry processor and Draw Stream Binning Rasterizer.

If the RX Vega64 still has 64 active compute units in these shader engines, AMD disables two CUs per shader engine on the RX Vega56, so it ends up at 56. Which explains the name. With 64 stream processors and four texture units per CU, you get 3584 stream processors and 224 texture units – i.e. approx. 88% of Vega NCU resources.

The lower computing power and the lower texture throughput are also due to the lower base and typical boost clock rates. Compared to the Vega64, the Radeon RX Vega56 only takes 1156 MHz instead of 1274 MHz, while the potential boost clock of the RX Vega56 is 1471 MHz compared to the 1546 MHz of the RX Vega64. The theoretical SP performance thus decreases from a maximum of 13.7 TFLOPS to 10.5 TFLOPS, although these are of course only absolute peak values.

Each of Vega 10's shader engines has four render back-ends that deliver 16 pixels per cycle, resulting in the 64 ROPS. These render back-ends hang together on the L2, which is now 4 MB in size, while Fidji would still have to cope with two MB of L2 capacity. Ideally, this means that the GPU will need to access the HBM2 less frequently and reduce Vega 10's reliance on memory bandwidth. Since the clock rates of Vega 10 on the 56-Cu card are up to 40% higher than Fiji's, but the memory bandwidth drops to 102 GB/s, a larger cache here should be helpful to prevent bottle necks on the flagship.

The use of HBM2 allows AMD to halve the number of storage stacks on the interposer compared to Fiji, almost halving the aggregated 4096-bit bus. Compared to the 4GB HBM modules of the Radeon R9 Fury X, the Radeon RX Vega56 uses quite comfortable 8 GB with 4-Hi stacks, similar to the Vega64. With a 1.6 GB/s data rate, you then create 410 GB/s bandwidth, which in theory is the values of a GeForce GTX 1070 or GTX 1080 with GDDR5 or GDDR5 GDDR5X.

Optics, haptics and connections

The approx. 1064 grams heavy card (14 grams more than the Frontier Edition) is 26.8 cm long (from outside edge slot aperture to end case), 10.5 cm high (from top board slot to top of housing) and 3.8 cm deep. This makes it a real dual slot card, even if the backplate is again approx. 0.4 cm on the back.

The housing cover and backplate are made of black anodized aluminium, which is valuable and cool. The surface structure of the lid was simply realized by cold deformation before anodizing. All screws are painted matt black. Only the printed red Radeon logo at the front stands out in colour.

The top is characterized by the two 8-pin PCIe clamping supply connectors, as well as the illuminated red Radeon logo. We also find a BIOS switch that provides access to a BIOS with significantly low power consumption. Quiet, cooler and of course a little slower. Together with the three new modes in the Wattman "Turbo" (max. Power Limit), Balanced, and Power Save (min. Power Limit) results in several variations, which we will discuss later.

The end of the card is closed and on the frame you will find the usual holes for mounting workstation cards. The matt black powder-coated slot aperture accommodates three DP connectors and an HDMI 2.0. A DVI-I has been cleverly dispensed with for reasons of flow technology, because the aperture is at the same time the outlet of the warm exhaust air from the cooling system.

Model Radeon
R9 Fury X
GTX 1080
GTX 1070
Gpu Vega 10 XTX Vega 10 XTX Fiji XT GP104 GP104
Chip size 484 mm2 484 mm2 596 mm2 314 mm2  314 mm2 
Transistors 12.5 billion 12.5 billion 8.9 billion 7.2 billion 7.2 billion
GPU base clock/
Boost clock
1247 MHz
1630 MHz
1156 MHz
1471 MHz
1050 MHz
1606 MHz
1733 MHz
1506 MHz
1683 MHz
Shader/SIMD 4096/64 3584/56 4096/64 2560/20 1920/15
Texture Units/ROPS
256/64 224/64 256/64 160/96 120/64
Memory connection 2048 Bit 2048 Bit 4096 Bit 384 bits 352 bits
Storage type HBM2 HBM2 Hbm GDDR5X GDDR5
Memory bandwidth
483.8 GB/s 410 GB/s 512.0 GB/s 320.3 GB/s 256.3 GB/s
Speed. Graphics
1.89 Gbps 1.6 Gbps 1.0 Gbps 10.0 Gbps 8.0 Gbps
Storage expansion
8 GB 8 GB 4 GB 8 GB 8 GB
DX12 Feature Level 12_1 12_1 12_0 12_1 12_1
PCIe sockets 2 x 8-pin 2 x 8-pin 2 x 8-pin 6 + 8-pin 8-pin
Tbp 295 watts 210 watts 275 watts <250 Watt  <150 Watt 

Test system and measurement methods

The new test system and the methodology have already been described in great detail in the basic article "How We Test Graphics Cards" (English: "How We Test Graphics Cards") and therefore, for the sake of simplicity, we now only refer to this detailed Description. So if you want to read everything again, you are welcome to do so. However, we have again improved CPU and cooling to largely exclude possible CPU bottlenecks for this fast card.

If you are interested, the summary in table form quickly provides a brief overview:

Test systems and measuring rooms
Intel Core i7-6900K -4.3GHz
MSI X99S XPower Gaming Titanium
Corsair Vengeance DDR4-3200
1x 1 TByte Toshiba OCZ RD400 (M.2, System SSD)
2x 960 GByte Toshiba OCZ TR150 (Storage, Images)
Be Quiet Dark Power Pro 11, 850-watt power supply
Alphacool Ice Block XPX
Alphacool Ice Age 2000 Chiller
2x Be Quiet! Silent Wings 3 PWM (Closed Case Simulation)
Thermal Grizzly Kryonaut (for cooler change)
Lian Li PC-T70 with expansion kit and modifications
Modes: Open Benchtable, Closed Case
Monitor: Eizo EV3237-BK
Power consumption:
non-contact DC measurement on the PCIe slot (Riser-Card)
non-contact DC measurement on the external PCIe power supply
Direct voltage measurement on the respective feeders and on the power supply
2x Rohde & Schwarz HMO 3054, 500 MHz multi-channel oscillograph with memory function
4x Rohde & Schwarz HZO50, current togor adapter (1 mA to 30 A, 100 KHz, DC)
4x Rohde & Schwarz HZ355, touch divider (10:1, 500 MHz)
1x Rohde & Schwarz HMC 8012, digital multimeter with storage function
Optris PI640, infrared camera
PI Connect evaluation software with profiles
NTI Audio M2211 (with calibration file)
Steinberg UR12 (with phantom power for the microphones)
Creative X7, Smaart v.7
own low-reflection measuring room, 3.5 x 1.8 x 2.2 m (LxTxH)
Axial measurements, perpendicular to the center of the sound source(s), measuring distance 50 cm
Noise in dBA (Slow) as RTA measurement
Frequency spectrum as a graph
Driver Radeon: 17.20.1035
Quadro: R381 U2 (382.05)
Operating system Windows 10 Pro (Creators Update, all updates)

Danke für die Spende

Du fandest, der Beitrag war interessant und möchtest uns unterstützen? Klasse!

Hier erfährst Du, wie: Hier spenden.

Hier kannst Du per PayPal spenden.

About the author

Igor Wallossek

Editor-in-chief and name-giver of igor'sLAB as the content successor of Tom's Hardware Germany, whose license was returned in June 2019 in order to better meet the qualitative demands of web content and challenges of new media such as YouTube with its own channel.

Computer nerd since 1983, audio freak since 1979 and pretty much open to anything with a plug or battery for over 50 years.

Follow Igor:
YouTube Facebook Instagram Twitter