AMD Radeon RX Vega 56 review: lucky star

Getting to know AMD's new flagship consumer video cards, the Radeon RX Vega 64, left a bitter taste of disappointment. While paying tribute to the impressive technical characteristics and innovations of the fifth-generation GCN architecture, we have to admit the fact that the Vega 10 chip cannot use its undoubtedly colossal computing power in gaming applications as efficiently as representatives of the competing NVIDIA Pascal family. In general-purpose calculations, the Radeon RX Vega 64 is not inferior to the GeForce GTX 1080 Ti either in terms of performance or flexibility in working with various data formats, but in the gaming accelerator market, AMD had to choose an easier target - the GeForce GTX 1080.

Although AMD's position may be strengthened in the future by the adoption of the DirectX 12 API, which reveals the capabilities of GCN more fully than DirectX 11, as well as the optimization of drivers and game engines for the new architecture, the successes that Vega 64 has achieved today came at a disproportionately high price - in terms of both the power consumption of the board and the cost of producing such a large and complex GPU. Low supply and high demand for Radeon RX Vega 64 refuted the optimistic recommended price of the accelerator ($499). With real prices in the range of $650–700, Vega 64 remains an excellent solution for non-graphics calculations, but in games the price-performance ratio is not in favor of AMD.

Fortunately, releasing an attractive second-tier card based on the Vega 10 processor is a much easier task. In this case, AMD engineers were not required to conquer the peak erected by their rival (more than a year ago, as our readers cynically remind us) at any cost. Since NVIDIA has built a fairly large distance between the GeForce GTX 1080 and GTX 1070, it was only necessary to choose the specifications and price of the Radeon RX Vega 56 in such a way as to give gamers the most advantageous offer while simultaneously maximizing the flow of Vega 10 chips into the market due to samples with defective computing units . Let's see how AMD solved this equation and whether the Vega 56 makes a more convincing bid for class leadership than its older sister.

⇡#GPU Vega 10

The Vega 10 GPU at the heart of the Radeon RX Vega family represents the biggest change to the Graphics Core Next architecture since it was first introduced in 2011-2012. However, Vega 10 is true to the priorities that AMD set at the dawn of GCN: the formula of computing units of the latest GPU is still shifted towards shader ALUs - based on predominantly non-graphics or mixed workloads. In fact, the new GPU repeated the Fiji configuration (Radeon R9 Fury series), including 4096 shader ALUs, 256 texture mapping units and 64 ROPs, adjusted for a different organization of cache memory access and double the L2 cache size. AMD spent the additional transistor budget (and in Vega 10 it is increased by 40% compared to Fiji) on optimizing the existing logic and introducing new functions.

In particular, AMD has expanded the computing capabilities of Vega by “compressing” operations of reduced bit depth. So, instead of the 128 operations per clock on FP32 numbers that a single NCU (the main building block of the GCN containing 64 shader ALUs) performs, 256 FP16 operations or 512 FP8 operations can be performed. Reduced-precision mathematics in discrete GPUs is currently in demand primarily in machine learning tasks, but over time it may find wide application in graphics, since now both NVIDIA and AMD provide such an opportunity.

AMD Vega 10 Block Diagram

Despite the continued emphasis on computational load in AMD chips, the creators of Vega did not ignore the fixed functionality blocks associated with 3D rendering. Polaris chips have already made big strides forward over their predecessors (Tonga and Fiji) in the efficiency of using RAM bandwidth and processing geometry early in the rendering pipeline. Vega received a completely new, partially programmable, geometry engine and a separate scheduler of operations on geometric primitives, and the relatively small number of texture mapping units for such a large GPU compensates for tile-based rendering. Finally, Vega has the most comprehensive support for DirectX 12 "in hardware" among modern GPUs - an area in which AMD has long been a catching up player.

The merits of AMD, and before that ATI, include the initiative in introducing new types of RAM and ways of working with it. Vega was the first GPU in the consumer market to use HBM2 memory, and for the first time on AMD GPUs, it used page access to the address space, which allows the GPU's local memory to be used as a third-level cache and access to "far" memory (system RAM) at the driver level - the same way it works in central processors.

You can read more about the Vega features listed here and other features in our review of the AMD Radeon RX Vega 64. In general, we can say that AMD has a more balanced, versatile and in many ways advanced graphics processor than in previous generations. However, many of the innovations in Vega 10 (as has happened more than once with AMD chips) represent an investment in the future, and do not bring benefits here and now, or are aimed at professional and prosumer market niches. On the contrary, in games AMD is only burdening the desire to make a universal chip for graphics and computing tasks. Or perhaps the GCN architecture itself has already encountered fundamental limitations that only a fundamental paradigm shift within the walls of the Radeon Technologies Group will eliminate.

Customer Reviews

Buyers unanimously say that the video card is well made, has a good cooling system and sufficient power for all modern games. The Pulse series is built on a simplified PCB, but this did not affect performance much

Some users also noted negative aspects in their reviews. This is the noise of the fans, the squeaking of throttles in games, the complexity of settings, and most importantly, the size of the video card, due to which as many as three slots are occupied.

Very presentable appearance and better performance than expected. November 28, 2022 [Verified purchase] For comparison, the ASUS R9 390 STRIX card in the photo - on the “sapphire” you can see a massive heatsink and an unusual “Energizer” style design. The video card is equipped with Samsung HBM and impresses with the absence of much noise and good performance.

Review from a satisfied customer, translation of review from Amazon

Nice card Oct 24, 2022 [Verified Purchase] At the time of publishing this card was ok for $479.99, unfortunately I purchased this card for $679 (It was even on sale) because I thought that its price will rise even more. The card is very quiet, the fans rarely come on when running in my system, and I would like to talk more about its performance, but my experience with this card is somewhat negative due to the lack of available drivers for Windows 8.1 and the fact that my 2700X is not win8 supported .1. Even though I don't have driver support, the card works quite well. It scored 4804 graphics points in 3DMark Firestrike Ultra.

Customer review from Amazon

Great for 1080p! July 14, 2022 [Verified Purchase]

I have a 1080p 144Hz monitor and I can game at a constant >60fps when streaming video to a second monitor with this GPU. The only time it went below 60 was with FO4, but I've heard it's terribly optimized. Make sure you have a good power supply, power consumption was above 180W. However, the temperature never exceeded 56 degrees. I've heard that you can flash a Vega64 BIOS on it for better performance, but I have no idea how to do that and there's really not much need for it yet.

Read also:  Blockchain Beta Compute Driver: Download and install

Customer review from Amazon.com

⇡#Technical characteristics, prices

The specifications of the Radeon RX Vega 56 are significantly reduced compared to accelerators based on a fully functional GPU. Here, 56 of the 64 NCUs are active, but this alone means a reduction in the processing power of the shader ALUs and texture mappers by only 13%. Vega performance, as we found out earlier, largely determines the card's power limit, and for the Radeon RX Vega 56 it is reduced by 29% compared to Vega 64 (from 295 to 210 W), so GPU clock frequencies, according to the table data, are not affected so much, in reality they will be noticeably lower. In addition, the clock frequency of the HBM2 RAM is reduced by 15%. Fortunately, AMD has kept the back-end of the chip intact - a set of memory controllers and ROPs.

The younger “Vega” is priced at $399, which corresponds to the cost of the GeForce GTX 1070 Founders Edition. By the way, the recommended price for non-reference versions of the GTX 1070 is lower - $349, but in reality, products from NVIDIA's original design partners cannot now be purchased for less than $400, so the positioning of the RX Vega 56 is quite fair, provided that AMD provides performance equal to the GTX 1070.

Alas, Vega 56 did not escape the fate of its older sister: the shortage drove the accelerator’s foreign exchange prices beyond $500, and in Russia offers start at 39 thousand rubles. For comparison: the average retail price of the simplest versions of the GTX 1070 in Moscow online stores is about 31 thousand rubles. In such a situation, the Radeon RX Vega 56 will have to demand not just equality, but a confident victory over the GeForce GTX 1070.

Design

The physical design of Vega 56 is no different from the older model in a configuration with an air cooling system. This will be the face of the Vega family until the first cards of the original design appear (which is not expected until November).

Both accelerators have a very advanced design by the standards of reference samples - in terms of both heat removal and power supply. The cooling system consists of a massive radiator based on an evaporation chamber (which is a necessary solution for top-end video cards with a closed casing) and a centrifugal fan. Taking into account the bitter lessons of the Radeon R9 290X, in the new generation AMD increased the diameter of the impeller, thereby reducing the rotation speed and noise level. Due to the fact that only modern compact HDMI and DisplayPort interfaces are located on the board, most of the area of ​​the mounting plate is occupied by the grille - this is the only way for hot air to leave the video card case.

Installing and removing a cooler on video cards with HBM memory is a delicate procedure in itself, and in some Vega 56 samples, the GPU and memory crystals also have different heights, so, to be safe, we did not disassemble the Vega. Fortunately, this won’t hurt to talk about board elements that deserve attention, based on data collected by more decisive colleagues. All products in the Vega family use the same printed circuit board, designed to handle much more serious power (up to 345 W in the Radeon RX Vega 64 Liquid Cooled Edition) than the 210 W allowed by the Vega 56. For this, two eight-pin power connectors are used (together with the PCI Express slot, providing a reserve of 375 W) and a twelve-phase voltage converter. The load on the VRM is indicated by a diode track near the connectors, glowing red or blue depending on the position of the miniature switch.

A distinctive feature of Sapphire boards, which has been producing reference samples for ATI and AMD since time immemorial, is two BIOS chips, switching between which is performed by a lever at the end of the board.

Graphics card design

The design of the video card is already familiar to us from previous reviews. It is worth noting that only the regular series was in our hands, but there are also limited editions with a steel-colored casing, as well as using a closed liquid cooling system instead of the turbine usual for reference cards.

In your hands, the video card feels like a single whole; the casing makes the card look massive. The front part is distinguished only by the RADEON inscription and the fan.

The reverse side is covered with a metal protective plate that feels like it can withstand a hammer blow.

Another inscription RADEON, located on the end of the video card, has a red backlight. Nothing superfluous, only two 8-pin additional power connectors are noticeable.

The connector latches themselves are located inside the casing, but this does not create any problems when connecting and disconnecting the power cable. Above the connectors are GPUTach LEDs, which indicate the loading of the GPU (8 LEDs) and the operating mode (far left).

A small switch is located closer to the video output panel of the video card.

This is a BIOS switch that has two positions - the main BIOS, which can be replaced with your version (left position), as well as a backup BIOS version with reduced consumption limits and write protection (right position).

⇡#Performance: 3DMark

3DMark shows AMD processors at their best. Just as the Radeon RX Vega 64 is confidently ahead of the GeForce GTX 1080 in “synthetics,” the Vega 56 turned out to be 14% faster than the GTX 1070. The advantage of the new product over the flagships of the previous generation (Radeon R9 Fury X and GeForce GTX TITAN X) was 23 and 19 % respectively.

Even at 1080p and 1440p screen resolutions, where the AMD GPU's potential is not fully realized, the Radeon RX Vega 56 outperforms the GeForce GTX 1070 in terms of average frame rates. Moreover, the AMD accelerator holds a confident lead in most tests, and only in two games (Crysis 3 and GTA V) does the GTX 1070 unconditionally dominate. As we remember from previous reviews, there were more games that were inconvenient for the Vega 64 in the fight against the GTX 1080. Vega 64 itself provides 7–9% higher frame rates compared to the younger model.

Vega 56 outshone the previous generation flagship cards (GeForce GTX TITAN X and Radeon R9 Fury X) with an advantage of 10–11% and 16–18%, respectively.

⇡#Performance: games (3840 × 2160)

Most games at 2160p resolution are too much for the Radeon RX Vega 56, but in this mode the accelerator works at its most efficient. The GeForce GTX 1070 can expect a tie in some tests, but the average frame rate gives the Vega 56 an 8% advantage.

The gap between Vega 56 and the former champions has also increased: the younger Vega is ahead of TITAN X by 12%, and Fury X by 30% due to twice the amount of RAM. 4GB of HBM memory was still sufficient for the Fiji GPU two years ago, but today resource-intensive modes like 2160p are no longer available on the Fury X.

Where can I buy?

The Sapphire Radeon Pulse RX VEGA 56 8GB video card (art. no.: 11276-02-40G) [1.32kg] can be purchased on computeruniverse for €307.01 (22,500 rubles). In Russia, this model is very rare, but if it appears in some online computer store, it usually costs at least 27 thousand rubles. It is more profitable to buy a video card on computeruniverse using the discount coupon FWANUVT (How to use?). It is better to pay for your order with a Kukuruza card so that the payment takes place at the most favorable rate - the rate of the Central Bank of Russia.

⇡#Performance: video decoding

In the task of decoding H.264 and H.265 (HEVC) streams, Vega family accelerators are almost equivalent, adjusted for the lower frequencies of Vega 56, and provide leading performance. However, in HEVC, the multimedia block of Pascal chips is almost twice as fast as AMD's.

⇡#Performance: Computing

Some of the computing tasks don't differentiate between the Vega 56 and Vega 64 (like particle physics simulation in CompuBench CL), but in most cases the older Vega is noticeably superior to the smaller model. However, the Vega 56 remains one of the best general purpose compute accelerators on the consumer market. According to the totality of tests, Vega 56 leaves no chance for the GeForce GTX 1080 and successfully competes even with the GeForce GTX 1080 Ti.

Zotac GeForce GTX 1070 Ti AMP! Extreme Edition

The best value for money is with the competing model from nVidia. Its price is about 35,000 rubles, respectively, ZOTAC GeForce GTX 1070 AMP! Extreme 8GB is cheaper by as much as 14,000 rubles. It consumes less energy with the same performance, but is a little noisier.

⇡#Clock speeds, power consumption, temperature, overclocking

The WattMan section in the Radeon RX Vega driver settings contains three preset power profiles - Power Save, Balanced and Turbo, with which you can either reduce power consumption or increase it relative to the nominal 100%. For Vega 64 we have specific profile data: Power Save reduces the power limit by 25%, and Turbo increases it by 15%. Plus there is a BIOS backup where each profile is trimmed by 7-10W. In the case of Vega 56, the effect of this option will have to be determined experimentally.

In Turbo mode, which we chose for the main performance tests of the Vega 56, the power consumption of the card is indeed much less than that of the Vega 64 (as can be seen from measurements in FurMark), but in terms of average system power in gaming applications, both Vegas compete with the GeForce GTX 1080 Ti. And even the most stringent restrictions imposed by the Power Save mode cannot bring the Vega 56 closer to its main competitor, the GeForce GTX 1070.

The maximum core frequency of the Vega 56 is 1590 MHz - much higher than the 1471 MHz indicated in the table. Let us recall that the “upper” frequency in the Vega specifications, unlike previous AMD accelerators, means the maximum that the GPU is guaranteed to be able to handle under a standard load, but not the maximum and, on the other hand, not a guaranteed stable frequency. Judging by measurements in Crysis 3, even in Turbo mode, the GPU frequency in Vega 56 fluctuates around 1431 MHz, although Vega 64 produces a hundred more in the same conditions. In power saving modes, the Vega 56 is forced to reset frequencies to 1341 and 1295 MHz, respectively, but the performance per watt increases: Power Save mode reduces frequencies by 10% and the overall system power by 14%.

The upper steps of the clock frequency at standard settings correspond to the maximum GPU supply voltage - 1.2 V, but considering that Vega 64 is enough to stably maintain a frequency of 1630 MHz, one could expect comparable results in overclocking from the younger Vega. However, even with the greatest power reserve (standard firmware allows you to increase it by 50%), the stable frequency of the Vega 56 GPU is locked in the range of 1513–1515 MHz, and any manipulations with the peak frequency in the WattMan settings do not make sense.

One way to correct the situation is to flash the BIOS from Vega 64. This way, GPU computing units blocked at production cannot be returned to operation, but at least the power limit will no longer affect the frequencies. Satisfactory results can be achieved by simply reducing the supply voltage to the GPU. Our sample, at a voltage as low as 1.05 V, allowed us to set the maximum frequency to 1612 MHz. At the same time, in the Crysis 3 test, the frequency reached 1591 MHz and stabilized at around 1575 MHz.

By the way, undervolting Vega 64, in our experience, only makes sense in order to calm the card's energy appetite and reduce the load on the cooling system. Even at a nominal voltage of 1.2 V, the Vega 64 core practically does not overclock above the default frequency of 1630 MHz. The road to higher frequencies will only be opened by a BIOS modification or future boards of original design, which will allow higher voltages to be supplied to the GPU.

There is a reason why the HBM2 memory in Vega 56 operates at lower frequencies relative to Vega 64. Apparently, our sample uses chips from a different manufacturer (AMD orders them simultaneously from SK hynix and Samsung) or simply lower-quality assemblies. If in Vega 64 the memory was easily overclocked from the standard 900 to 1095 MHz, then in Vega 56 stable operation is impossible at frequencies above 950 MHz.

As a result of overclocking, the system's gaming power increased by 52 W compared to Turbo mode, but as we learned with the Vega 64 example, without reducing the voltage, everything would be even worse. Now, at least, we see a standard picture, when a fan operating at maximum speed (4647 rpm) reduced the temperature of the overclocked GPU by 17 °C compared to the temperature at standard settings, which means that a compromise can be found in everyday use between heat and CO noise without sacrificing clock speeds.

Note: here and below in the diagrams and tables for overclocked Vega cards the maximum GPU frequency is indicated, for non-overclocked cards the Boost Clock frequency is indicated according to the specifications.

Lots of settings to optimize performance and high clock speed

In addition to the "Efficiency BIOS" setting, which has lower power consumption and slower fan speeds, there are also "Standard" and "Nitro+-Boost BIOS" settings.

At its highest performance level, the graphics card reaches clock speeds of up to 1,592 MHz, which is 121 MHz higher than the reference card (its maximum of 1,471 MHz).

However, at such values, energy consumption increases significantly. At full load, our test system consumes 415 watts, the second-highest result on our leaderboard. And this card’s capabilities have not yet been exhausted! The WattMan tool makes it possible to force the video card to squeeze out even more performance.

⇡#Overclocking performance: 3DMark

Since AMD's architecture demonstrates enviable efficiency in 3DMark, it is not surprising that even an overclocked GeForce GTX 1070 cannot compare with the Vega 56 running at stock frequencies. Vega 56 itself is practically equivalent to Vega 64 when overclocked, but the latter, due to its significantly higher power reserve, allows you to increase the frame rate by another 10%.

Benchmark tests

3DMark Cloud Gate GPU: 125359

Better than NVIDIA GeForce RTX 2060, but worse than AMD Radeon RX Vega 64.

3DMark Cloud Gate Standard Score: 47018

Better than NVIDIA GeForce RTX 2070 Super Mobile, but worse than NVIDIA GeForce RTX 2070 Super Max-Q.

3DMark Fire Strike Graphics: 20759

Better than NVIDIA GeForce RTX 2080 Max-Q, but worse than NVIDIA GeForce GTX 1070 SLI.

3DMark Fire Strike Score: 17098

Better than NVIDIA Quadro RTX 4000, but worse than NVIDIA GeForce GTX 1080 Mobile.

3DMark Ice Storm GPU: 412820

Better than NVIDIA GeForce RTX 2060 Mobile, but worse than NVIDIA GeForce RTX 2070 Max-Q.

3DMark Time Spy Graphics: 6482

Better than NVIDIA Quadro P5200, but worse than NVIDIA GeForce GTX 1070 Ti.

3DMark Time Spy Score: 6746

Better than NVIDIA Quadro RTX 4000 Max-Q, but worse than NVIDIA GeForce GTX 1080.

3DMark Vantage P: 54586

Better than NVIDIA Quadro RTX 5000, but worse than AMD Radeon RX Vega 64.

3DMark06: 36286

Better than AMD Radeon RX Vega 64, but worse than AMD Radeon RX 5700.

3DMark11P: 22937

Better than AMD Radeon RX Vega 64, but worse than NVIDIA GeForce RTX 2060.

3DMark11P GPU: 29086

Better than NVIDIA GeForce GTX 1080 Mobile, but worse than NVIDIA GeForce RTX 2070 Super Mobile.

ComputeMark v2.1 Normal, Score: 17030

Better than AMD Radeon RX 6600, but worse than NVIDIA GeForce RTX 2080 Super Mobile.

LuxMark v2.0 64Bit Sala GPUs-only: 5459

Better than AMD Radeon RX 5700, but worse than NVIDIA GeForce RTX 2060.

Passmark: 13579

Better than NVIDIA GeForce GTX 1070, but worse than NVIDIA GeForce RTX 2080 Super Max-Q.

SPECviewperf 12 - 3ds Max: 145

Better than AMD Radeon RX Vega 64, but worse than NVIDIA GeForce GTX 1660 Ti Mobile.

SPECviewperf 12 - Catia: 141

Better than AMD Radeon RX 5700, but worse than NVIDIA Quadro P5000.

Unigine Valley 1.0 Extreme HD DirectX: 76.4

Better than NVIDIA Quadro P5000, but worse than NVIDIA GeForce GTX 1660 Ti.

⇡#Overclocked performance: games (1920 × 1080, 2560 × 1440)

Overclocking the Vega 56 increases the average frame rate in games by 11-14%, depending on the resolution, and in some tests the effect reaches 24%. The overclocked Vega 64 is easily ahead of the Vega 64 running at stock frequencies, and when both cards are overclocked, the difference comes down to 5–7% in favor of the Vega 64.

Comparing the Vega 56 and GeForce GTX 1070 in overclocking gives the AMD card a 6-10% advantage, and even the reference GTX 1080 is only 5-6% ahead. However, the overclocked GTX 1080, in turn, leaves no chance for any of its rivals.

⇡#Overclocked performance: games (3840 × 2160)

At 2160p, overclocking the Vega 56 is just as useful (13% average frame rate) as it is in less resource-intensive modes. Moreover, the younger Vega overclocked here is 12% superior to the overclocked GTX 1070 and only 3% inferior to the reference GeForce GTX 1080. The overclocked Vega 64 and GTX 1080 also differ by 3% in favor of the latter.

⇡#Conclusions

The Radeon RX Vega 64 was unable to unconditionally beat its main rival, the GeForce GTX 1080, and only at the cost of incomparably higher power consumption did AMD achieve the level of gaming performance that NVIDIA conquered more than a year ago. But Vega 56 hit the target. The younger model in terms of gaming performance won a difficult victory over the GeForce GTX 1070 in the key resolutions for this category of accelerators 1080p and 1440p, and is confidently in the lead at 2160p.

Another advantage of Vega 56 over its main rival is its overclocking potential. The GTX 1070, as a rule, cannot be overclocked to the level of the GTX 1080. NVIDIA has separated the two models both by the type of RAM and by selecting the most promising GPU crystals for the GTX 1080. At the same time, the Vega 56 can easily reach the level of overclocking Vega 64 is getting menacingly close to the GeForce GTX 1080. All this partly justifies the actual prices of the new product, which significantly exceed the amount recommended by the manufacturer due to the general shortage of Vega cards. If the Vega 56 weren't so overpriced right now, we wouldn't hesitate to give it our Best Buy award.

How to increase Vega 56 hashrate using overclocking

The Vega 64 hashrate may be much higher than stated. It is better to overclock the GPU for mining using MSI Afterburner. The utility interface is very simple. To overclock the video card, we need two sliders:

• Core Clock – core overclocking, increases the core frequency;
• Memory Clock – memory overclocking, increases the memory frequency.

You need to change the frequency values ​​little by little. Then test stability. If all is well, then you can increase the frequency further. If the video card starts to fail or the drivers crash, then simply reduce the frequency and that’s it. There will be no harm.

Optimal overclocking parameters: