The wait has finally come to an end: AMD Radeon RX Vega64 in test

Already at the end of last year we were able to marvel at the Vega64 in its first form and since then the manufacturer (and thus also the readers, as well as the potential customers) has fed us again and again with small bits of information and announcements about the n... Also at the risk of us repeating ourselves, we want to go back to the most important innovations at Vega. With the Vega64, AMD is now offering a new generation of GPUs after the Vega Frontier Edition, in which more than 200 changes are... 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 Vega64 and Vega Frontier Edition have the same board, 100% identical components and differ only by the soldered package with the halved memory expansion, as well as a customized firmware. The la... 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) Introduction Actually, we wanted to remove this game from the benchmark suite for good this time, because the quality of the DirectX12 implementation is not something you really want for yourself and the players. Nevertheless, the results were, so noticeable... 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) Good question, short answer: Jain. The latest version of Claymore's Dual Ethereum AMD/Nvidia GPU Miner (v9.8) includes support for Radeon RX Vega, so we used it for our mining benchmark. All AMD cards run in the so-called... Preliminary remark on the three power modes In the Wattman, in addition to the standard setting (ex works), which is called Balanced Mode, you will also find the possibility to save energy with the Power Save Mode or, conversely, to use the Turbo Mode to... Temperature curve and clock rate The fan control is quite conservative, so that the maximum temperature of 85°C is reached relatively quickly. But then the card already has approx. 6% of their performance lost from the cold state, which is almost out of the... Better late than never and yes, we are really glad that this chapter of endless salami-slice-hold tactics is finally over. But you could have saved the whole media circus in advance, because with a slightly more relaxed...

Board layout

The Vega64 and Vega Frontier Edition have the same board, 100% identical components and differ only by the soldered package with the halved memory expansion, as well as a customized firmware. The length, if you look at the right side of the PCB, is only due to the radiator structure with the large radial fan. By a sensible shortening, it would be easy to achieve board sizes in the format of an R9 Nano.

AMD has therefore thought a bit about the division of the board, especially since the elimination of the external memory modules opens up new possibilities. Exactly in their place you now place the individual power supplies. We see an almost classic 6+1 phase design for GPU and memory, as well as other voltage converters for the respective partial voltages. We will go into all these things in more detail, because it is really interesting.

The two 8-pin connectors are each connected via a coil, which helps to smooth certain voltage peaks. However, we could not find larger capacitors here.

On the back, besides the very tightly equipped base of the package, we see the PWM controller as the most obvious part, as well as other components whose positioning we should remember, because we want to show a video right away.

Power supply of the GPU

First of all, the focus is on the IR3521 from International Rectifier, a dual output digital multi-phase controller that can provide both the 6 phases for the GPU and 2 other phases, which we will talk about in a moment. But back to the GPU. After all, we count twelve voltage transformer circuits, not six. However, since only six real phases can be created, you can double each of these phases in order to divide the load into two converter circles. We suspect that this effort will not be done later with the slightly simpler XL version. for this, you either simply owe away the space for the second circle or simplify the board as such.

A total of six IR3598s are used for this so-called doubling, which are located on the back of the board (we remember). The actual voltage conversion of each of the twelve converter circuits is performed by an IRF6811 on the high side and an IRF6894 on the low-side, which also contains the required Schottky diode. Both are previously USED by AMD HEXFETs from International Rectifier. In the case of coils, AMD relies on encapsulated and front-soldered ferrite core coils. The inductance is slightly lower at 190 nH than the usual 220 nH for THE GPUs.

Power supply of the memory

As mentioned briefly, the IR35217 also provides one phase for storage. One phase is sufficient because the memory is much more adequate. The gate-driver CHL815 is back on the back of the board, while an NTMFD 4C85M from ON Semiconductor is used for voltage conversion. This dual N-Channel MOSFET realizes both the high and the low-side.

It is also interesting that AMD generally dispenses with all cup capacitors and only relies on flat SMD caps. The slightly lower capacity can be compensated by simply switching two of these caps in parallel and usually also using the back of the board. A sensible equalization of the thermal hotspots and a larger heat dissipation also have the nice side effect that you can also enter a temperature class lower and thus also save costs. The finance department is certainly equally pleased with this.

The coil is slightly larger this time with 220 nH. At 820 nH, on the other hand, the coils for the significantly slower clocking converters of the other partial voltages are even larger, which, however, also have to withstand significantly lower currents.

Other voltage converters

The production of VDDCI is not a big item in terms of performance, but it is important. It is used for GPU-internal level transition between the GPU and memory signals, something like the voltage between the memory and the GPU core on the I/O bus. In addition, two constant sources for 1.8V and 0.8 volts are produced.

Below the GPU you can also find the APL5620 from Anpec for the VPP. This ultra-low dropout chip generates the very low voltage for the Phase Locked Loop (PLL) range.

With the HC238A from ON Semiconductor as a de-multiplexer, the "mausekino" is realized for the LED bar, which shows the utilization of the power supply. Nice gimmick, but in intensity almost disturbing. Especially at night.