AMD Radeon RX 590 benchmark, review, features and performance

AMD Radeon RX 590 benchmark

AMD has moved again in the GPU market and today, it has released AMD RX 590 GPU.  Although it is not the movement that we would all like, if it is a movement that provides security, it is evolutionary in nature and increases performance. Its Polaris 30 architecture offers a performance bonus at a price similar to that seen by Polaris 20 thanks to the new RX 590 cards.

AMD Radeon RX 590 benchmark, review, features and performance

Back in 2016 AMD launched what would be the most “exploited” architecture in its history. Not only in longevity, but in the number of total models available, where it has been able to cover 3 generations with slight tweaks that have been boosting performance generation after generation.

As we say, since the RX 480 practically nothing has changed, so, we found a GCP graphic layout with the same distribution and components within the die in these RX 590:

In it we find 2304 shaders, 32 ROPs, 1 GCP, 4 ACE, 36 CUs, 4 geometry processors, 144 textures units, 2 MB L2, a 256-bit bus, 8 GB GDDR5 and a long etc. that is shared from Polaris 10 up to Polaris 30.

Therefore, the way of rendering is exactly the same: a command processor distributes the workloads among four Shader motors, which distribute the load to a geometry processor, rasterizer and 8 ROP, to finally perform the hard work on the 9 computing units (CU).

The computing unit (CU) is the indivisible subunit of the “Polaris” architecture and performs most of the processing on the GPU. Each CU contains 64 flow processors, sharing them with small data caches, a programmer, a scalar unit and four TMUs.

But then where are the improvements ? Everything revolves around the “new” manufacturing process for these chips, where Globalfoundries with its 12 nm LP process makes the difference.

Compared with the 14 nm LPP the 12 LP transistors improved and achieved a CA performance around 15% at the same DDQ level. All with comparable levels of production and above all reliability, but better at 14 nm.

As far as energy is concerned, the 12 nm contain new libraries of seven and a half tracks (compared to 9 of the 14 nm LPP) where, for identical frequencies, a reduction of the power consumed by up to 16% is achieved.

The heights of the cells have also been improved, since GF has gone from 576 mm in its 14 nm LPP to 480 mm in its 12 nm LP, but interestingly it has not modified the End Pitch, the Poly Pitch and the Metal 2 which are maintained at 48 nm, 84 nm and 64 nm.

This is how we say, curious, since both Samsung and Intel or TSMC vary these distances in their improvements in the processes of lithography. Without going any further, Intel in its “new” 14 nm +++ varied the Fin Pitch to achieve an improvement in efficiency by requiring less power between transistors, optimizing their current flow.

The problem with not doing this is that you obviously optimize the use of energy worse, which AMD needs to compete with NVIDIA in this field, but it seems that Lisa’s guys do not care about this factor as much as the performance gain. what that node can contribute

Thus, despite physically containing the same chip only in a smaller manufacturing size and being more efficient at the same frequency, AMD has opted as we have said above for the performance factor. Therefore, it has raised the frequencies up to 1545 MHz with boost (reference GPUs) from 1266 MHz with boost that the RX 480 has, so this RX 590 manages to raise the bar in this section by 22%.

With this increase comes the rise in consumption, where we went from 150 watts in the RX 480, to 185 watts in the RX 580 to 225 watts of these new RX 590, which implies a 50% increase in consumption since Polaris 10 up to Polaris 30.

As if this were not enough, since at the moment the figures do not accompany, we will only have available at the moment and today 4 different models of 4 different assemblers.

For this short availability of models is not so hard, AMD has made available to buyers of any of the 3 models a launch Bundle with three games:  Resident Evil 2 (2019), The Division 2 and Devil May Cry 5.

And here’s the news of these RX 590 that AMD has officially launched today. But what about performance and consumption?

Well, considering that AMD has not contributed a reference card to any media in the world, all the available reviews have then custom models for their revisions (which we hope to have from the assemblers when they are available in our country).

Therefore the data of performance and consumption are a little high in both cases, since the custom models include higher speeds and therefore higher consumption.

Under 1080p its relative performance positions it 10% above its direct rival in the market, since at the moment NVIDIA has not made available to the global public its GTX 1060 6 GB with GDDR5X nor the supposed RTX 2060.

Also, this RX 590 gets 11% off its predecessor RX 580. This shows the “bad” performance scalability Polaris architecture, because with a 17.9% rise in clocks in the XFX Fatboy model left almost a 8% down the road.

In 2K the gap is reduced with the RX 580 but is enlarged against the GTX 1060 6 GB, where the roles between these two are reversed and get to be two figures of the new RX 590. In turn, this really stays close to leading models of NVIDIA in previous generations, where the 980 Ti barely surpasses it by 5%.

In 4K the RX 580 maintains the distance, but the GTX 1060 6 GB goes to 13% below the RX 590, product of its lower bus and bandwidth. In contrast the GTX 980 Ti manages to raise its difference by 1% despite having more muscle than the AMD card.

If we focus on consumption, the data is much worse than in the previous generation but very scalable in terms of the MHz / watt ratio. With 232 watts of average consumption the difference with the RX 580 grows up to 17.1%, while as we have said the clocks have increased by 17.9%.

The problem is that we start from a smaller node where the improvement of the wafer manufacturer (GF) is 16% at the same clocks. This only shows that the architecture as such is scalable in said ratio, but it is very limited in terms of frequencies for said ratio.

These RX 590 have a PVP set by AMD from $ 279, but the reality is that by not having reference models the future buyer is forced to acquire a custom model.

This has an impact on better performance, lower temperature and better targeta in general, but also increases the cost of this, so the comparison in this case is not fair.

Even not having performance benchmarks to compare the two reference cards, if we know that the price of which has been increased one 21.8% if we look at the data provided both at its output by AMD.

This shows a worse performance / watt ratio seen when seen, where a custom card and higher range of available barely gets off 10%, but its price as we say has risen 21% to the same range.

This curiously is not fulfilled in the custom models, where we can currently find models like the Sapphire cheaper in its RX 590 version than in its RX 580 version.

Curious fact without a doubt, so if we want to get one of these small should look well at current offers and especially the stock, since it is limited and usually in these cases tends to raise the price due to lack of availability.

Related topics