Well since my last tech post a number of interesting things have occurred within the graphics card tech industry. AMD have released their RX480 both the 4gb and 8gb version with a little controversy about power draw through the PCI slot. Nvidia returned fire with the GTX 1060 and the announcement of a new flagship card the GTX Titan X. Though this might not be the flagship card for all that long the GTX Titan X might be usurped by a Ti version early 2017. So lets dive into the deep end…
Sooo the RX 480, comes in 2 flavours the 4gb and 8gb versions priced at around £180 and £230 respectively with a reference cooler. This price point is the sweet spot for a gaming according to steam statistics. The GTX 1060 is just a little bit more expensive and im not really going to focus on this card.
There was a lot of hype surrounding the release of this card (I’m looking at you reddit!) So much hype, which hadn’t come from AMD and was purely speculation. I felt that the card was unfairly compared to the GTX 1080, people seemed to think a $200 dollar card could compare with a card costing multiple times that amount. People and media forgot that all though the cards were released close together they where aiming at different ends of the market. The 480 targeted at gamers and the 1080 aimed at enthusiasts. I’m not saying that enthusiasts arent gamers but they are something above that level of commitment. Another disappointment for many was that the 480 only iterated on the GCN architecture and most of the obvious improvement came from the shrink of the transistor size which related to more energy efficiency and cooler core temperatures.
At the moment there are certain features within DX12 that nvidia cards don’t take full advantage due to the architecture and use a method of brute force and ignorance to get the high clock speeds! Early DX12 builds of ashes of the singularity where showing AMD’s GCN architecture having a distinct advantage over the Nvidia cards. When Vulkan drivers for Doom were released a major step up was made by the Red team. With the Bench marks for DX 11 and DX12 you see individual games favour one brand of card over another. which is not something is likely to change as each company wants their logo on the top-selling games to increase brand awareness. Not that there is much competition!
Reference coolers are not known for their cooling efficiency or noise levels (well only in a negative way!) the RX 480 is a ‘blower’ style card so it has a single fan that directs air across the whole board and is exhausted out of the grill on the back. It is a single fan design and the major design feature is that the hot air from the card is exhausted out of the case unlike other designs where the hot air is circulated into the case and relies on the case fans to remove the hot and replace it with cool air, which is then blown across the heat sinks of both the GPU or CPU depending on the setup. After market cards that are more expensive than reference cards are aimed at either noise levels to make the card as quiet as possible or for maximum cooling or at a spot in-between. With improved cooling their opens up the possibility of overclocking the card by increasing the speed that the chip processes information by increasing the clock speed which also means you have to increase the voltage going through the chip which increases the temperature of the chip, memory and other components. All chips are not created equal in both the GPU and CPU markets some chips will just have a larger headroom for increasing speed of the chip. as an example lets say a chip has 4096 ‘whatmaflips’, these whatmaflips are batched together in groups of 256 giving 16 ‘bobbins’ per chip. If during the process of making a chip one whatmaflips doesn’t work a whole bobbin does not work and is generally switched off. That is why when AMD and nVidia release a group of graphics cards the price for a top end card is expensive because there are no errors on the chip and the yield/availability is low, the lower end cards have errors on the chip and have had there bobbins switched off! So a top end expensive card will feature the full 16 bobbins the mid range card might have 12 bobbins and the entry-level card might just have 8 bobbins. But the chip is exactly the same just with bobbins switched off and the chip been exactly the same. There will be fewer chips on a silicon wafer with no errors at the start of the process but as the silicon manufacturer gain experience at working at a certain level yields of 16 bobbins per chip will improve. Bringing down the price of the top end cards. So although the chip is the same size there are fewer bits of silicon actually working so heat efficiency will actually improve because of the easier dissipation. Occasionally during the process of manufacture there will be a significant change in the process which could improve yield or heat efficiency, these are called ‘stepping.’ For example Intels Q6600 ‘G0’ stepping became famous in overclocking circles from changing a good chip into a beast by allowing massive overclocking ability. The G0 stepping improved both thermal specification and power consumption as well. Certain graphics card manufacturers will ‘bin’ the top end cards. Which means they reserve these cards and make special cards with them that are massively overclocked. It can be hard to spot them because each manufacturer uses different naming protocols. for example you will get OC (overclocked), SC (super clocked), SSC (super super clocked) etc. Overclocking though not encouraged by Nvidia and AMD has the tools within software to improve performance of your card for no extra cost. AMD changed the name of their software from Overdrive (which I liked) to ‘Wattman’ that in my head is a cartoon character very much like Duff Man from the Simpson’s.
As graphics cards have become more powerful they have needed more power than the PCi slot could supply so they have drawn more power directly from the PSU via 4,6 or 8 connectors. The major specification of PCi slots allows 75 watts of power to be drawn, but the PCi specification is rather complicated and does allow for more power to be drawn from this source. It was found that the RX 480 drew more than 75 wats from the motherboard and the back lash from a lot of the internet/youtube press. When overclocking a graphics card the power can come from 2 locations the PCi slot or the PSU. As overclocking is not part of the manufactures warranty you do it at your own risk. Many graphics cards that are overclocked unless they are well below the power draw of the combined rating of the PCie slot and the PSU cable are likely to draw the extra power though the motherboard. AMD has worked on this bad press and released a software patch that alters the power draw, but it seemed odd for them to receive so much bad press for something that happens when you over clock any card.
When the RX480 has been overclocked it has been found that as with other cards in the GCN line of chips it does not have much room for overclocking, it is no GT8800. The low overclock capabilities seems to be a major contention within the Youtube tech show brigade. Which for their audience will both be gamers and enthusiasts is understandable or maybe it’s gamers who wish they could be enthusiasts.
Over the next month the after market manufacturers will get more custom coolers out to retailers which will mean hopefully more interesting cooling options available. Waterblocks for the custom cards were announced as the RX 480 was released. But maybe a RX 480 with a vapour chamber cooler will be nice and quiet which will get the noise/performance/price sweet spot.
The RX 480 is in the correct price bracket and does do amazing things at that price point it delivers in 1080p gaming and can be pushed to 1440p with 8gbs but is no 4k player. Even the mighty GTX 1080 struggles at 4k at present there is no single card that is able to deliver a 4k gaming experience at ultra settings some compromise has to be made. The RX 480 is a small card the and could see a place in many mini-itx builds in small form factor cases.
With nVidia announcing the GTX Titan X for a price that might make you take out a second mortgage on your house and maybe a Ti version in the new year. Intel releases Broadwell-E chips at the the same price as a GTX Titan X! The Zen chipset has had more information ‘leaked’ the Vega graphics chipset which is going to be AMD’s high-end range not coming out to Q1 2017 it is giving nVidia all the market share at the moment in the mid-top end but will there be a backlash due to the price that nVidia are charging for these top end products. Will AMD follow suit and try to price gouge or will they go for a cheaper option with slightly less performance.
At the moment if I had £300 to spend on a graphics card I would buy the RX480 with a none reference cooler. Because I game at 1080p and I could spend the spare change on something else shiny! AMD didn’t deliver the giant killer that people hoped for, what they did deliver was a great card like the ancient but at the time great HD4870.
I would say console gamers have it easy not having to worry about upgrading their gaming rig to keep up with all the shiney graphics but with the annoucement of the NEO and Scorpio both Sony and Microsoft have muddied the water there as well. Well done!
See you on the other side…