We try and get our heads around the complexities of the GPU
A fair few years ago, we were spoiled for choice when it came to picking a graphics card. Should we buy an ATI card, Voodoo Banshee, Matrox Millennium or a Savage S3? It could lead to some interesting, if a little confusing discussions.
These days there are the two main players left: AMD (formally ATI) and Nvidia. But even though there are only two manufacturers now, it hasn’t got any less confusing. Where we used to simply look for the make and the amount of memory the card had, now there’s the GPU clock frequency, GPU shader clock, number of shaders and cores, memory type, memory bandwidth and TDP to take into consideration. And all that is before we even look to see if our case is big enough to fit the card in the first place.
The technical specification of a modern graphics card often give experts a headache, so for the layperson, who simply wants a card that will fit in their PC and run Assassin’s Creed: Syndicate without too much of a problem, the long list of tiresome numbers can be quite bewildering.
Factor into all that what exactly a budget, mid-range and high-end card actually is and what it’s capable of, and you can hardly blame people for simply turning to a console when it comes to gaming.
That said, we thought we’d try and put together something that can help to explain the various jargon associated with graphics cards. We’ll look at some examples of budget, mid-range and highend cards, and we’ll see if we can come up with some examples of what sort of cards you’ll need for particular situations. All being well, you should have enough knowledge to be little less confused the next time you decide to shop for a graphics card.
Budget Cards
We’ll start off with some examples of the price ranges, beginning of course with the budget range of graphics cards.
A budget graphics card is a fine balance of performance and cost; a little too much or too little either way, and you’re either out of your price range or you’ve got something that’s really not worth bothering with.
Nvidia
In our opinion, the best budget graphics card from the Nvidia camp is the GeForce GT 730. This is an exceptionally capable card, but before you go out and buy one, there are a couple of things we need to address first.
There are a number of GT 730 cards available. Some offer 1GB of GDDR3 memory, and others have 2GB; some are 128-bit, whereas others are 64-bit. They’re more or less the same, in that the GPU used is an upgraded GT 630 model with a better memory bandwidth, but the specific GPU to look out for is the GK208 Kepler.
This is a better processor, and you’ll usually find this GPU on the GDDR5 64-bit models, such as the Asus GT 730 2GB priced at around £58-60. This particular model has a 902MHz core clock, a memory clock speed of 1250MHz and 384 CUDA cores with a TDP of around 35W.
It’s a single-slot card but manages to feature VGA, DVI and HDMI ports, which is ample for most users. In terms of gaming power, the GT 730 should achieve a 3DMark score of around 1,540, which translates to running most of the top games from last year (the likes of Watch Dogs and so on) at 1280 x 720 on medium graphical settings. You’ll need to make sure the other system specs are up to scratch, of course, but the GT 730 2GB should be a good baseline Nvidia card to work from.
AMD
One of the best AMD budget cards you can get is the Asus R7 250X 1GB GDDR5, priced at around £67.
This Cape Verde GPU is an upgraded Radeon 7770 and has a 1000MHz GPU clock (over 5% more than the stock R7 250X) and a memory clock speed of 1125MHZ. There are 640 shader units and it has a TDP of around 80W.
It’s a dual-slot card and one that features DVI, HDMI and DisplayPort. As for gaming power, you should expect to see a 3DMark score of around 2,600 with this particular model.
Since it’s a tad more powerful than the GT 730, the R7 250X will be able to run the same titles as previously mentioned but probably at a higher resolution with the graphical details still set to medium. Or you could opt for a lower 720 resolution but increase the detail slightly. Either way, it’s an excellent budget gaming card.
Mid Range
The cost of a mid-range card varies depending on who you talk to. We usually look at the £70 to £160 as mid-range, but others tend to take the price range all the way up to £250.
A mid-range card may seem like a pretty safe place to be, but it all depends on which rung of the ladder in the mid-range you are. If you look at a card that’s toward the bottom end of the scale, then it won’t be long until it’s classed as budget and struggling with the current games.
On the other hand, a card that’s toward the top end of the mid-range scale will last you a lot longer. Taking that into consideration, we’ll start some way into the mid-range price bracket.
Nvidia – Lower End
Starting at £100, we have the PNY GTX 750Ti with 2GB of GDDR5 memory. This uses a GM107 GPU with a clock speed of 1020MHz, a boost clock of 1085MHz and a memory clock speed of 1350MHz. You’ll find 640 CUDA cores and a TDP of 62W on this dual-slot card, and as far connectivity goes, it has DVI, VGA and HDMI ports.
Performance is certainly good, with a 3DMark score of around 5,680. What this translates to is a surprisingly well-balanced gaming card. For example, we’ve had a GTX 750Ti 2GB with a 3.5GHz Intel i5-4690, 8GB of memory and Windows 10 running Assassin’s Creed: Syndicate on medium settings at 1920 x 100, and it was a playable 28fps with only a graphical hiccups here and there. Likewise, Fallout 4 on the same system, with medium settings and at 1920 x 1080, managed a pretty impressive average of 27fps.
Considering it’s only £100, it’s a good graphics card for the money.
AMD – Lower End
The MSI R7 370 Gaming 2G is our mid-range AMD card of choice and one we’ve reviewed and scored highly in the past. For around £116, this Trinidad GPU has a clock speed of 925MHz with a gaming boost of up to 1050MHz (over 6% faster than the stock R7 370 GPU speed) and a memory clock speed of 1400MHz.
It has 1,024 shader units and a TDP of 110W, much higher than the aforementioned GTX 750Ti. It’s a dual-slot card and features a pair of DVI ports, as well as HDMI and DisplayPort connections.
You can expect a 3DMark score of around 5,856 when running this card, which puts it slightly ahead of the GTX 750Ti, and will no doubt yield the same results, albeit with one or two extra in-game effects pushed to higher graphical setting.
Of course, it depends on whether you think the extra £16 is worth the 200 3DMark points or not.
Nvidia – Higher End
Toward the end of the mid-range scale, the Gigabyte GTX 960 OC 4GB GDDR5 graphics card makes a pretty good choice at £160.
This GM206 GPU has a base clock speed of 1216MHz, which is more than 7% faster than the stock GTX 960. It has a boost clock 1279MHz with a memory clock speed of 1753MHz. It also features 1,024 CUDA cores and a TDP of 120W, with connectivity in the form of one DVI, one HDMI and three DisplayPort connections.
In terms of the performance, you can expect this card to hit somewhere in the region of 8,500 in 3DMark, which means it’ll happily play Fallout 4 or Assassin’s Creed: Syndicate on high settings at 1920 x 1080 or slightly above.
AMD – Higher End
The MSI Radeon R9 380 Gaming 2G at £162 is the top of the mid-range scale for us and a very impressive graphics card.
The Antigua GPU has a base clock speed of 970MHz, moving up to 1000MHz in overclock mode, with a memory clock speed of 1425MHz.
It has 1,792 shader units and a huge TDP of 190W, with a pair of DVI ports, a single HDMI and a single DisplayPort for connectivity.
Although the clock speed is slightly slower than the GTX 960 above, the MSI Radeon R9 380 2G does a pretty good job when it comes to benchmarking. We’ve had a 3DMark score of 8,374 from this, along with an average fps of 39 in The Witcher 3 with high settings at 1920 x 1080.
High-end Cards
If you plan to get the best possible gaming experience at much higher resolutions, well into the 4K region, then you’re going to need to splash out for a high-end graphics card.
As you would expect, though, a high-end graphics card will cost anywhere from £160 to £270 and upwards. The top-of-therange card at the moment is the Nvidia GTX Titan X, a monster of a graphics card with 12GB of memory and a whopping 3,072 CUDA cores. In this case, the top-end price bracket starts at £1,050, but it’s a phenomenal graphics card if you have the system capable of handling it and you have pockets deep enough to enjoy it without having to quit eating to pay for it.
Thankfully, there are alternatives in the high-end range that are a little more affordable.
Nvidia
The MSI GTX980 Gaming 4G with 4GB of GDDR5 memory at £399 is an exceptionally impressive graphics card. The GM204 GPU has a base clock speed of 1127MHz and a boost clock of up to 1216MHz – more 8% faster than a stock GTX980. The memory clock speed is 1753MHz, and the card boasts 2,048 CUDA cores and a TDP of 165W.
Connectivity consists of a single DVI and HDMI port and three DisplayPorts on this dual-slot card.
As for the performance, we’ve seen this card hit a 3DMark score of 16,101, and as a result you can run the likes of Fallout 4 with everything graphical setting maxed out at 2K resolutions with the frame-rate hitting between 55 and 60 constantly. And with Star Wars: Battlefront at 1920 x 1080 with everything set to ultra, it played beautifully and looked spectacular.
If you can afford £400 for a graphics card, then you’ll be extremely happy with this.
AMD
The equivalent AMD graphics card would be the Gigabyte R9 Fury. Priced at around £440, this is as impressive a graphics card as the GTX980.
The Fiji GPU is clocked at 1010MHz, slightly faster than the stock R9 Fury, with a 1000MHz memory clock, 3,584 shader units and a TDP of an eyewatering 275W. And as with the GTX980, the connectivity consists of a single DVI, single HDMI and a trio of DisplayPorts.
As for benchmarking, we haven’t managed to get our hands on one of these yet for testing, but we have heard that some users are boasting a 3DMark score of over 18,300. If that’s the case, then you can expect the Gigabyte R9 Fury to run every game available at the highest possible resolutions your monitor can handle and with every graphical detail set to ultra.
What Do You Need?
The question of which one of these graphics you need is a difficult one to answer. After all, it depends on what you plan to do with your system. If you want a low-cost system that’s capable of some lightweight gaming, mainly with older titles from a year or so ago, and you want to watch some decent HD films, then you can settle for a budget graphics card.
If you want to play games and enjoy a good level of graphical detail while still keeping the frame-rate high and you want to have a good media system as well, then one of the lower-end mid-range cards will suffice.
For something with a little more performance, the higher end of the mid-range will cater for the latest games at good graphical levels with high frame-rates. It will also future proof you in terms of VR for next year.
If, however, you want to knock the socks off everyone else and dominate the gaming scene, then you’ll probably opt for one of the £400 graphics cards. Or if you happen to have a hefty income, then we’ll probably expect you to go for the Titan X range of ultra-high-end cards.
Those who simply want their computers to do day-to-day work, with maybe some video watching could get away with using the GPU that comes with a modern processor. In all honesty, while a dedicated graphics card will take some of the pressure off your system resources, if you’re only ever going to watch YouTube videos and look at Facebook, you can do that without even the budget card fitted.
As we said, though, it depends on your situation and what you plan to use the computer for.
Terminology
We’ve purposely left the terminology of graphics cards till last, purely because it can be quite a lengthy and confusing subject, and it’s the kind of thing that causes people to flick over the pages of a magazine without reading the content. We know – we do it ourselves sometimes.
The GPU
The GPU of a graphics card is its processor, in the same way that the Intel or AMD chip driving your computer is the processor. It might sound simplistic, but there’s a lot of confusion when it comes naming components.
For example, a GTX 970 isn’t a GPU; it’s actually the make of graphics card. The GPU on a GTX 970 is called a GM204, and it’s processed and made slightly differently to, say, the GPU on a GTX 960, which incidentally is called a GM206.
In these examples, a GM204 GPU has a process size of 28nm, 5,200 million transistors and a die size of 398mm2. The GM206, meanwhile, has a process size of 28nm, 2,940 million transistors and a die size of 228mm2. In terms of performance, the more transistors you can pack into that 28nm, the more powerful the GPU is likely to be.
However, other factors affect the performance, such as memory bandwidth and the speed and type of the memory, as well as the physical clock speed of the GPU – 1,050 on the GM204 versus 1,127 on the GM206.
You’ll also notice that Nvidia has a model code, if you will, for its GPUs, such as GM204, whereas AMD likes to give its card GPUs names, such as Fiji or Caicos.
GDDR Memory
GDDR memory is basically the same memory as you’ll find on your motherboard, but with different voltages and running at higher clock speeds.
The latest memory specification is GDDR5, which has a lower voltage and reduces the heat generated. Furthermore, GDDR5 allows for a better memory bandwidth, which will push the data through the GPU at a higher rate – in the simplest terms.
Basically, the more memory you have on your graphics card, the higher the resolutions you’ll be able to hit and the more details you’ll be able to push through at those resolutions.
Memory Clock Speed
GDDR5 memory runs from around 1000MHz up to 1500MHz; this is the actual speed of the video memory.
The memory can do four transfers of data every clock cycle, so often you’ll see memory clock speeds rated at 4000MHz, which isn’t the actual clock speed of the memory but rather the total data rate – at four times the real clock speed.
There are even times when you’ll see a memory clock speed of 2000MHz. This is just nonsense, as it’s only twice the clock rate and half the data rate.
Memory Bus
To make memory bus simple: there are several memory chips mounted on a graphics card, and a GPU can read 64 bits of memory at a time from a single memory chip.
To speed up the memory bandwidth, though, some GPUs can read from two memory chips at the same time, so that’s two times 64-bit, which equals 128-bit. That’s the memory bus. The memory bus increases all the way up to 4,096-bit in the case of a R9 Fury X, where it has eight lots of 512-bit memory controllers. Which is quite a lot.
CUDA Cores And Shaders
Stream processors, shaders/shader units and CUDA cores are all, essentially, one and the same thing.
The terms all refer to the cores within a modern GPU, which go to processing all that graphical and mathematical data in the form of shading pixels according to how light hits them in a 3D environment. These cores, though, perform differently depending on the manufacturer of the GPU.
In the case of AMD GPUs, the stream processors or cores are called shaders or shading units. So the MSI Radeon R9 380 Gaming 2G has 1,792 shader units or 1,792 cores to help form the pixel shading routines.
Nvidia, on the other hand, has developed a more complex method of shading with its CUDA (Compute Unified Device Architecture) cores, which improve the overall processing power of the GPU. In the case of the MSI GTX980 Gaming 4G, with its 2,048 CUDA cores, there are 2,048 advanced shader cores.
As to which version of stream processors is better, at the moment we’re tempted to say Nvidia’s CUDA cores have the edge over AMD’s shader units, purely because the CUDA core is a little more advanced and can often be found helping boost the parallel processing power of supercomputers.
TDP
Thermal design power (TDP) or, as it’s sometimes called, the thermal design point. Basically it means the maximum amount of power/energy a processor – GPU in this case – needs to dissipate the heat generated to keep the component below its maximum temperature range.
So in the case of the MSI GTX980 Gaming 4G, it needs 165 watts of cooling power to keep the heat down. It doesn’t necessarily mean that the component will be consuming that amount of power all the time, but it means that the amount will possibly draw that much energy over a sustained period of time.
You can, however, draw some conclusions based on a GPU’s TDP, in that the higher the TDP, the more heat needs to be siphoned off by the cooling solution. That usually means the more heat, the more work the GPU and other components are doing, so effectively the card may well be performing better than an equivalent model with a lower TDP.
3DMark
3DMark is the benchmarking tool created by Futuremark to test the performance of a graphics card’s 3D rendering and GPU workload.
The better the number, the better and more detailed graphics can be had from the card. The example of the MSI GTX980 Gaming 4G with a 3DMark score of 16,101 is really quite high and means it’ll easily be able to handle the intense 3D work needed to play the latest games at high resolutions and with the highest graphical setting.
It’s not always the best real-world benchmark, though. You’re always better trying to find a system that’s similar to your own in terms of processor, memory, hard drive and even Windows version, to compare a graphics card with. If a similar system can play Fallout 4 at 60fps at 1920 x 1080 resolutions with one card, but only 30fps with another, then you can happily say that the 60fps card is going to be better.
Conclusion
There’s a lot to get your head around when it comes to graphics cards, which is why we often crave the old days of simply buying 4MB graphics cards and being able to play Doom on our DX2-66.
However, if you take the time to look into the GPU the memory, its bandwidth, the number of cores or shaders and what the TDP is, then you’ll be able to judge which is the right card for you and the best value for money and performance.