You’ve probably heard quite enough about “12 teraflops” this week. But if you’ll indulge us one more article, we’ll break down what the figure means – in useful, real-world, value terms – for the Xbox Series X.
We still haven’t been revealed the actual specs of the Xbox Series X. We’ve been given glimpses of performance. Snippets of capability. A general hand-waving sense of what to expect.
That includes things that make a practical difference to players, including further support of Xbox Game Pass, backwards compatibility, and a sort of “forwards compatibility” that means, even if you buy a game like Cyberpunk 2077 on Xbox One, you’ll get to play it for no extra charge on Xbox Series X.
On the technical side that includes SSD storage (which will all-but eliminate loading times), real-time ray tracing (for hyper-realistic lighting effects), and something called “Dynamic Latency Input” which means, we think, that controls will be more responsive. Somehow. Microsoft hasn’t gone into the details for that one.
The closest thing to actual specs we’ve seen is the promise that games will be able to run at 120 frames per second, and that the Xbox Series X will be capable of 12 teraflops. And that’s been all over the internet since the announcement yesterday. 12 teraflops, plastered all over the press. 12 teraflops, repeated like a mantra. 12 teraflops. 12 teraflops. 12 teraflops.
So, what exactly is a teraflop, and why is having 12 of the things a big deal?
What the flop?
A teraflop – the nice, spellcheck-friendly way of writing out TFLOP/s – is a unit of measurement. Here’s how it breaks down:
- “Tera” the size, like kilo, mega, or giga. (If mega is million, and giga is a thousand million, then tera is a million million. Yes, like gigabytes, megabytes and terabytes.)
- A “flop” is a floating-point operation. (In really simple terms, that’s a complex mathematical calculation involving either really big or really small numbers, i.e. lots of decimal places.)
- And the “/s” on the end signifies per second. (That bit’s easy. We’ll give you a simple one to finish.)
So the Xbox Series X does 12 teraflops, or 12 TFLOP/s, or twelve million million floating-point operations per second. But why do we need to use teraflops to measure performance? Surely you can look at the specs of a device and see at a glance which one is better?
The truth is that that figure of 12 teraflops, being shouted about town in relation to the Xbox Series X, isn’t in itself a specification. You can’t look at “12 teraflops” and instantly get a handle on how powerful something is. “12 teraflops” is, in isolation, a bit meaningless.
The value in using teraflops as a relative unit of measurement is that it’s a measure of theoretical performance that isn’t system-dependent.
Let’s take the performance of cars as a for-instance. You can look at the engine of one car, and see that it’s got a displacement volume of 2 litres. If you see another car that’s got a 1.6-litre engine, you’d think you were safe to assume the latter is slower. On paper, that makes sense. Bigger is better, right?
But what if the smaller engine has a turbocharger? What if the smaller engine produces more torque? What if the smaller-engined car weighs less? What if it has a Fast and the Furious nitrous oxide button? In real-world terms, the bigger engine isn’t necessarily better. That’s why we use other units of measurement – miles per hour for top speed, time to 62mph for acceleration – to offer an absolute comparison between things that might be very different.
Now consider computing, and more specifically, graphics processing hardware.
The processor in your PC is easy. A core’s a core. It’s easy to compare, no matter which manufacturer you’re buying from. Graphics cards are a little different. Nvidia measures the number of “cores” in its GPUs with its proprietary CUDA cores. AMD measures its “cores” in Compute Units, or CUs, which are made up of Stream Processors. These are not exactly the same thing.
You can fudge it, a bit. You can make comparisons. If you multiply the number of CUs in an AMD graphics card by 64 (because a Compute Unit contains 64 shader processors) you can come out with an approximation of Nvidia’s core count. But even when you make that sort of equivalency it’s still not an entirely fair fight. One graphics card might have a faster base clock speed. The other might have a faster boost clock. But the boost clock might not be on all cores. One might be more efficient. The other might be able to run hotter. Graphics card specifications may look similar across generations, even, but newer variants can offer significant performance increases for less power. (And when you take inflation into account, they’re not always that much more expensive. (Well, when cryptocurrency-mining shortages aren’t driving prices.)
All this adds up to is a very difficult way to compare technology. So we use the number of teraflops a graphics card can output as a sort of vendor-neutral, agnostic expression of maximum mathematical throughput. And the sort of mathematical throughput that teraflops are used to measure (in the enterprise IT and scientific worlds) is very similar to the workload that’s done by graphics cards when doing rendering, shaders, and all that pretty stuff.
And in the absence of any other specs to go on, we’ll have to use the “12 teraflops” number provided for the Xbox Series X to work out whether it’s a “good” number or not.
So is 12 teraflops a good number, then?
So now you know (roughly) what a teraflop is, (mostly) how it’s calculated, and (broadly speaking) why it’s a good idea to use teraflops to compare the power of something like the Xbox Series X. But is 12 teraflops a “good” number?
That depends. (Sorry, there are no easy answers here, but hopefully, easy-to-follow explanations.)
Given the teraflop is a unit of peak theoretical measurement that is, in isolation, a bit meaningless, we have to compare the 12 teraflops of the Xbox Series X against some other things (also measured in teraflops) to see how it stacks up. The sensible place to start is video game consoles. Here’s how the 12 teraflops of the Xbox Series X stacks up against other console hardware:
- Xbox Series X – 12 teraflops
- Xbox One X – 6 teraflops
- PS4 Pro – 4.2 teraflops
- Original PS4 – 1.8 teraflops
- Xbox One S – 1.4 teraflops
- Original Xbox One – 1.3 teraflops
- Nintendo Switch – 1 teraflop
- Nintendo Wii U – 350 gigaflops (note the change in measurement unit here)
- Xbox 360 – 240 gigaflops
- PS3 – 230 gigaflops
- Original Xbox – 20 gigaflops
- Nintendo Wii – 12 gigaflops
- Nintendo Gamecube – 9 gigaflops
- PS2 – 6 gigaflops
And we’ll stop there. (You could argue that the numbers stopped being relevant when we dipped into gigaflops, but we thought it was interesting. It’s a great illustration of how far hardware has come in the last 20 years.)
What’s obviously missing from that chart is the PlayStation 5. It feels like we’ve been playing a big game of chicken, with neither side wanting to show their cards while there was still time for their opposite number to change tack or undercut their offering. Given that the Xbox One X went second – and beat the performance of the PS4 Pro by 30% – you have to think that Microsoft must be feeling pretty confident about the 12 teraflops figure in relation to what the PS5 will offer.
The really interesting thing about the Xbox Series X is that it’s, in some ways, more akin to a gaming PC than a games console. It’s bigger. It’s more powerful. It doesn’t exactly look like it fits under your telly. So how does the 12 teraflops figure for the Xbox Series X stack up against PC graphics cards that are currently on offer?
- Nvidia Geforce RTX 2080 Ti – 13.5 teraflops
- Xbox Series X – 12 teraflops
- Nvidia Geforce RTX 2080 Super – 11.2 teraflops
- Nvidia Geforce RTX 2080 – 10.1 teraflops
- AMD Radeon RX 5700 XT – 9.8 teraflops
- Nvidia Geforce RTX 2070 Super – 9.1 teraflops
- AMD Radeon RX 5600 XT – 8.1 teraflops
- AMD Radeon RX 5700 – 8 teraflops
- Nvidia Geforce RTX 2070 – 7.5 teraflops
- Nvidia Geforce RTX 2060 Super – 7.2 teraflops
- Nvidia Geforce RTX 2060 – 6.5 teraflops
- AMD Radeon RX 5600 – 6.4 teraflops
- AMD Radeon RX 5500 XT – 5.2 teraflops
- AMD Radeon RX 5500 – 5.2 teraflops
(We’ve not included the RTX Titan or AMD Radeon VII because they’re not exactly easy to come by. We also haven’t included any professional or enterprise cards like Nvidia Quadro or Tesla.)
So there aren’t many graphics cards above the Xbox Series X on that list, and most of them will cost you upwards of a grand. The cheapest variant of the card directly above it, the RTX 2080 Ti, will give you change from £1000, while the most expensive comes in closer to £2000. The card immediately below it in the performance stakes, the RTX 2080 Super, will set you back £650-850.
Now think about how much a brand new, high-end games console – like the Xbox Series X – might cost. £450? £500? £550? £600? We reckon the PlayStation 5 will come in around the £450-500 mark, so if the Xbox Series X is more powerful (as Microsoft’s bravado might suggest) then it makes sense that it might cost a little more.
But the prices listed for the cards above are literally just the graphics cards. That doesn’t include the case, the motherboard, the CPU, the RAM, the SSD, the power supply, the cooling. Or the peripherals, like a fifty quid controller. Or the operating system, which isn’t free, either. Comparing it to pre-built systems – which is just easier, but you’d be able to do this cheaper in a self-build – you’ll be able to pick up an Alienware Aurora desktop, with an RTX 2080 Ti (and all the other high-end components, including a high-end processor, 16GB of RAM, and an SSD) for three grand. Yes, that’s three thousand pounds.
And then think about console pricing. Microsoft might be able to offer the Xbox Series X for under a grand. That’s under the cost of just the equivalent graphics card on its own. It might even be in the console pricing Goldilocks zone, closer to £500. That’s just remarkable.
So if your question is, “is the quoted 12 teraflops figure in the Xbox Series X good?” then the answer is, “it depends” and “that’s a bit of a vague question, to be honest, good’s a bit subjective.”
But if your question is, “is the quoted 12 teraflops figure in the Xbox Series X good compared to equivalent-power PC gaming hardware, particularly as a value proposition?” then the answer is, “yes, absolutely” and “it might seem expensive for a new console when the price is announced, but there’s no way you’d be able to build anything similar for anywhere close to console prices.”