Introduction: Analyzing the Price of Mobility

Computers have been getting faster over the years, and with the increased performance we eventually passed the point where most systems were “fast enough” and the various features and use cases became more important. It used to be that to get similar performance to a desktop, a laptop would generally cost two or even three times as much – and even then, sometimes it was simply impossible to match desktop performance with a laptop. Has that changed with the era of “fast enough” computing? One of our readers suggested we take some time to investigate this topic to help enlighten the general public, so we pulled together results from recent laptop and desktop/CPU reviews to see how much of a premium we’re now paying to go mobile.

There’s a related topic that I’m not even going to get into right now: tablets. The short summary is that at the low-end of the price spectrum, tablets can actually fill quite a few requirements. They’re slower, but battery life and portability is also better. Typing on a screen is not something I really enjoy at all, though, so adding a keyboard would almost be a requirement, which means at a minimum we’d be looking at closer to $500 for a decent tablet with a keyboard (e.g. ASUS Transformer TF300T with the keyboard dock). Okay, I said I’m not getting into this subject; basically, it’s possible to get a $500 tablet with keyboard (perhaps even $400) but performance is a major step down from even a budget laptop. That’s changing but for now I’m going to focus on Windows laptops vs. desktops.

Naturally, when we talk about performance, there are many factors at play. CPU and GPU performance are usually the biggest items, but in some cases the performance from the storage subsystem can actually trump the other two. A modern desktop with the fastest CPU and GPU available will handle pretty much anything you want to throw at it, but if it’s using a hard drive (HDD) for storage even a moderate Ultrabook equipped with a solid state drive (SSD) can be faster at booting into Windows or launching several applications at the same time. That might seem like an odd performance metric, but if you’ve ever experienced the dreaded “turn on the PC and wait five minutes after Windows loads before the system is actually ready for use” scenario, you’re running into storage bottlenecks.

We’ve advocated the use of SSDs for the OS and applications for several years now and we’ll continue to do so. In terms of storage performance, a good SSD will be at least 2-3X as fast as the best HDD for sequential transfers, but more importantly it can be 50-100X (or more) faster in random accesses, which is similar to what happens during the Windows boot process or when you launch a bunch of applications simultaneously (or launch a browser with dozens of tabs).

The good news is that nearly all laptops can be easily upgrade with an SSD if you’re willing to pay the price and take the time to do the upgrade yourself; the laptops that can’t be upgraded with a typical SSD are usually Ultrabooks that already have SSDs. The only drawback for SSDs is capacity: a typical 1TB 5400RPM 2.5” HDD will cost around $80; Seagate’s hybrid 1TB HDDs (with a bit of solid state cache to improve performance) will set you back around $130. The least expensive 240GB SSD in contrast costs around $165, with “better” models (faster, more reputable, and/or larger) costing up to $230 (or more). That’s 2x to 3x the cost of a hard drive for 1/4 the capacity, but the performance benefits are tangible. We’ll stick with comparisons between SSD-equipped systems for this article, just to keep things easy.

CPU/General Performance Discussion

With that out of the way, let’s focus on the remaining two areas, starting with CPU performance. I’ve grabbed several generations of desktop CPUs from both AMD (K10.5, Llano, Trinity, Bulldozer, and Piledriver – we don’t currently have Richland results, but they’re generally 5-10% higher than Trinity) and Intel (Sandy Bridge, Sandy Bride-E, Ivy Bridge, Ivy Bridge-E, and Haswell) and combined those results with the latest mobile CPUs (Haswell quad-core, Haswell ULT, and Richland). We’ve got two benchmarks that are common to our CPU and laptop we’re going to look at: Cinebench 11.5 (single- and multi-threaded) and x264 HD 5.x video encoding (first and second pass).

Unfortunately, we don’t have results for all of the desktop CPUs in x264 HD 5.x, though you can get an idea of where they should fall by looking at the x264 HD 3.x results. We’ve gone ahead and used the x264 3.x scaling for desktop CPUs where we don’t have 5.x results, and marked those results with an asterisk (*) in the following graphs. The first pass tends to be about 2x as fast with x264 HD 3.x, unless you have four cores with Hyper-Threading in which case scaling is closer to 1.5x (and 6-cores results in scaling of around 1.08x). The scaling in the second pass is much more uniform, with 3.x being around 2.75x as fast as 5.x. Since we’re already comparing apples and oranges in a sense (desktop and laptops, with different motherboard, RAM, etc.) we’re really just interested in an estimate of performance. Do not take the following charts as 100% definitive, but rather as a baseline comparison.

It used to be that if you wanted top CPU performance, a desktop system was absolutely required. These days, for lightly-threaded workloads the gap has narrowed substantially: a moderate i7-4700MQ Intel Haswell processor will get you around 85% of the performance of the desktop i7-4770K, and the i7-4930MX will get you 98% of the single-threaded performance. Even the i7-4500U Ultrabook is able to deliver 75% of the 4770K performance. Move to heavily threaded workloads and the gap grows (particularly if we included the hex-core processors), but the 4700MQ and 4930MX are still 81-87%/91-98% of the 4770K performance – it’s the dual-core mobile CPUs that really take a hit, with the i7-4500U delivering roughly one third (37-41%) the performance of the 4770K. Of course, if you’re not doing CPU intensive tasks like video encoding, the single-threaded performance will be the more pertinent result.

The AMD side of the equation isn’t nearly so rosy, at least if you want performance anywhere near that of a top-end desktop. Llano, Trinity, and Richland have been far more focused on reducing power use and increasing integrated graphics performance than on boosting CPU performance. The result is that in single-threaded performance, the fastest mobile Richland APU is less than half the performance of the 4770K, and in the multi-threaded tests it’s 26-31% of the 4770K performance. Comparing mobile Richland to desktop AMD parts doesn’t really help much either: in the single-threaded Cinebench result, it basically matches the Llano A6-3650 and is 12% slower than the A8-3850, while it trails most of the other AMD CPUs/APUs by nearly 30%. Move to multi-threaded workloads and it’s only able to deliver 30-60% of the more powerful AMD desktop APUs (and 60-80% of the desktop Llano APUs). Richland is still “fast enough”, but it’s definitely a value proposition as opposed to a performance option – the iGPU of Richland may be faster than HD 4000/4400/4600, but dGPUs like the GT 730M are still roughly twice as fast and available for not much more money. But we’re skipping ahead….

Summarizing the general performance of our mobile platforms, we’re able to come pretty close to the performance of desktop systems, and while there’s still a price premium in effect it’s definitely getting smaller than in the past. The least expensive laptop/notebook I can find right now with a quad-core Haswell CPU is the HP Envy 15t-j000, which you can currently get for $700 after the $100 instant discount. That will get you 8GB RAM, 1TB HDD, and Windows 8, but not too much else. Adding a 1080p LCD ($50), GT 740M ($70), and a backlit keyboard ($30) brings the total to $850. That’s roughly the same performance as a Sandy Bridge i7-2600K, and slightly faster than the Haswell i5-4670K. You can get a pre-built desktop from Newegg for $700 (which adds in an HD 7750 GPU), while a DIY system with similar specs to the laptop (e.g. no dGPU) will set you back $672 (see table below) – not including a mouse, keyboard, display, or speakers.

In other words, if all you want is a computer for moderate needs, it’s not hard to see why laptops are displacing desktops for many people. In this case we’re able to get the same level of performance for potentially less money (depending on whether or not you need to buy an LCD, keyboard, and mouse). Desktops are still going to be far more upgradeable, and you can take a “basic use” desktop like the one above and upgrade it with a high-end graphics card for $300 to get a potent gaming PC. Adding more memory, more storage, upgrading the motherboard and CPU, etc. are all things that you can do with a desktop, whereas with laptops typically only the RAM and storage can be upgraded/replaced and if anything else breaks you’ll have to send it in for repairs (or buy a new laptop).p

We can go down to lower price points of course – there are desktop PCs with moderate hardware starting at just over $300, and they’re still more upgradable than laptops, but similar performance laptops can be had for about the same price, especially if you look at some of the previous generation hardware. It’s mostly a question of how much performance you want/need, and whether you’re interested in the potential to upgrade components down the road. If you don’t have any plans of adding more RAM, storage, or a GPU the case for laptops is stronger now than ever. And let’s not forget that most laptops have an idle power draw south of 20W, compared to 40-60W for a typical desktop (and that’s not including the display or speakers on the desktop), so the energy factor also favors laptops.

GPU/Graphics Performance Discussion

The other area of performance we want to look at is the GPU/graphics side of the equation. Where CPU performance has really started to level off – only overclocked CPUs or hex-core/octal-core options really hold serious performance advantages over the fastest mobile CPUs, and then only in heavily threaded workloads – the same can’t be said of graphics chips. Part of the problem is power requirements, as high-end desktop GPUs can draw up to 300W under load, which is three times what the most powerful mobile GPUs are rated to draw. The form factor also comes into play, but really I think power is the far more limiting factor.

Of course, there's a big difference in terms of what we do with CPUs vs. GPUs and how we do it. Most CPU-intensive tasks don't easily scale beyond a certain level of multi-threadedness, and some tasks really don't scale at all. The easiest example is anything that requires user input: no matter how fast your CPU gets, Word isn't going to type any faster than your fingers can go – or if you use speech recognition, any faster than you can dictate. Graphics on the other hand is practically infinitely parallel, where we can break things down into smaller and smaller pieces of a screen being rendered/computed by individual GPU cores. Many tasks on the CPU don't scale beyond four or maybe eight cores, but graphics is easy to scale to thousands of cores and beyond. So GPUs can get as large/fast as we can conceivably make them and there are still ways to tap into that performance potential; that's why modern systems typically have GPUs that consume two to ten times as much power as the CPU(s).

Let’s again put together some comparisons, only this time we’re a bit more limited in what we’ve tested with modern GPUs – on our desktop tests, we only have an overclocked GTX 680 and a GTX 780 with all of our gaming results, while on mobile we have a few additional reference points. I’ve also added Bioshock Infinite and Sleeping Dogs numbers for additional GPUs from our GPU 2013 test suite, but that’s only two of the seven titles we’re testing on laptops. We’re now looking at entire systems, but the GPUs are the primary factor in gaming performance. Also note that our reference point (1080p Max Detail, usually with anti-aliasing) is far more demanding than what most gamers actually require, but in terms of a performance discussion at least we can get an idea of where mobile GPUs stand.

As far as replacing gaming desktop performance with a notebook is concerned, we’re nowhere near close to accomplishing this task. CPU requirements for the vast majority of users have reached a plateau during the past five or so years, but for graphics (games) we can still use pretty much all the horsepower available. The fastest mobile GPU we have right now, the GeForce GTX 780M, ends up delivering performance that falls roughly half way between the GTX 660 and GTX 760, or if we look at the Digital Storm system Dustin tested, it offers 60-70% of the performance of that desktop. Moving down to the lower performance mobile GTX parts like the GTX 765M, we haven’t tested any desktop GPUs in recent history that are anywhere near that slow – even the GTX 660 tends to be more than twice as fast as the GTX 765M and the GTX 560 Ti is still 50% faster. Drop one level further to the GT 750M (DDR3) and performance is about 60-80% of the GTX 765M! It’s not that you can’t game on a GT 750M or GTX 765M, but you’ll definitely need to drop the details down from max and turn off anti-aliasing. Finally, no matter what you do with a notebook, you’re not going to match the performance of the fastest desktop GPUs – the GTX Titan nearly doubles the performance of the GTX 780M, and even SLI GTX 780M isn’t likely to beat a Titan on a consistent basis.

Looking at pricing is also illuminating. Most notebook vendors charge around $650 (give or take) for a second GeForce 780M on the Clevo P370SM, compared to the desktop GTX 760 (which has slightly higher performance) that can be purchased for $250. Meanwhile, the desktop GTX 780M sells for the same $650 price as the GTX 780 while offering only 60-70% of the performance you get from the desktop GPU; the GTX 680 and GTX 770 have the same number of CUDA cores (1536) and go for $400, delivering superior performance at a substantially lower cost. The pattern continues as we move down the product stack; the slowest desktop Kepler, the GeForce GT 630, is a $65 card that has slightly higher clocks than the GT 750M, never mind the GT 730M and GT 740M, but the GT 730M is typically a $70 upgrade on a notebook. The GTX 765M has the same number of cores as the $130 GTX 650 Ti but with lower clocks; it reputedly sells for close to the same price to laptop OEMs as the desktop GTX 760 we mentioned as a performance equivalent of the GTX 780M.

Of course, where you can easily add a GPU to most desktops, with laptops you’re basically stuck with whatever you initially purchase. I noted in the Gaming Laptop Guide last month that the lowest price I could find for a notebook with a GTX 780M ends up around $1750 (including the OS with a 1TB HDD). For equivalent performance (more or less), we can take the desktop we listed above and add a GTX 760 GPU, with a slight PSU upgrade as well just to be safe. Here’s what we end up with:

Opting for a gaming notebook thus carries a hefty price premium, nearly doubling the price of a similar performance desktop. Yes, the notebook still gives you a battery, display, keyboard, touchpad, and speakers as part of the package, but if you already have those for an existing desktop they’re not really necessary. The price premium for high-end gaming notebooks has certainly come down from the 2x-3x we used to see (though you can still get to that range with SLI notebooks), but we’re nowhere near parity like we are on the CPU side of things. As an alternative, desktop GPUs are also able to hit much higher performance levels at a similar cost to a high-end gaming notebook.

Given the various options, many people have concluded that it’s far better to have a decent gaming desktop like the one above and then to spend another $500-$1000 on a good laptop for when you need to go mobile. That’s generally my recommendation as well – for all the gaming notebooks I’ve reviewed over the years, I still turn to my desktop system 95% of the time when I’m looking to play a game. On the other hand, not everyone wants to worry about having two computers and syncing data between them, and there are also space limitations to consider. Finding a spot for a 10 pound gaming notebook in a college dorm is easy enough; finding space for a gaming desktop with display, speakers, keyboard, and mouse can be a bit more difficult. And if you’re regularly on the road and want to take your games with you, you’ll need the laptop regardless.

There’s no right answer for how to do things, of course. Some people love gaming notebooks and others don’t see the point. But while you can certainly make a case for typical notebooks/laptops now being close to price parity with similar performance desktops, once you add gaming into the mix you’ll need to pay the mobility piper his due.