Windows on ARM benchmarks show performance isn’t its strong suit

What’s also hilarious is that you’ve been talking about UWP being universal and AAA UWP gaming on these Windows on ARM laptops, yet not a single AAA UWP game right now actually supports ARM or even 32 bit OS’s. UWP is not some magic universal platform. There are very good reasons developers only build for x64 or a specific platform.

Well duh…devs wouldnt have a reason to compile for ARM64 because the devices weren’t there and support in Visual Studio is recent for ARM64, i think starting with Spring Creators Update.

There are plenty of free to play GameLoft UWP games with ARM32 support… Age of Empires: Castle Seige has arm support. But those arent AAA games. MineCraft runs butter smooth on ARM64 Even though it is still only ARM32 support.

Minecraft Windows 10 Edition (aka Pocket Edition) runs perfectly fine because that’s what the pocket edition was meant for. I’ve had Minecraft Pocket Edition running super smooth as far as my original HTC EVO 4G, back when the leading 4G standard was still WiMax instead of LTE. If we start getting into the actual full fledged PC version of Minecraft or even the console editions, the story changes there.

As far as ARM64 support goes, I still don’t see it happening for most AAA games. The system requirements for many of the newer games make it unfeasible for even a Surface Book with a GeForce 940M to run it acceptably. Now with ARM devices, we are suddenly taking a nosedive in performance and TDP. Not to mention the RAM requirements.

Lets not discuss the crappy java minecraft. The pocket edition is simply the C# edition. I didnt say you are going to see AAA games on Windows on Arm right away or automatically, my point was the potential is there for many games. Obviously the graphical detail would be reduced greatly and locked to 30 or 60 fps just like console but the Windows 10 SDK and unreal engine are designed to Scale from 20 teraflops multiGPU PC to 1.3 teraflops xbox one. ARM has to surpass atleast 1.5 teraflops to 2 teraflops gpu single precision compute before AAA gaming on ARM becomes more viable with UWP.

I’ll give you one thing about UWP’s potential success though and that’s Windows Mixed Reality and VR Apps in general. SteamVR is still more popular, but it’s not totally dominant like Steam is for buying games, so there’s room in VR for UWP and the Windows Store to grow and thrive. And that’d be a great thing too since I’m all on board with WMR. I’m ordering a headset soon and already own a HoloLens. That being said, Windows Mixed Reality has pretty high system requirements. My 1st gen Surface Book, even with its i7 and GeForce 940M still misses a few minimum requirements. So with that in mind, you can forget about WoA devices being Windows Mixed Reality ready. Now you could say WoA isn’t meant for that. Ok, that’s fine, for today. But you can clearly tell that Mixed Reality is Microsoft’s vision going forward and they’re pretty… Read more »

As I mentioned on MSPoweruser, FLOPS is not an accurate measure of GPU performance. It doesn’t mean much at all. Even if it matched the Xbox One in TFLOPS, that really doesn’t mean anything. Let’s use a different metric for example. Cryptomining. My Galaxy S8 with the SD835 gets a hashrate around 30 at best for mining XMR, with crazy heat and battery drain. You say it’s not that far behind the Xbox One in FLOPS, but the Xbox One’s hashrate for mining XMR is around 400-500. The Xbox One’s FLOPS count certainly isn’t 10-15x higher than that of the SD835 though. You see, comparing TFLOPS between two difference architectures and using that to measure performance, is pretty much comparing apples and oranges. The metric is totally meaningless. You have so many other factors that affect performance ranging from transistor count, architecture, TDP and cooling, optimizations, even fab process. The… Read more »

I knew you’d be here spreading your false propaganda about Microsoft and ARM. Clearly you didn’t even read the article this time. If you did, you’d have seen native benchmarks were done and it shows that even native ARM64 performance at best is comparable to a low end entry level Celeron CPU.

Once again, no one wants a watered down UWP experience. And no dev wants to waste time and resources to recompile for UWP when they’re already strapped for time and resources at it is. It’s cheaper, easier, and more profitable to just release applications like games on Steam and instantly reach a wider audience than UWP ever will. Why shoot yourself in the foot and release for such a limited audience?

You love talking in absolute certainties don’t you, like you speak for the rest of the windows users. lets correct you on few things: 1. The native ARM64 test done was only on Edge using Google Octane…..if you see nothing wrong with that…..i won’t say more. 2. The difference in performance of a celeron n3450 vs core m3 on cpubenchmarks (passmark) is only like 1500 points. If as you say…SD835 is celeron n3450 levels, lets go with that base level, SD845 is 30% faster than sd835. SD855 coming next year in 10 months will be even faster, SD865 will be even faster in 22 months. ARM has been doubling in cpu and gpu compute every three years. and these are only snapdragon chips, the Nvidia ARM chips destroy SD in performance, their latest Xavier reaches xbox one level 1.3 teraflops in GPU single precision compute. 3. There are 50 million… Read more »

ARM has been around for almost as long as x86, and as always, ARM is still and will always be farrrr behind x86 in terms of performance.

There are benchmarks showing ThunderX2 (32c) outperforming Broadwell (18c), and being competitive with Skylake (22c). So not far behind, though obviously none of those are low-power devices.

That’s the thing though, it may be competitive with more mainstream Broadwell and Skylake CPU’s. But Intel has shown they can scale up x86 at a higher level than ARM can scale up. If that’s what ARM has to offer on the high end, Intel has Xeon Phi to offer on their high end which brings us into super computer territory, but still based on the same architectures in mainstream CPU’s. ARM simply isn’t meant to compete with Intel in the mainstream and high end computing market. ARM doesn’t scale up as well, but it scales down very well whereas Intel has struggled to scale down x86 competitively. ARM trying to enter the high end space and compete with Intel Xeon, is like Intel Atom trying to compete with Snapdragon and Tegra.

ARM beats the xeons in performance per watt, that is what the giant server farms are interested in more than having extra performance that would never get used. They would save up to 50% in electricity costs right off the bat.

The ultra high end Super Computing will belong to quantum computers soon. IBM and MS are both working on quantum.

You totally misunderstand quantum computing. First of all, we’re still a long way off from quantum super computers. Second, quantum computers aren’t magically more powerful than classical computers. They’re able to crunch through certain algorithms magnitudes faster than computers today, but that doesn’t mean outside of specialized computing, quantum computers are faster at everything. Take cryptomining for example. The consensus right now is that quantum computers wouldn’t have much of an advantage over regular computers when it comes to mining crypto currencies.

Intel does not need to worry about quantum computers cutting into the high performance super computing market anytime soon. And as for today, they also don’t need to worry about ARM. It’s the same situation as WoA computers. 50% less performance for 25-50% power efficiency. It’ll be an interesting experiment for some people, but the world will largely move on and stick with what they know works.

“ARM” can scale up to whatever their licensees need it to. Most of their customers so far are in the embedded/mobile space, so that’s what the implementations have been optimized for. The ThunderX2 is one of the first implementations of the ARM architecture targeting HPC applications, and has been selected for at least two supercomputers; the EU Mont-Blanc (with 3000 ThunderX2 cores), and the British GW4 Isambard (over 10000 ThunderX2 cores).

There is absolutely nothing about ARMv8 that would make it inherently un-scaleable, just as there is nothing about x86-64 that makes it inherently scaleable.

It’s an architecture designed for low power/embedded computing. That’s what prevents it from scaling up as well as x86-64. Likewise, x86-64 was designed for the opposite, preventing it from scaling down well to compete with ARM in mobile. Idk about more recent ARM servers, but at least as of a year ago, ARM servers have always been more BUS limited compared to other competitors. x86-64’s pipeline can throughput more data through the different layers of caches faster than ARM can. Not to mention that the x86-64 and Xeon’s specific architectures contain numerous additional instructions that give significant optimizations to applications (eg JIT and AOT compilation). Not to mention a big use case of server farms these days is virtualization for cloud applications, which there’s no doubt the x86-64 architecture contains incredible instructions and optimizations for virtualization. It’s unlikely that will ever change for ARM either since adding more instructions defeats… Read more »

Arm has been around but i was correcting him… MIPS, ARM, PowerPC are all RISC architectures.

Fair enough