The Review - Introduction and Synthetics

This review is of the MSI Gaming X 1080 provided by Overclockers UK.

Test rig for this review:

MSI X99S XPower

Intel Haswell-E i7-5960x overclocked to 4GHz

Noctua NH-U12S CPU cooler

32GB Corsair Vengeance 2400 DDR4

Storage: Kingston 240GB SSDNow M.2 SATA (OS), 1TB Samsung 850 Evo (games), 4TB WD Black (data).

EVGA SuperNova 1200W P2 PSU

EVGA Titan X SuperClocked 69.0% ASIC

Stock Configuration: Core 1127MHz, Boost 1216MHz, Memory 7010MHz GDDR5. Temp at idle 42C, temp under load 83C.

Custom Overclock Configuration: Power Target 110%, Temp Target 91C, GPU Clock Offset +110MHz, Mem Clock Offset +325MHz, 75% constant fan (3600rpm). Temp at idle 34C, temp under load 66C. Final core clock 1414MHz, final memory clock 7656MHz.

MSI Gaming X 1080 Unknown ASIC Quality (GPU-Z can’t read Pascal ASIC quality yet).

Stock Configuration: Core 1708MHz, Boost 1847MHz, Memory 10108MHz GDDR5X. Temp at idle 42C, temp under load 70C.

Custom Overclock Configuration: Power Target 121%, Temp Target 83C, GPU Clock Offset +159MHz, Mem Clock Offset +499MHz, Auto fan, temp at idle 42C, temp under load 70C. Final core clock 2113.5MHz, final memory clock 11016MHz.

4K

That little two character term has had a lot of meaning for me in recent years. It’s cost me a lot of money, it’s been my goal, just two little characters.

Of course for those on this site, you know exactly what it means. Enough to know that if I don’t point this out explicitly, I’ll be bombed in the comments so let me be clear. I know that what most people refer to as 4K is in fact not 4K but UHD. Still, the rest of the non-technical world doesn’t particularly care now and as you can tell from tons of TV’s currently marketed as 4K/UHD, the world has moved past the debate.

But the point is, whether you care about 4096x2160 (ACTUAL 4K) or 3840x2160 (UHD), the GPU horsepower needed to drive gaming at these kinds of resolutions is still awesome and, more to the point, has so far been generally out of reach.

Oh sure, I get it, you’ll all talk about various SLI and Crossfire configurations, that can drive 4K gaming at reasonable levels. You’ll tell me that with turning down settings I can use a 980 Ti or a Fury X at 4K and get playable framerates, but here’s the thing.

That’s a load of hogwash.

Yes, multi-gpu has come a long way, but not far enough to make me want to live with the compromises that it always entails.

Yes, I can compromise settings, and this is what I’ve spent the last year and a bit doing ever since I got a Titan X as I had waited long enough and when the first single GPU card came along that seemed somewhat capable of 4K gaming, I pulled the trigger and bought it knowing I would need to compromise on settings, because I’d waited long enough and wanted my 4K NAOW!

But I’m kind of sick of that situation. You tweak settings and play around with this and turn off anti-aliasing by default and turn down this and disable that.

Can you get a playable experience? Absolutely, is it what I and a lot of other gamers out there want? Absolutely not.

Really what I’m looking for is a genuinely capable 4K60 Ultra card. I had resolved to wait for GP102/Vega based cards to come along when the 1080 came out but the nice folks over at OverClockers UK somehow managed to twist my arm into giving a 1080 a try so I figured I’d give it a go.

Picking one was slightly problematic, as I’m sure many of you are aware. Supply on 1080’s is still not really able to keep up with demand. A quick look at the OCUK 1080 page (here) shows that there are lots of Founders Edition available, but little in the way of aftermarket cards which most people are rightly interested in. Whether the stories are true and Nvidia has a supply problem or they simply massively underestimated demand (maybe some combination of the two), the simple fact is that there aren’t many of these going around and you need to keep an eye on the page to get hold of one. I checked one day and they got the MSI Gaming X in stock and had 10+ available so thought I’d pick one up that evening when I got home. What do you know, by the time I got home, all sold out again. Hint, if you want to get hold of one, click the link above and F5 it regularly.

A quick chat with them and I was assured more Gaming X’s were coming in the next week so pre-ordered one and now here I am writing up the review. EVGA Classifieds are still unknown ETA, same with Strix (Strixes? Strix’s?) and Gaming Z’s.

Historically speaking both graphics vendors Nvidia and AMD have successfully maintained a cadence that ensured significant performance and power efficiency advances are delivered with every new architecture and Pascal is no exception. The very basic building block of Nvidia’s Pascal architecture has been stripped apart and redesigned. This building block which Nvidia dubs Streaming Multiprocessor or SM for short is the engine that drives the graphics and compute horsepower of every Pascal chip.

With Maxwell, Pascal’s predecessor powering the GTX 900 series, Nvidia introduced the Streaming Maxwell Multiprocessor. The SMM built on the strengths of Nvidia’s Kepler SM – introduced with the GTX 600 and 700 series – which Nvidia dubs the SMX. It also done away with many unnecessary complexities which enabled the engine to deliver more throughput and higher clock speeds. The Pascal SM in its own right is an evolution of the Maxwell SM, a smarter, more streamlined engine.

Inside Nvidia’s full GP104 powering the GTX 1080 (a cut back version of the GP104 GPU is used in the GTX 1070) – there are four Graphics Processing Clusters or GPCs. Each GPC consists of five Streaming Multiprocessors or SMs – each SM contains 128 CUDA cores – and sixteen Texture Mapping Units or TMUs. Each SM includes eight Render Output Units, ROPs. In turn GP104 houses 2560 CUDA cores, 160 TMUs and 64 ROPs. Finally the engine is connected via eight 32-bit memory segments – 256bit memory controller – to 8GB of GDDR5X memory.

As mentioned, each GP104 streaming multiprocessor includes 128 FP32 CUDA cores, the same as Maxwell. Within each GP104 streaming multiprocessor there are four 32 CUDA core partitions, four dispatch units, two warp schedulers and a fairly large instruction buffer. Twice as large compared to Maxwell.

This arrangement is almost identical to what we’ve seen with the much larger 3840 CUDA core GP100 GPU that’s powering the Tesla P100. Only inside GP100 each SM contains exactly half the number of CUDA cores, dispatch units and warp schedulers vs GP104. But in turn there are twice as many SMs per GPC.

So the only difference with GP104 is that Nvidia is grouping 64 CUDA core SMs in pairs made of 128 CUDA cores each and in turn naming the larger 128 unit an SM instead. This is all while maintaining the exact same ratio of dispatch units, warp schedulers and instruction buffers per CUDA core that we’ve seen with GP100.

So think of it as Nvidia just pairing 64 CUDA core groups together in a single SM. This decision is likely influenced by the significant reduction of FP64, double precision, CUDA cores per SM inside GP104 vs GP100. GP104 only contains 1 FP64 CUDA core for every 32 FP32 CUDA cores. While GP100 has 1 FP64 CUDA core for every 2 FP32 CUDA cores, 16 times more than GP104.

Additionally because each GP104 SM has twice the number of registers as Maxwell, this in turn means that not only can Pascal accommodate more threads than Maxwell, but each thread has access to more registers and thus a lot more throughput. Finally, each warp scheduler can dispatch two instructions per clock.

Nvidia’s Senior Architect, Lars Nyland admitted that the 16nm FinFET process played an important role in realizing the team’s power efficiency goals for Pascal, but maintains that numerous architectural improvements aided in further reducing the energy footprint of the architecture. Including the use of new on-chip voltage signalling technologies.

And so to the question. Is the 1080 a genuine 4K60 capable card? Let’s find out.

We’re going to do something slightly different for this review. A lot of people overclock, so I’m going to compare the MSI 1080 Gaming X to the Titan X on both a stock for stock basis, as well as from an overclock to overclock perspective. Please do keep in mind that overclocking if not done carefully can potentially damage your card/system and also keep in mind that the results you see here are obviously not guaranteed to be reached on any given 1080 you buy. Even on identical model cards, chip quality varies and this will result in different overclocked capabilities.

Furthermore, this isn’t an overclocking guide. There are plenty available on the web but for those curious, I’ve taken a basic approach and opted not to overvolt. I bump up the core clocks, faster at first, then by smaller amounts until the card displays instability (either via artefacts or the display driver crashing etc), then dial it back a bit and run the 3DMark Stress test.

Once that’s done, I do the same with the memory clock. In the end I managed to get the card relatively stable (98.6% on 3DMark Firestrike Ultra Stress Test) with a core clock at 2114 and Memory at 11016.

Also please keep in mind that for people considering overclocking their Titan X, half of the memory chips are on the underneath of the card so are not directly cooled. With no way of monitoring the temperature of these chips, one should absolutely be extremely cautious in overclocking Titan X memory (which GM200 responds very well to in performance testing so is extremely tempting to do!)

So, let’s take a look at the results.

Synthetics – 3DMark Firestrike

First up, Firestrike. Our review is focussed on 4K gaming but for the synthetics we used all three of the mainstream resolutions (being 1920x1080 or 1080p, 2560x1440 or 1440p and 3840x2160 or 2160p). Firestrike covers the 1080p benchmark, Firestrike Extreme is for 1440p and of course the punishing Ultra gives us our 2160p readout.

The callout point here is that for normal Firestrike (1080p) I am at least somewhat CPU bound once I start overclocking the Titan X (the “Brawn” achievement in Firestrike). Even so, the 1080 gives a commanding lead over the Titan X both in stock and overclocked form, giving about a 20% increase stock to stock and an 18% increase overclock to overclock.

As you raise the resolution, the difference becomes greater with an almost 25% increase on Extreme stock to stock and the same again overclock to overclock.

Ultra sees a little under 20% increase stock to stock and just under 18% increase overclock to overclock.

Meanwhile, overclocking gives a reasonable return on both cards with Titan X giving about 10% to 1080’s 8% boost in Firestrike, 9% to both in Extreme and just over 9%/just under 9% to Titan X and 1080 respectively in Ultra.

Scores are all well and good, but we want frames! For the various Firestrike results, the FPS scores are the average across the 2 graphics tests performed.

As we can see from these, the frame rate boosts line up and scale basically as the scores do. Standard Firestrike matches the increase in scores. Extreme gives slightly less bang for buck on frame rate vs. score giving only a 23% improvement in frame rates vs. a 25% improvement in scoring terms stock to stock and overclock to overclock. While Ultra again gives just over 20% and just under 20% boost to frame rates.

Synthetics – Unigine Heaven

Heaven is a similar story to Firestrike with the 1080 again laying down its serious gaming capability credentials. Unfortunately for some reason, Heaven didn’t want to play nicely with the Titan X in overclocked mode regardless of how much I experimented with it and tried to get any decent level of boost over stock. As such, all Heaven slides are looking at Titan X at stock clock speeds only. It’s not ideal, but rest assured that most of the tests were fine with the Titan X being overclocked.

In using Heaven, we of course use Ultra with Extreme Tessellation, 8x AA and full screen at the various resolutions.

Heaven doesn’t give us any fancy achievements if we’re CPU bound. It does seem to scale better at 1080p than Firestrike did giving a 23.8% increase stock to stock and 34.4% boost stock to overclock (unfair yes but you can only compare what works at the end of the day).

1440p gives a worse return here than 1080p with a touch under 16% stock to stock and 26.6% stock to overclock.

Finally, Heaven gives just over 10% stock to stock with 21% stock to overclock at 2160p.

On the frame rate side of things, we see a broadly similar story.

Frame rate percentage increases stock to stock at 1080p gives us identical percentage performance increases to the scores at 23.4% and 34.4%, as well as also identical jumps with 16% and 26.5% at 1440p and 10%/21% stock to stock/stock to overclock respectively at 2160p.

We start out with standard games which are not DX12, then we’ll also take a look at a couple of DirectX 12 games at the end of the review and compare both stock and overclocked performance on the two cards with both versions of DirectX. All of these games were tested at 2160p in exclusive fullscreen unless otherwise noted.

Batman: Arkham Knight

Arkham Knight was largely slated when released on PC as an unplayable and buggy port. Unfortunate given the reception the series of games had for its storyline and gameplay. All but those with the highest performing cards found the game to be a largely unplayable mess at launch, particularly given the use of Unreal which seems to have struggled a bit lately with XCom 2 also not having the greatest performance at launch, especially when you consider it’s not a ridiculously killer engine like Cryengine for example.

Well, a cancelled port to Mac and some PC updates later, we’re now in a situation where the performance of Arkham Knight isn’t a complete travesty anymore so I thought it’d be interesting to see how these cards perform on the in game benchmark. Slide below is average throughout the benchmark with a table beneath showing minimum and maximum values. Overclocking both cards presented no problems in Arkham Knight and the game ran smoothly without artefacts, glitches or crashing drivers.

So as we can see, at stock clocks the 1080 holds a 15% advantage over the Titan X, while hitting a little over 11% lead looking overclock to overclock. Unfortunately we still need another 20% improvement over the overclocked 1080’s FPS to hit 60 frames per second at 4k but it’s a decent return.

Arkham Knight minimum and maximum frame rates come in as below.

	Titan X Stock	Titan X Overclock	1080 Stock	1080 Overclock
Min FPS	25	27	28	31
Max FPS	52	58	60	64

Doom (2016)

At the other end of the scale in terms of successful launches from Arkham Knight was the Doom reboot. An excellent game lauded webwide as a brilliant return to form and a great technical launch given the overall smoothness of the graphics and detail in game.

Doom doesn’t include a benchmark so for this testing we set ourselves up with a custom 60 second benchmark in the Argent Tower level opening segment. Frame rates here are getting much better, although as you will see from the minimum values in the table after, there is still ample room for improvement.

As you can see from the above, we get a nice boost in OpenGL using the 1080 over the Titan X with the stock to stock registering a 24.6% boost in average frames per second vs. a 16.26% increase overclock to overclock. It’s worth noting that given the lack of an in game benchmark, although relatively controlled in terms of the level, direction and actions taken, there will always be differences in performance here, as indeed there can be even in some in game benchmarks which use a degree of random generation in the benchmark (such as Ashes of the Singularity later).

	Titan X Stock	Titan X Overclock	1080 Stock	1080 Overclock
Min FPS	37	33	47	47
Max FPS	70	72	82	87

Shadow of Mordor

A couple of years old now but still a game which pushes cards, particularly at 4K with the high resolution texture pack installed, Shadow of Mordor was an example of a very well done port to PC in many ways showing that WB can do things right in the PC space when the mood takes them. I have had issues with running Shadow of Mordor in exclusive fullscreen at 4K however for quite a while and for some reason it doesn’t see 3840x2160 as my native resolution. To get around this, I run it in borderless mode which allows for the resolution without problem, although potentially giving up some performance.

For Shadow of Mordor we use the in game benchmark with minimum and maximum frame rates in the table below.

Again we see a decent performance boost from the 1080 at 4K with a stock to stock gain of 27.5% although only a 20% gain overclock to overclock. Even so, we’re well over the 60FPS for average frame rate here, although as below, you’ll see that there are dips into the fifties even for the 1080 in overclocked mode.

	Titan X Stock	Titan X Overclock	1080 Stock	1080 Overclock
Min FPS	37.81	44.43	46.07	52.92
Max FPS	67.5	74.64	114.56	92.04

Total War: Rome 2

Another difficult PC launch and several years old now but still a challenging game for modern machines to run. Rome 2 was a commercial if not technical success at launch but was regularly patched to get the game into a playable state. Rome 2 was another game that had issues with my normal Titan X overclock but I did manage to get it stable at +100 on the core clock and +300 on the memory clock.

We use the in game Forest benchmark for our testing here at 4K.

Stock to stock, we see a decent jump again with a 26.81% gain for the 1080 although only an 18.4% gain overclock to overclock.

	Titan X Stock	Titan X Overclock	1080 Stock	1080 Overclock
Min FPS	20	19	25	28
Max FPS	44	52	57	61

The final two games of our review probably need no introduction. Here we take a look at both games in DirectX 11 and DirectX 12. As many will know, Nvidia hasn’t seen anywhere near the kinds of performance gains in DX12 as AMD has and in some games (Total War: Warhammer springs to mind) has had significant performance loss.

Things got hairy for a bit when I realised that FRAPS didn’t capture screenshots in DX12. Asking a mate, the print screen key was suggested and the day was saved, although there was suggestion that I should perhaps be obliged to turn in my nerd card!

DirectX 12 is still in its infancy and although it is true, the gains made are basically little to nothing, to my mind I look for the best performing card, not the biggest gainer. It’s always nice to get a free performance boost, hence overclocking but at the same time there is also something to be said for the best absolute performer at a given price/performance level. Maybe for you that’s an AMD card, maybe an Nvidia one.
DirectX 12 certainly gives a lot more to the developer as an API than 11, but it’s also certainly true that developers have had a lot of time to gain familiarity with DX11 and have become very proficient at using it. I expect as time passes and developers shift their plans, we will continue to see improvements both in the way it is used, the performance it gives and the uptake on the consumer side with of course wider adoption of Windows 10.

All of this being said, it is still somewhat disconcerting that Nvidia cards are currently either roughly maintaining performance levels or significantly losing performance under DX12 scenarios. Whether this is something that Nvidia can easily fix or not, only time will tell. Perhaps Nvidia are simply devoting more of their resources to what they know is still the dominant API (DX11) and will do more on 12 once the market has matured. Perhaps there is a fundamental issue in the design of Maxwell/Pascal which means that they will simply not be able to do much beyond maintain existing performance levels if they’re lucky, only time will tell.

Ashes of the Singularity

Ashes is of course in many ways the defacto standard for DX12 given its early access availability and the many results which have been put out and analysed to death in its benchmark since it first became available.

The situation with respect to DX11 vs DX12 performance is unsurprising (although obviously not ideal) given the current state of Nvidia GPU’s in DX12, but we do still see gains as expected from the Titan X to the 1080 in both 11 and 12.

DirectX 11 first then. Stock to stock, we get a 23.8% gain, with a 19.8% gain overclock to overclock.

DirectX 12 in all but one scenario (1080 stock clocks) represents a drop in performance for the two Nvidia cards and even where it makes a gain, it’s so insignificant as to likely be within the margin of error for the generated benchmark run although the losses themselves are not huge.

Stock to stock in DX12 we see a 32.5% gain for the 1080, mostly generated by the fact that the Titan X loses almost 6% of its performance when jumping from DX11 to 12 whereas the 1080 gains an anaemic 0.7% improvement (but at least doesn’t lose in the transition like the Titan X does!)

Overclock to overclock in DX12 we see a 16.8% gain, again compounded by the fact that the Titan X loses almost 5% of its performance in the shift from the older API to the newer one, whereas the 1080 in this case only loses 0.8% of its DX11 performance, so neither is particularly a standout result in itself and it’s likely that Nvidia users will need to keep using DX11 versions of games until the company does something to sort out its DX12 drivers or comes up with a better chip design for the API.

Rise of the Tomb Raider

Rise of the Tomb Raider sees the continuation of Lara Croft’s adventures after the excellent reboot in 2013. Once again the 1080 puts in a reasonable turn here, however it should be noted that once again the Titan X struggled to maintain stability at its normal overclock, so I once again reduced it to a +100/+300 overclock on the core and memory respectively.

DX11 first then and service resumes with a stock to stock increase of 21.2% for the 1080 over the Titan X. Overclocked is a similar story with a 20.7% boost.

In DirectX 12 the familiar story continues. All scenarios suffer performance drops from their DX11 peaks with a 3.6% drop for Titan X at stock, a 2.6% drop for Titan X overclocked, 4% drop for 1080 at stock and a 3.2% drop for the overclocked 1080.

Stock to stock increase is a respectable 20.7% increase for the 1080 and 19.9% increase in overclocked form.

Minimum and maximum frame rates are similar for both API’s again.

DX11	Titan X Stock	Titan X Overclock	1080 Stock	1080 Overclock
Min FPS	23.78	30.82	31.61	37.77
Max FPS	71.82	63.84	66.59	69.05

DX12	Titan X Stock	Titan X Overclock	1080 Stock	1080 Overclock
Min FPS	31.11	34.71	34.44	40.09
Max FPS	48.16	53.6	60.72	66.7

So, what have we learnt? Well, at the start of this review I posed a question. Are we now in the era of truly capable 4K gaming? Realistically, unfortunately the answer is still a no from me.

Can we game at 4K? Absolutely. But as things stand, compromises still need to be made to do so. Either in putting up with the inherent problems of multi-gpu setups or in compromising graphical fidelity in the search for playable frame rates. The tests done in this review were on fairly punishing detail levels and this isn’t to say that you can’t have a good experience with a 1080 at 4K. The first thing I tend to do upon installing any game is turn off any kind of AA since it’s questionable whether it’s really needed at the resolution, but even without AA, we’re not yet into the realms of smooth 60FPS gameplay.

Certainly we’re getting closer, but we’re still not there yet and if you’re looking for 4K gaming and have a 980 Ti or Titan X, chances are the added cost isn’t going to be the wisest investment in terms of return for your dollars, pounds, euros or yen. Additionally, given that 1080’s are still in fairly short supply and (certainly in the UK) prices have been increasing, likely as a result of the collapse in the pound vs. the dollar since the UK voted to leave the European Union recently, it makes it a difficult sell for anyone who already has a top end card, particularly given that in all likelihood, Nvidia has a new Titan/1080 Ti coming sometime in the next 6 to 9 months while AMD has Vega chips also due in early 2017 to resume battle at the high end with Nvidia.

The 1080 is a good card, a great card even if you’re going to be looking for silky smooth, super high frame rates at 1080p or very good frames at 1440p. As for 4K? Certainly it’s the best available at the moment, but buyers should still be aware that there is still some way to go before we get what I’d consider to be a real “fire and forget, install a game, turn everything up to ultra and play at 4K” GPU.

Also worth considering (less of a problem for water or 980 Ti users) is that the overclocks I used for the Titan X are generally the ones I use when I’m going for maximum performance. Half of the Titan X’s memory chips are on the underneath of the card and have no cooling on them other than the air the case moves around inside itself. As such, I tend not to use that aggressive a memory overclock when gaming, simply because I have no way of monitoring the memory temperature. This is a problem for me given that the Titan X gets a decent amount of its performance boost from the memory overclock so in terms of the differential from overclock to overclock, it’s probably greater than represented here. As a rough rule of thumb, I’d probably achieve only about 50% - 60% of the performance boost I have here with the Titan X, whereas with the 1080, I’m pretty comfortable with the level of overclock.

Additionally, the blower design on the Titan X (as well as reference 1080’s) certainly makes for a louder experience. Than the MSI 1080 Gaming X’s Twin Frozr VI.

Ultimately, it’s a definite overall improvement on the Titan X in all areas, AIB 980 Ti’s will perform better than the Titan X anyway, but I’d still expect a reasonable boost in performance. Really though the 1080 is aimed at previous “medium chip” owners.

If you’re not looking for real 4K ultra gaming, the 1080 will be a great card for you depending obviously on the price performance point you’re looking at. For those that want the top end though? We’re still in compromise territory. I’ll be keeping it and selling the Titan X, but even so, I’ll still be buying a GP102 (Nvidia) or Vega (AMD) based card in the not too distant future I suspect.

Follow Wccftech on Google to get more of our news coverage in your feeds.