marți, 16 august 2016

MSI GeForce GTX 1070 Sea Hawk X

Manufacturer: MSI
UK price (as reviewed):
£499.99 (inc VAT)
US price (as reviewed): $519.99 (ex Tax)

Nvidia's board partners are now out in full force with multiple variants of the GTX 1070. Take MSI, for example; by varying the cooler used with the card and the clock speed of its specific SKUs, it's able to offer a whopping 11 different options just for this one GPU. With us today is one of its more unique offerings: the GTX 1070 Sea Hawk X. It's actually a joint venture with Corsair, effectively pairing the GTX 1070 with a custom all-in-one liquid cooler.For the privilege of out-of-box water-cooling, MSI is charging a hefty £500. With reference-style cards available for £375, this is a 33 percent markup, and only £100 away from a basic GTX 1080. A price-performance offering this is not, then, but water-cooling is of course much more about reducing temperatures and noise than adding performance, although there is the potential for higher overclocks too.
As this is the Sea Hawk X card, it comes not only with built-in water-cooling but also a factory overclock. MSI does list a GTX 1070 Sea Hawk on its website, which runs at stock speeds, although this non-X variant is hard to come across in retail. Back to the Sea Hawk X, and you get a base clock of 1,607MHz (1,797MHz boost), a 7 percent increase on reference speeds and the same as MSI's popular Gaming X card. Relative to the rest of the market, it's a fairly middle-of-the-road overclock, but should still provide a modest bump in performance. These speeds are applied when the card is operating in its OC Mode. Technically, the default frequencies (Gaming Mode) are 1,582MHz base and 1,771MHz boost and a third, Silent Mode offers reference clocks: 1,506MHz base and 1,683MHz boost. The modes are easily switched between using the MSI Gaming App – sadly there's no integration into Afterburner.The memory has been overclocked too, but only very slightly, from 2GHz to 2.025GHz, pushing the effective speed from 8Gbps to 8.1Gbps. This overclock is also only applied in OC Mode. Even so, it's nice the memory hasn't been totally ignored.
The actual card here looks just like your standard reference blower – 270mm long and 111mm tall with a closed shroud and a single radial fan. The plastic shroud is a little flimsy and plain looking, although the MSI logo along the top is backlit with a white LED, and the Gaming App can be used to control it with a few effects. The card is also fitted with a matt black metal backplate, strengthening it and improving the aesthetics over what a bare PCB would offer.
The power input is kept as a single 8-pin connector; there probably wouldn't be much benefit in upgrading this so that's fine. MSI also sticks to the tried and tested reference outputs with three DisplayPort headers, one DVI-D and one HDMI.
The reason MSI and Corsair have kept the radial fan in place is to cool the memory chips and VRMs. The cooling apparatus draws heat away from these via thermal pads and a metal plate, feeding a heatsink above which the fan cools, evacuating pretty much all air neatly out of the rear I/O panel thanks to the closed shroud. Cooling these components is of course important, but it's a shame not to see a more complete liquid cooling solution that manages them all and removes the need for a blower fan altogether. Then again, as long as the setup does its job, who are we to complain?
The GPU is of course the most critical component, and it's this that has been treated to Corsair's AIO liquid-cooling. The pump unit is fitted to the GPU, pulling heat away via the copper base that's outfitted with micro-fins for improved heat transfer. The resultant liquid feeds a slimline 120mm aluminium radiator – keeping the radiator small improves the case compatibility and helps to keep costs down too. As ever, the all-in-one loop is fully sealed and pre-fitted so the card requires no maintenance or assembly beyond screwing in the radiator.
A single 120mm fan is also supplied and comes pre-fitted to the radiator. It's a white LED model, so this adds a bit of pizazz, but it's powered by a standard 3-pin header and thus not temperature-controlled by the GPU. This is a shame, and it gives us the feeling of the cooler being somewhat incomplete. On the other hand, it does make the fan dead simple to replace, and opens it up to fan control through dedicated controllers or your motherboard.
The PCB also looks like a standard reference board, with no fancy features and a 4+1 phase power layout that doesn't make use of upgraded components like MSI's Gaming range of cards does. The card does support SLI, although it's likely to look messy pairing up two of these, and you'd need to ensure adequate spacing between them to give the blower fan of the top card room to breathe.


  • Graphics processor Nvidia GeForce GTX 1070, 1,607MHz (1,797MHz boost)
  • Pipeline 1,920 stream processors, 120 texture units, 64 ROPs
  • Memory 8GB GDDR5, 8.1GHz effective
  • Bandwidth 259.2GB/sec, 256-bit interface
  • Compatibility DirectX 12, Vulcan, OpenGL 4.5
  • Outputs/Inputs 3 x DisplayPort 1.4, Dual Link DVI-I, 1 x HDMI 2.0b
  • Power connections 1 x 8-pin PCI-E, top-mounted
  • Size 270mm long, 111mm tall, dual-slot (plus radiator and 120mm fan)
  • Warranty Three years

Test Setup

Our GPU test rig uses Intel's X99 platform. Specifically, we use the Asus Rampage V Extreme paired with an Intel Core i7-5960X. This combination gives us 40 PCI-E 3.0 lanes to play with, meaning that even if we test 4-way SLI set-ups, we'll have available the maximum possible bandwidth. With eight cores and 16 threads via HyperThreading, the CPU ensures even the most multi-threaded games won't be bottlenecked, and we've overclocked it to 4.2GHz as well with a baseclock of 127.3MHz, a multiplier of 33 and a vcore of 1.25V, further alleviating CPU bottlenecks. The CPU is paired with 16GB (4 x 4GB) of Corsair Vengeance LPX DDR4 clocked at 2,667MHz and cooled by a Corsair Hydro Series H75.The rig is powered by the Corsair AX1500i, which has enough wattage and 8-pin PCI-E power connectors to deal with any multi-GPU set-up we care to throw at it. It's all housed inside a Corsair Graphite Series 760T, which is spacious enough for 4-way set-ups with any card. The final component is an OCZ Vector 180 960GB SSD, giving us plenty of space for game installs.Our results come from a mix of built-in benchmarks and custom, manually-played sections, depending on the game. We typically use FRAPS to record the average and minimum frame rates, but may use a game's own benchmark measurement tools in the event FRAPS doesn't work (e.g. with certain DX12 titles). All benchmarks are repeated at least three times to ensure consistency. We test at 1,920 x 1,080 (1080p) and 2,560 x 1,440 (1440p), as well as at 3,840 x 2,160 (4K) with higher end cards. The operating system is an up-to-date copy of Windows 10 64-bit.

Test System

  • CPU: Intel Core i7-5960X at 4.2GHz (33 x 127.3MHz)
  • Motherboard: Asus Rampage V Extreme
  • Memory: 4 x 4GB Corsair Vengeance LPX 2,800MHz DDR4 (at 2,667MHz)
  • PSU: Corsair AX1500i
  • SSD: OCZ Vector 180 960GB
  • Case: Corsair Graphite Series 760T
  • CPU cooler: Corsair Hydro Series H75
  • Operating system: Windows 10 64-bit

AMD Graphics Cards

  • AMD Radeon RX 480 8GB - 1,266MHz GPU, 8GHz GDDR5 (Radeon Software Crimson 16.6.2 Beta)
  • AMD Radeon RX 480 4GB - 1,266MHz GPU, 8GHz GDDR5 (Radeon Software Crimson 16.7.2)
  • AMD Radeon R9 Fury X 4GB - 1,050MHz GPU, 1GHz HBM (Radeon Software Crimson 16.6.1)
  • Sapphire Radeon R9 390X Nitro 8GB - 1,080MHz GPU, 6GHz GDDR5 (Radeon Software Crimson 16.6.1)
  • Sapphire Radeon R9 380 ITX Compact 4GB - 1,000MHz GPU, 5.8GHz GDDR5 (Radeon Software Crimson 16.6.1)

Nvidia Graphics Cards

  • EVGA GeForce GTX 1080 FTW 8GB - 1,721MHz GPU (1,860MHz boost), 10GHz GDDR5X (GeForce 368.39)
  • Asus GeForce GTX 1070 Strix 8GB - 1,657MHz GPU (1,860MHz boost), 8GHz GDDR5 (GeForce 368.39)
  • MSI GeForce GTX 1070 Sea Hawk X 8GB - 1,607MHz GPU (1,797MHz boost), 8.1GHz GDDR5 (GeForce 368.81)
  • Nvidia GeForce GTX 1060 Founders Edition 6GB - 1,506MHz GPU (1,708MHz boost), 8GHz GDDR5 (GeForce 368.64)
  • Nvidia GeForce GTX 980 Ti 6GB - 1,000MHz GPU (1,075MHz boost), 7GHz GDDR5 (GeForce 368.22)
  • Nvidia GeForce GTX 980 4GB - 1,126MHz GPU (1,216MHz boost), 7GHz GDDR5 (GeForce 368.22)
  • Asus GeForce GTX 970 DirectCU Mini 4GB - 1,088MHz GPU (1,228MHz boost), 7GHz GDDR5 (GeForce 368.22)
  • Palit GeForce GTX 960 Super JetStream 2GB - 1,279MHz (1,342MHz boost), 7.2GHz GDDR5 (GeForce 368.39)

Ashes of the Singularity

This is effectively the game that introduced the world to DirectX 12, with it offering one of the very first easily repeatable benchmarks. It's a real-time strategy game that features large scale battles with hundreds of units and individual ballistic calculations for every shot. In short, there's a lot going on, and the game is developed specifically to exploit DirectX 12 capabilities like large draw call counts and asynchronous compute (scheduling and executing multiple queues all at once). The game even supports the API's Explicit Multi-Adapter feature, allowing AMD and Nvidia cards to be paired up multi-GPU rendering scenario.
Using the DirectX 12 version, we run the game's built-in benchmark using the 'High' preset, as we've found settings above this to be too extreme. We record the first 30 seconds of the benchmark, as this includes the most challenging part and is representative of the whole thing.

Fallout 4

From our Fallout 4 review:

'In both theme and structure Fallout 4 is very similar to 3. The dangerous wastes of the commonwealth wrap around Boston's skyscrapers like an exclusion zone around a leaking reactor, and the eerie juxtaposition of post-apocalyptic horror with 1950s' optimism is threaded through with references to America's emergent history, though this time the focus is on revolution and abolitionism, rather than Fallout 3's deconstruction of Washington's political infrastructure. Bethesda counter these similarities by giving Fallout 4 a much more vibrant aesthetic. Gone are the Instagram-filtered tones of Fallout 3, replaced with a colour palette that better reflects how light works. During the day, the Wasteland is remarkably bright and colourful, from the garish scarlet outline of a Red Rocket gas station, to the motley tarpaulins and neon signage of Diamond City, Boston's central urban hub. At night, those colours fade away to a moody blue, and it's then you start to feel the ghosts of the apocalypse around you - the stillness of the air, the cracked roads and crumbling buildings, the dozens of pre-war skeletons scattered about like children's toys, twisted and broken. The mannequins -oh god the mannequins - standing in dusty corners and shattered shop-fronts, always staring, always judging.'
Our Fallout 4 benchmark is a 30-second FRAPS recording of a manual playthrough, where our character runs forward through a woodland area just outside the Corvega Automation Plant. The scene is very challenging relative to the rest of the game, with massive draw distances and complex volumetric lighting. This means the results below are not representative of typical gameplay, but rather of the most challenging points in the game. We test at the game's 'Ultra' preset, the highest available, and v-sync is disabled in the game's .ini file.


From our Hitman review:

'I enjoyed Hitman, and I think that the episodic nature is going to keep me coming back to the game again and again to pull and prod at the game's systems. If you're the type of person that wants to devour a game whole, you might want to wait till the disc based release coming next year. It's hard to put a score on what is, by its own admission, an unfinished game, and with that much still up in the air I'm not able to slap one of our recommended stickers at the bottom of the review - but this tiny slice is the finest Hitman title so far, and if they can keep the quality this high, I think a lot of people will enjoy it.'
We test Hitman using DirectX 12 and the highest available detail setting for each option, with anti-aliasing set to FXAA. The game has a built-in benchmark, which is what we use, recording performance over the first 90 seconds - essentially all of it.

Tom Clancy's The Division

From our Tom Clancy's The Division review:

'At its core, The Division plays like a traditional third-person cover shooter. Press space when near an obstacle, and your agent will stick to it like a Velcro spider, making them all-but invulnerable to attack. But enemies are similarly difficult to hit when in cover, and they also attack aggressively and use grenades to flush you out. Hence, your team needs to out-think as well as outgun your opponent, using one or two agents to lay down suppressive fire, while the others relocate to a vantage point or flanking position to take them out.'
We record pretty much the entirety of The Division's built-in benchmark with FRAPS set to 90 seconds. This ensures we capture the more challenging second half, where the cards really begin to strain. We use the game's own 'Ultra' preset in order to make it as challenging as possible.

Total War: Warhammer

From our Total War: Warhammer review:

'Total War: Warhammer is something genuinely very special. Somehow the team at Creative Assembly have managed to build a game that's the real deal both for the Total War faithful and the Warhammer fans coming to the franchise for the first time.'
For this game, we use a custom-made benchmark that shows a heavily populated battle between the Empire and the Greenskins. In the interest of full disclosure, this benchmark was supplied by AMD. This is another DirectX 12 test, and since FRAPS does not currently work with this benchmark we rely on its built-in reporting, which still give us an average and minimum frame rate. We test with the 'Ultra' preset, the highest available.

The Witcher 3: Wild Hunt

From our The Witcher 3: Wild Hunt review:

'But these flaws don't even come close to undermining CD Projekt's achievements in the Witcher 3. It's the culmination of a lengthy improvement process over three games. The Witcher proved they could make an interesting game, and the Witcher II proved they could make an intelligent and entertaining game. The Witcher 3 does something entirely different. It relieves CD Projekt of the burden of proof, and places it squarely on the shoulders of every other RPG developer working in the industry today. That's how good it is. Consider the bar officially raised.'The Witcher 3 is an extremely demanding game. As such, we test using the 'High' preset rather than the 'Ultra' one, as the visual difference between them is small. However, we disable Nvidia HairWorks due to its massive performance impact. We also use the 'High' preset for the postprocessing effects, which is the highest possible setting. The benchmark is a 45-second recording of a manual playthrough in the level Lilac and Gooseberries, where we run on horseback up to and through a village just after Geralt has first encountered a Griffin. There's plenty of water, shadows and lighting to deal with and a good amount of people, buildings and foilage to render as well.

Unigine Valley 1.0 Benchmark

Unigine's free Valley 1.0 benchmarking tool works well as a graphics benchmark as it is GPU limited and is thus incredibly taxing on the GPU whilst placing the CPU under very little stress. Unigine's scoring system is effectively linear: a card with 2,000 points is considered twice as fast as one with 1,000 points, and half as fast as one with 4,000 points. As such, you can easily replicate and run the test on your own system to gauge roughly how big a difference an upgrade would likely make for you. Currently, Nvidia hardware tends to fare much better than AMD's in this test, so it is mostly useful for comparing AMD cards with other AMD cards and likewise with Nvidia.

Power Consumption (Idle and Gaming)

In order to get an idea of a GPU's power draw, we run our Unigine Valley 1.0 benchmark, which is our most GPU limited test. We use a watt meter to measure the maximum total system power draw during the test, and also take an idle reading at the Windows desktop (3,840 x 2,160).

Thermal Performance (Idle and Gaming)

Thermal output is also measured using Unigine's Valley 1.0 benchmark. The GPU is installed in a closed, but fairly well ventilated case (Corsair Graphite 760T) with HDD cages removed and the case fans at full speed. We leave all GPU fan profiles and settings as they come and leave the benchmark running for ten minutes so that temperatures plateau. We record the peak GPU temperature using GPU-Z, and present the data as the delta T (the difference between the GPU temperature and the ambient temperature in our labs). We also take an idle reading at the Windows desktop (3,840 x 2,160).

MSI GeForce GTX 1070 Sea Hawk X Overclocking

As ever, we bumped to power limit and temperature limit to their maximum values: 105 percent and 92°C respectively. We also increased the voltage threshold to 100 percent; this is a liquid-cooled card, after all.Eventually, we managed to add 110MHz to the core, giving us new speeds of 1,717MHz base and 1,907MHz boost – a 7 percent increase, going by base clock. The core was able to peak at 2,100MHz, but was most often in the 2,025-2,050MHz range. The memory, meanwhile, was stable all the way to 2.35GHz, a 15 percent jump that took the effective speed to 9.4Gbps.

Performance Analysis

Performance is pretty much in line with the other overclocked GTX 1070 we've seen, the Asus Strix model, although that does come out a touch faster (just over 1 percent on average) thanks to its higher overclock. The MSI card here offers close to 10 percent more performance than a reference GTX 980 Ti or R9 Fury X, although with the latter the advantage is considerably less in DirectX 12 games and this is balanced by much better performance in the DX11 titles. Ultimately, this card will blast through games at 1080p, have very few issues remaining smooth at 1440p and will even make a good go of 4K, although at this resolution you'll have to dial down the details occasionally, as the Fallout 4 and The Division tests show. It will also probably offer strong VR performance, but until we're able to benchmark that effectively we won't know for sure.Clock speeds in the card's OC Mode were not constant, but averaged out to 1,900MHz or thereabouts. Dropping to Gaming Mode saw 1,886MHz as the most common speed and the card was boosting to between 1,797MHz and 1,835MHz in Silent Mode. Clearly, there is very little difference between the three modes, and the Unigine Valley results expose this clearly with only a 3 percent difference in scores between Silent Mode and OC Mode.
The power consumption figures are impressively low, especially considering there's a pump to deal with – the card probably has a relatively strict power limit, which would also explain the fluctuating clock speeds. We did remove the rear 140mm fan from our system to accommodate the radiator here, but as this comes with a 120mm fan of its own this shouldn't make too much difference to the overall numbers. Either way, performance per watt is definitely high.The benefit of water-cooling is clear in the delta T results – 19°C is a fantastic result, and 12°C less than the R9 Fury X, which is the logical comparison. As usual, there is no real difference in power consumption or temperature between the different clock speed modes.While the water-cooling clearly delivers in terms of temperatures, on the noise front it's not so good. The issue isn't the pump, which is quieter than Corsair's recent offerings and definitely quieter than the H75 unit in our test system. Nor is it the radial blower fan, which operates at a fixed speed of 33 percent (about 1,100 RPM and the minimum you can set it to) in Silent Mode and Gaming Mode or 36 percent (about 1,200 RPM) in OC Mode – in all instances it is practically inaudible. That 120mm fan, however, is certainly very noticeable. However, the big caveat is that it runs by default at full speed. Clearly, as the delta T results show, you could lower the speed significantly while still maintaining healthy temperatures, and it wouldn't be all that difficult to do so.
Overclocking the card netted us performance uplifts often above 10 percent and put the card just ahead of the GTX 1070 Strix OC when it too was overclocked, although in real terms the difference between them is none. Power consumption rose by about 20W, with the system now peaking at 322W. Clock speeds peaked at 2,100MHz, but were more commonly at 2,025MHz or 2,050MHz. The delta T increased by just 1°C, and since both fans were running at fixed speeds there was no change in noise.


The MSI GTX 1070 Sea Hawk X is a unique product – it's the only all-in-one liquid-cooled GTX 1070 on the market. Clearly, if you just want maximum GTX 1070 performance for as little money as possible, or you're happy with an air-cooled card (and there are plenty to choose from), this isn't the card for you.What the GTX 1070 Sea Hawk X purports to offer, however, is all (or most) of the benefits of a proper, custom water-cooling configuration without any of the hassle or warranty voiding that comes with it and at a lower price. And for the most part, it succeeds. As the temperature chart shows, the card offers superior cooling to anything you'll see outside of a custom loop. It doesn't require any maintenance or assembly beyond the radiator installation, and has a three year warranty too, so it ticks those boxes as well.In terms of cost, a Founders Edition card, waterblock and backplate will already run you close to £500, and that's before you've added a pump, tubing, fittings, reservoir and radiator to the mix. Unless you already have a custom loop capable of slotting a GTX 1070 neatly into, there's simply no cheaper way to water-cool this GPU. NZXT's Kraken G10 and a cheap AIO cooler could work, but it won't look as good and this route adds installation hassle as well as potential issues with keeping the VRM and memory chips cool. Our only real gripe with this card is the noise, or rather the inability to control the noise by default. The impressively low temperatures are all well and good, but much of the reason for choosing liquid over air is noise and while the card isn't loud with the fan at full pelt it is more audible than we'd like. It's not the end of the world, however. As we said, there are advantages to this – being able to replace the fan easily, for example, and being able to hook it up to any fan controller you like. Even a simple Molex adaptor could be used to limit the speed, or alternatively the fan control suites of modern motherboards, especially those from Asus and MSI, are excellent nowadays. Still, in the wake of Asus's FanConnect feature that has fan headers onboard the graphics card to allow for GPU temperature controlling of standard fans, not having something similar here is a real shame. Being able to control the fan (and its LED) through MSI Afterburner or the Gaming App, for example, would be a real boon. Speaking of software, we'd like to see a more integrated approach, and better use of the clock speed modes to produce tangible differences between them. Even so, the Sea Hawk X is still a strong product for what it is. It fills the niche well for those looking for a low-cost (relatively), low-hassle alternative to full liquid-cooling for the GTX 1070.

Niciun comentariu:

Trimiteți un comentariu