Take your Pascal on the go
We take a quick look at the new Pascal-based GTX 1080, GTX 1070 and GTX 1060.
Easily the strongest growth segment in PC hardware today is in the adoption of gaming notebooks. Ask companies like MSI and ASUS, even Gigabyte, as they now make more models and sell more units of notebooks with a dedicated GPU than ever before. Both AMD and NVIDIA agree on this point and it’s something that AMD was adamant in discussing during the launch of the Polaris architecture.
Both AMD and NVIDIA predict massive annual growth in this market – somewhere on the order of 25-30%. For an overall culture that continues to believe the PC is dying, seeing projected growth this strong in any segment is not only amazing, but welcome to those of us that depend on it. AMD and NVIDIA have different goals here: GeForce products already have 90-95% market share in discrete gaming notebooks. In order for NVIDIA to see growth in sales, the total market needs to grow. For AMD, simply taking back a portion of those users and design wins would help its bottom line.
But despite AMD’s early talk about getting Polaris 10 and 11 in mobile platforms, it’s NVIDIA again striking first. Gaming notebooks with Pascal GPUs in them will be available today, from nearly every system vendor you would consider buying from: ASUS, MSI, Gigabyte, Alienware, Razer, etc. NVIDIA claims to have quicker adoption of this product family in notebooks than in any previous generation. That’s great news for NVIDIA, but might leave AMD looking in from the outside yet again.
Technologically speaking though, this makes sense. Despite the improvement that Polaris made on the GCN architecture, Pascal is still more powerful and more power efficient than anything AMD has been able to product. Looking solely at performance per watt, which is really the defining trait of mobile designs, Pascal is as dominant over Polaris as Maxwell was to Fiji. And this time around NVIDIA isn’t messing with cut back parts that have brand changes – GeForce is diving directly into gaming notebooks in a way we have only seen with one release.
The ASUS G752VS OC Edition with GTX 1070
Do you remember our initial look at the mobile variant of the GeForce GTX 980? Not the GTX 980M mind you, the full GM204 operating in notebooks. That was basically a dry run for what we see today: NVIDIA will be releasing the GeForce GTX 1080, GTX 1070 and GTX 1060 to notebooks.
Mobile Pascal Specifications
These are not cut back parts in anyway – NVIDIA claims that performance between the desktop and mobile versions of these chips will essentially be identical. They are able to achieve that with a combination of voltage tuning, silicon binning and even a CUDA core configuration change in the middle of the stack.
| GTX 1080 | GTX 1080 (mobile) | GTX 1070 | GTX 1070 (mobile) | GTX 1060 | GTX 1060 (mobile) | GTX 980 (mobile) | GTX 980M | GTX 970M | |
|---|---|---|---|---|---|---|---|---|---|
| GPU | GP104 | GP104 | GP104 | GP104 | GP106 | GP106 | GM204 | GM204 | GM206 |
| GPU Cores | 2560 | 2560 | 1920 | 2048 | 1280 | 1280 | 2048 | 1536 | 1280 |
| Rated Base Clock | 1607 MHz | 1556 MHz | 1506 MHz | 1442 MHz | 1506 MHz | 1404 MHz | 1126 MHz | 1038 MHz | 924 MHz |
| Texture Units | 160 | 160 | 120 | 128 | 80 | 80 | 128 | 96 | 80 |
| ROP Units | 64 | 64 | 64 | 64 | 48 | 48 | 64 | 64 | 48 |
| Memory | 8GB | 8GB | 8GB | 8GB | 6GB | 6GB | 4GB | 4GB/8GB | 3GB |
| Memory Clock | 10,000 MHz | 10,000 MHz | 8000 MHz | 8000 MHz | 8000 MHz | 8000 MHz | 7000 MHz | 5000 MHz | 5000 MHz |
| Memory Interface | 256-bit G5X | 256-bit G5X | 256-bit | 256-bit | 192-bit | 192-bit | 256-bit | 256-bit | 192-bit |
| Memory Bandwidth | 320 GB/s | 320 GB/s | 256 GB/s | 256 GB/s | 192 GB/s | 192 GB/s | 224 GB/s | 160 GB/s | 120 GB/s |
| TDP (watts) | 180 | ~150 | 165 | ~115 | 120 | ~75 | ~150 | ~100 | ~75 |
| Peak Compute | 8.2 TFLOPS | 7.9 TFLOPS | 5.7 TFLOPS | 5.9 TFLOPS | 3.85 TFLOPS | 3.5 TFLOPS | 4.61 TFLOPS | 3.18 TFLOPS | 2.3 TFLOPS |
| Transistor Count | 7.2B | 7.2B | 7.2B | 7.2B | 4.4B | 4.4B | 5.2B | 5.2B | 2.94B |
| Process Tech | 16nm | 16nm | 16nm | 16nm | 16nm | 16nm | 28nm | 28nm | 28nm |
| MSRP (current) | $599/699 | N/A | $379/449 | N/A | $249/299 | N/A | N/A | N/A | N/A |
Both the mobile version of the GeForce GTX 1080 and the GTX 1060 are near identical to their desktop counterparts in terms of specifications with modest changes in clock speeds. The GTX 1080 that you’ll find in a gaming notebook will have 2560 CUDA cores and will run at clock speeds over 1.5 GHz. How is it possible to squeeze a GPU with a TDP of 180 watts into a mobile chassis? They aren’t - the mobile variant will likely have a TDP somewhere around 150 watts. I saw “around” because NVIDIA continues to be cagey about the specific TDP of the GPU, memory and power delivery systems for notebooks. I honestly don’t know why, these are reference specs meant to set a baseline minimum for performance and cooling on mobile platforms. Just as they did with the GTX 980 in gaming notebooks, NVIDIA is cherry picking GPUs from the line that have the least leakage, that perform the best at the lowest voltages possible. You could view it as the mobile market getting the “best” GPUs from TSMC.
The GTX 1070 is different though. The mobile version of the GTX 1070 actually has more CUDA cores than the desktop card, 2048 rather than 1920, leaving on more SM enabled on the die. Meanwhile, the base clock of 1.4 GHz is about 100 MHz lower than that of the desktop product. The goal was to create a different part that has (nearly) identical performance to the desktop GTX 1070 Founders Edition product, but would be able to run at a TDP in the 115 watt range rather than 165 watts. It’s an interesting move, and one that will likely create controversy around the brand of the GTX 1070. Could we see desktop variants with 2048 CUDA cores and lower clocks to enable low power, SFF add-in cards? It seems like it would make sense.
I still have more testing to do to see if the (nearly) identical performance claim sits well with me - if so then I see no harm in NVIDIA’s move here. If there are noticeable differences in more than a couple of gaming titles, we’ll debate the decision more directly.
All things considered, I do wish that NVIDIA had added the “M” suffix these parts: the GTX 1080M, GTX 1070M, GTX 1060M. Obviously NVIDIA wants to avoid the assumption that the 1080M is slower than a 1080, more in line with a 1070, but I see the added binning, and the change in the GTX 1070 mobile, as a reason to stick with the older naming scheme and instead educate users that the perf delta is gone.
There is still space here for a GPU under the ~75 watt TDP that I expect the GTX 1060 to have, leaving room for AMD and its Polaris GPUs to make some noise. You are very likely looking at a minimum price for a notebook based on these GeForce parts of $1300-1500. If AMD could work with a partner to build an $800-1000 machine with competent performance it has a chance to be successful.
New Features Coming to GeForce Gaming Notebooks
While the performance upgrade going from the GTX 980 to the GTX 1080 should be impressive enough to warrant the purchase of a new gaming notebook on its own, NVIDIA is also taking the opportunity to add some new features and capabilities to the platform along the way.
The most exciting of these to me center on displays. NVIDIA’s partners will be selling gaming notebooks based on Pascal that include 120Hz 1080p panels and 2560x1440 120Hz screens, all with G-Sync support. Until today, the best mobile panel I could find is a 1080p 75Hz G-Sync option, with some 4K 60Hz G-Sync screens creeping in as well. For gamers that are serious about frame rates, the 120Hz G-Sync option is going to be awesome. For those more concerned about resolution (which I think I would favor), the 2560x1440 60Hz G-Sync options will stand out. Update: it turns out the 2560x1440 screens are also available in 120Hz versions!
NVIDIA also improved its Battery Boost technology, the feature that enables better and longer gaming sessions in those times when you are away from a power outlet. As you should know by now, when running a gaming notebook that might normally draw 200+ watts while gaming without a power adapter, total power consumption is limited. Modern battery technology limits power draw to ~100 watts for the entire system, meaning the GPU must be pulled back to allow the CPU, screen, etc. to have room to function as well.
Battery Boost helps manage this by lowering in-game settings and enabling a frame rate limiter through GeForce Experience when running on the battery. All of this is configurable by the user if they wish to change default settings, but the defaults are managed by NVIDIA’s performance and experience labs to present the best gaming to the consumer. A game in which you might run at Ultra settings at 65 FPS while plugged in, might be turned down to medium settings at limited to 30 FPS. With that option enabled you should be able to play the game longer, though obviously these machines are still better off when tethered.
I actually tested Battery Boost back in 2014 with the GTX 980M release and found it work well. When paired with GeForce Experience it prevented bad user experience gaming sessions and the did increase gaming lifetime of the notebook.
The Swarm is Coming
NVIDIA couldn’t stop bragging about how many notebooks using Pascal were coming out this summer and fall. Lines like “every major OEM and system builder” and “quicker adoption than Maxwell” were muttered over and over. Clearly the team inside NVIDIA is proud of this accomplishment. On-hand at the event I attended I saw ASUS, MSI, Clevo, Razer and several others. I saw models using the GeForce GTX 1080, the GTX 1070 and the GTX 1060. Models as slim as 18mm and as light at 4 pounds were there to be fondled and gamed on.
That is quite an impressive list and I'm interested to see the range of designs and styles that are created with Pascal at the heart of these gaming notebooks.
VR Considerations
While half of the demo room at the mobile tech day catered to benchmarking notebooks on display, the other side had three dedicated VR stations, all with HTC Vive’s configured and running on mobile machines using Pascal. Every single gaming notebook built on the GTX 1080, GTX 1070 and GTX 1060 is “VR ready” and meets the recommended specifications for Oculus Rift and HTC Vive. While VR growth still has a long way to go to reach the levels we hope they will, this NVIDIA launch will make the idea of portable VR gaming even more attractive and accessible than ever before. If you can purchase a 4 lbs. machine that is 18mm thin that you can carry with you to your office, your friend’s house or a LAN event, then you have the ability to showcase your VR purchases or just bring it along for you enjoy on the go.
It’s a niche of a niche market, to be sure (gamers that want notebooks that want to use VR) but NVIDIA’s new Pascal products have it covered in a wide range of system designs.
SteamVR Performance Test - GTX 1070 in the ASUS G752VS OC Edition
Overclocking Mobile GPUs
Overclocking mobile graphics chip is always more complicated than it is with desktop components. Most importantly, the coolers used to keep the GPUs running smoothly differ dramatically from chassis to chassis. And the thermal constraints are more restrictive - if a cooler is limited to 150 watts, you don't have any room to arbitrarily adjust voltages, etc.
That said, both NVIDIA and its partners are pushing the overclockability of the Pascal GPUs in mobile chassis. It is going to be more of a case-by-case setup though; my ASUS G752VS OC Edition was able to accept an offset of just 100 MHz, well below the insinuated 300+ MHz from NVIDIA's slide.
Using the ROG Gaming Center I enabled manual settings and pushed the GPU clock offset to +100 MHz and +300 MHz on the memory. There is no control for voltage and NVIDIA told me that would be the case for all systems.
First, note that ASUS has overclocked the GTX 1070 by 50 MHz automatically when you install ROG Gaming Center, so the +100 MHz offset only results in a 50 MHz increase over out-of-box settings. Using Unigine Heaven and looping it for more than 10 minutes gets us to a stable clock speed and temperature for both configurations. The ASUS G752VS provides an average clock rate of 1570 MHz on the GTX 1070. That is above the 1442 MHz base clock but is well under the targeted Boost clock of 1645 MHz. This will vary from game to game, as we have always seen with NVIDIA's GPU Boost technology, but it seems likely that the GPUs will struggle to reach those clocks in constrained thermal environments.
When overclocked by another 50 MHz, the average clocks level out perfectly, hitting 1620 MHz over our extended test run. That is still under the 1645 MHz targeted Boost clock but still provides a performance increase across the board.
Quick Performance Preview
Having only just returned from the mobile tech day with NVIDIA, I haven’t spent much time benchmarking the system we have on hand, the ASUS G752VS OC Edition. NVIDIA did have notebooks at the event to use for benchmarking, but the pre-installed titles didn’t match up with my normal test suite and not being able to install our own games, drivers, manage setup or even use our Frame Rating performance analysis tools kept that from being a feasible option for us.
I will be following up on this story with a full review of the ASUS notebook, including a full range of gaming tests, but for now, I only wanted to verify that NVIDIA claims of performance parity between the GTX 1070 desktop and mobile SKUs was indeed true.
ASUS G752VS OC Edition with GTX 1070
Keep in mind there are significant differences between the platform on the GTX 1070 desktop results and the mobile results from the ASUS G752VS OC Edition.
- Desktop
- Intel Core i7-5960X (8-core)
- ASUS X99 Rampage
- 16GB DDR4
- Windows 10 Pro
- GeForce GTX 1070 8GB Founders Edition
- Mobile
- Intel Core i7-6820HK (quad-core)
- 32GB DDR4
- Windows 10 Pro
- GeForce GTX 1070 8GB (mobile)
In our three different 3DMark tests, looking only at the Graphics score to try to avoid bringing the CPU and platform differences into play, the mobile implementation of the GTX 1070 is within 5-6% of our desktop Founders Edition.
Our games testing shows a bit more variance. NVIDIA did warn us that the platform differences would result in some frame rate changes as a handful of titles still place emphasis on CPU performance. Take a look at Hitman and GTA V for example: our desktop GTX 1070 setup is 26% and 17% faster than the ASUS G752VS OC Edition running at 1080p. Other games track much more closely between the mobile and desktop variants: Rise of the Tomb Raider is within 1% and Dirt Rally is within 3% - very impressive! Fallout 4 and The Witcher 3 measure <10% slower than the 8-core desktop system with a GTX 1070 Founders Edition card.
To get a real apples-to-apples comparison I'll need to build a system around a Core i7-6700K soon, but even the few results we see matching performance between the two systems proves that NVIDIA has built some beefy GPUs in modest thermal envelopes. NVIDIA is once again pushing the envelope forward on what we should expect from gaming notebooks. I am looking forward to getting my hands on a GTX 1080 and a GTX 1060 based machine in the near future to really see if the full range is just as impressive.
Availability and Closing Thoughts
Based on what NVIDIA has told me and what I know from talking with the notebook vendors, the claims of immediate availability for notebooks based on these GPUs should stand. Both ASUS and MSI have had notebooks in their hands for a couple of weeks, just waiting for NVIDIA to pull the trigger and let them loose. If you looked closely over the past few days, you might have even seen them find their way into the wild a bit early!
Look, I know that gaming notebooks aren’t for everyone that reads PC Perspective. In terms of performance for your dollar, they are almost never a “good deal” when compared to a desktop PC. If you want the best possible gaming experience and you don’t need to be on the road with it, building a PC is still the best option. If you want portability with your gaming, if you frequent LAN parties or you want game during those boring family events at Grandma’s (shame on you), the a gaming notebook with as Pascal-based GeForce GPU is going to provide you the best experience anywhere.
Look for more reviews based on this hardware soon!
I’m not into notebook gaming,
I’m not into notebook gaming, but this is very impressive. Mobile 1070 within 10% of a desktop 1070 in something like The Witcher 3. Amazing stuff!
They won’t be in real world
They won’t be in real world use. You can’t make up,for the far better cooling desktop cards have. As the article states the boost is what to,look at. In actual gaming the clock difference will be much more. The mobile chips just won’t have the cooling to boost as much. Still impressive for mobile.
We’re there any rumours about
We’re there any rumours about AIOs that will use pascal?
Well the Digital Storm Aura
Well the Digital Storm Aura is available for preorder and I am fairly certain it has Pascal GTX1080 as an option.
The performance looks
The performance looks amazing.
We had to reach 16nm FinFETs
We had to reach 16nm FinFETs to see mobiles and desktops so close in performance. For people who can’t have desktops – I have a friend working on cruise ships – this is extremely important.
I’ll take a freesync enabled
I’ll take a freesync enabled 460 based laptop please. Hopefully it will be around $500 cheaper than 1060 with Gsync.
I know there are some people that need gaming laptops instead of desktops, but for the rest of us, I think it’s alot easier to justify an average performance laptop around $1000 to compliment our desktops than one of these incredibly impressive, but likely incredibly expensive parts.
So pascal on desktop were
So pascal on desktop were basically mobile parts masquerading as desktop parts, when can we expect actual desktop gpu’s from team green at a decent price?
Ryan could you do clock to
Ryan could you do clock to clock comparison with gtx980(desktop version would do just fine). A bit OC to gtx980 and underclock to gtx1070 mobile, if possible(~1.5GHz). Same to vrams.
Would be quite interesting to see most close to plain architecture changes between maxwell vs pascal.
MSI GT72s 6QE: Will I be able
MSI GT72s 6QE: Will I be able to upgrade to the new graphics cards?
These will cost Arm + Leg +
These will cost Arm + Leg + kidneys + eyeballs + every other organ including the hide. I’ll wait for Zen/Polaris and no organs sold for a more affordable solution.
Nvidia, like Intel, has the money to buy its way into more OEM devices so here is to hoping for a 6+ core Zen laptop CPU, or APU, to pair with Polaris graphics without breaking the bank. If AMD can get a 6 core Zen APU laptop SKU with Polaris graphics then that Zen APU will make for a more attractive and affordable laptop solution for many who would be stuck with Intel’s dog food graphics that comes with Intel’s SOCs, and some very overpriced dog food graphics at that.
Afaik, amd haven’t had very
Afaik, amd haven’t had very competitive high end gpu for mobile lately. Hawaii, Fiji while were performance vice very competitive, neither were suitable for laptops. And Tonga was not very competitive against gm204. So OEMs does not actually have amd alternatives for high end laptop.
Sadly though, I can’t really see this will change. After seeing Polaris 10 perf/W, I have doubts about Vega being suitable for mobile even if it use hbms.
Well intel being intel, it actually has most powerful igpu on mobile now. But intel just keep slapping it to most expensive i7s and i5s.
Polaris only appears to have
Polaris only appears to have bad performance to watt because the desktop parts are overvolted to hit clock speeds to hit the performance targets. I’m sure Polaris 10 at 1 volt and 1Ghz would be under 100 watts. Further hardware revisions with lower leakage should get gloflo closer to tsmc too (hopefully)
Polaris 10’s gaming
Polaris 10’s gaming performance/watt does not have to beat Nvidia’s gaming performance/watt as Polaris 10 has way more FP units and total Flops/$ relative to what Nvidia offers for its similar priced consumer SKUs. I do not measure performance in FPS, or any Frame time percentile metrics. I do more Blender 3d rendering and Intel’s graphics lacks the total GPU resources on its GPUs relative to Nvidia or AMD.
For Ray Tracing rendering all those extra Polaris 10 FP/Flops resources/$ come in very handy using every FP unit for ray tracing acceleration. That power figure metric is not as important as having the extra FP/Flops that Polaris 10 provides at a much lower $/Flops. I’m not rendering for FPS, as my single frame render times take minutes. I want the compute in AMD’s Polaris 10 GPUs and I do not care about the extra power usage to get that extra FP compute that comes with AMD’s affordable GCN GPUs.
It’s very easy to take a gimped down GPU and make it perform well on the simplified gaming mesh models/textures that are used for gaming and even Intel’s very gimped GPU/graphics do fine with some games, but for other graphics workloads like ray tracing calculations the more FP/Flops the better for getting those very realistic reflection/refraction, lighting/shadow, and AO/other effects and the single frame times measures in minutes and hours depending on the amount of FP compute available on the GPU/s.
Nvidia charges a whole lot more for its FP/Flops than AMD does, and usually for Nvidia getting all that extra compute in a single GPU requires purchasing Nvidia’s even more outrageously priced Pro GPU SKUs. For the price of 2 RX 480s for around $500 I can get as much/more FP/Flops as a Titan X(Pascal) that costs 1200. even the GTX 1070 is going to cost around $900 for 2, and I could get 3 RX 480s for $750, or 4 RX 470s for around $900. Also most non gaming graphics software can scale well across multiple GPUs, and for my rendering workloads it’s not about FPS its about FPUs.
I have had few issues with
I have had few issues with AMD Desktop Graphics Cards, but on Mobile, they still rely on the OEM to provide GPU drivers instead of directly supporting it like Nvidia does.
For that reason I never buy laptops with AMD graphics.
For the same reason (software) I also never buy anything with Nvidia Optimus. Software behaves unexpectedly and causes more problems for me.
Laptop OEM’s modding the
Laptop OEM’s modding the graphics drivers should be banned, with everyone forced to go with generic drivers that can be updated by the GPUs’ makers. Hopefully more of the games will go to using Vulkan/DX12 with the very simple close to the metal GPU drivers and all the other fancy feature controls turned over to the Games/OS-API/Software makers. With more simple and easy to manage GPU drivers at hand OEMs should not have any problems with the generic GPU drivers for Vulkan and DX12 and any of the fancy features like Multi-GPU load balancing will move to the OS/API and be accessible by all software, in addition to games software.
Hopefully more computing systems will be designed with HSA types of features in mind that are standardized across all device makers and markets so there will be less customizing by OEM’s necessary. Really any device specific hardware features setup and enumeration should be under the control of a standardized API so those OEM hardware/software/firmware development costs can be spread across the entire market for PC/Laptops and Mobile Table/Phone system under a fairly standardized set of Software/Firmware/hardware tools.
Perhaps OEMs still refuse to
Perhaps OEMs still refuse to rely on AMD for drivers, I don’t blame them
Out of character I paid $300 extra for 680M over 7970M based on drivers alone… not only have I had maybe three games in 5ish years I’ve had to force onto the dGPU being the only problem with Optimus (compare problems with Enduro) but history proved correct with AMD’s solder fail on their 6 and 7 series
Even if Polaris 10 kicks perf/price butt they have a lot of ground to make up before even the value conscious buyers like me will consider a Radeon
Typo…
“Pascal is still more
Typo…
“Pascal is still more powerful and more power efficient than anything AMD has been able to product.”
*should be “produce”
If you are only interested in
If you are only interested in gaming, Nvidia strips more of the compute out so the power savings goes hand in hand with the compute gimping from Nvidia. But watch out for VR gaming as that market wants to move more of the gaming compute to the GPU and away from the CPU to save on any Latency issues, and Nvidia may be forced to put some of that compute back.
And Powerful for gaming does not necessary equate to powerful for other compute uses as the popularity of AMD’s GPU SKUs for those other compute markets has shown, and still does for any new bit/coin mining algorithms that are yet to be implemented in ASIC form. Lot’s of extra FP compute on AMD’s SKUs also comes in handy for Ray Tracing acceleration on the GPU so users can forget about the need for any expensive multi-core workstation grade CPU SKUs from Intel for Ray Tracing workloads done on the GPU. AMD’s GPUs have always more compute so they use more power, and the VR gaming market will be using that extra compute even more for that market’s VR games.
Hello. Please share bios of
Hello. Please share bios of 1070 mobile. I cant find it 🙁