With CES kicking off this week, we have a spate of laptop-related announcements. As Intel is launching their Kaby Lake quad core (4+2) SKUs for laptops, so too are new video cards are being launched to use in those systems. To that end, today NVIDIA is taking the wraps off of their latest mobile video cards, the GTX 1050 Ti and GTX 1050 for laptops.

As you might recall from last summer, NVIDIA has reworked how they are positioning and promoting their laptop products starting with the GeForce 10 series. Rather than having separate laptop SKUs with their own (lower) specifications, the 10 series’ SKUs all have (nearly) the same specifications as their laptop counterparts. This means that the launch of most laptop parts has been shifted to something of a formality – we already know roughly what their specifications will be – but it still marks an important milestone for NVIDIA and their customers as these laptop parts finally become available.

To that end, NVIDIA and their partners are launching the laptop versions of the GeForce GTX 1050 Ti and GTX 1050. The original desktop versions of these parts were launched back in late October, built using NVIDIA’s GP107 GPU. The successor to NVIDIA’s now-venerable GM107 that kicked off the Maxwell generation, GP107 is NVIDIA’s mainstream performance, high-volume GPU. And GTX 1050 Ti and GTX 1050 in turn are the formal SKUs that will be the backbone of a number of mainstream gaming-capable laptops.

NVIDIA Laptop GPU Specification Comparison
  GTX 1060 GTX 1050 Ti GTX 1050 GTX 960M GTX 950M
CUDA Cores 1280 768 640 640 640
Texture Units 80 48 40 40 40
ROPs 48 32 16 16 16
Core Clock 1404MHz 1493MHz 1354MHz 1097MHz 914MHz
Boost Clock 1670MHz 1620MHz 1493MHz Undefined Undefined
Memory Clock 8Gbps GDDR5 7Gbps GDDR5 7Gbps GDDR5 5Gbps GDDR5 5Gbps GDDR5
Memory Bus Width 192-bit 128-bit 128-bit 128-bit 128-bit
VRAM 6GB 2GB/4GB 2GB/4GB 2GB/4GB 2GB/4GB
FP64 1/32 1/32 1/32 1/32 1/32
GPU GP106 GP107 GP107 GM107 GM107
Transistor Count 4.4B 3.3B 3.3B 1.87B 1.87B
Manufacturing Process TSMC 16nm Samsung 14nm Samsung 14nm TSMC 28nm TSMC 28nm
Launch Date 08/16/2016 01/03/2017 01/03/2017 03/12/2015 03/12/2015

Looking at the individual SKUs, the GTX 1050 Ti for laptops will be the full-fledged GP107 SKU. It ships with all 6 SMs enabled, for a total of 768 CUDA cores, just like its desktop counterpart. On the clockspeed front, officially it can boost up to 1620MHz; surprisingly this is some 228MHz (16%) higher than the desktop part it’s based on. NVIDIA has already played with the clockspeeds of some of the other laptop SKUs, so that it doesn’t match the desktop clockspeed isn’t unexpected, however this is the first time we’ve seen the clockspeeds increase like this.

Unfortunately I don’t have a good answer right now to explain the difference. GP107 is something of a special chip to begin with – it’s the only Pascal chip made over at Samsung on their 14nm process – so it already throws out part of the rulebook. NVIDIA of course bins parts for laptop/desktop usage, so it may just be that the better chips being held back for laptops can clock so much higher, or it could be that Samsung is producing better chips since the initial batches that defined the desktop launch a few months ago.

Either way laptop clockspeeds are a lot more variable from product-to-product regardless: the design of a laptop’s cooling system has a lot more to do with sustained clockspeeds than NVIDIA’s formal specifications. NVIDIA for their part doesn’t disclose the TDPs of their laptop SKUs, so it’s impossible to say just how much cooling is required to get close to the GTX 1050 Ti’s peak performance, but it’s likely not too far removed from the desktop part’s 75W TDP.


Unofficial GP107 Block Diagram

Moving on, paired up with the GP107 GPU in the GTX 1050 Ti for Laptops will be paired with 7Gbps GDDR5 for VRAM. As is usually the case with laptops, the amount of VRAM is an “up to” specification – in this case up to 4GB – so I’m expecting we’ll see both 2GB and 4GB laptops. This in turn is running on GP107’s 128-bit memory bus, so we’re looking at 112GB/sec of memory bandwidth.

Also launching alongside the GTX 1050 Ti for Laptops is its lower-end sibling, the GTX 1050 for Laptops. As with the desktop part, this is a GP107 SKU with 5 of 6 SMes enabled, trading off about 17% of the shading/texturing/geometry hardware in exchange for lower prices. However past this point, it more significantly diverges from both its desktop counterpart and faster laptop sibling. NVIDIA has disabled half of the ROPs, leaving it with 16 ROPs instead of the full 32 ROPs found on GP107. To date this is the second time NVIDIA has shipped a GeForce 10 series for Laptops part with a different core configuration, and the first time that modification has been the ROPs (and is why we say laptop parts are configured similarly to their desktop counterparts).

GTX 1050 for Laptops has a low enough SM count that I’m not fully sure what the impact is of disabling half the ROPs, as the ROPs likely weren’t always being fed to begin with. At the same time, NVIDIA has never done something quite like this before, as GTX 1050 retains its 1MB of L2 cache, something that doesn’t usually happen when ROPs are disabled. So this isn’t something that can be logically reasoned out, and we’ll have to see what the benchmarks find. However I think it’s safe to say that it will certainly drive a larger wedge in performance between GTX 1050 Ti and GTX 1050 than what we’ve seen on the desktop side.

In any case, unlike the GTX 1050 Ti, this SKU is going to be clocked much closer to its desktop counterpart, coming in at 1493MHz for the boost clock (~2.5% higher). On paper then, GTX 1050 for Laptops can deliver about 77% of GTX 1050 Ti for Laptops’ shader/geometry throughput, or 46% of its ROP throughput.

As for memory, things are unchanged from the GTX 1050 Ti. This means “up to” 4GB of GDDR5, running at 7Gbps.

Overall, given that these SKUs are meant to replace GM107 in so many designs, I’m very interested in seeing just what the real-world performance uplift is. Replacing the 3 year old GM107, GTX 1050 Ti for Laptops enjoys a significant clockspeed, ROP, and SM count advantage over parts like the GTX 950M/960M, so there’s quite a bit of potential, possibly more than we’ve seen in the desktop. Which if GP107 ends up being a major backbone part for midrange laptops like GM107 was for the past few years, would be very welcome news.

Wrapping things up, like the rest of the GeForce 10 series for Laptop parts, the new GTX 1050 series parts gain access to NVIDIA’s full laptop feature set. This includes the latest iteration of Battery Boost, G-Sync support (something that might be especially useful for these lower-performance parts), and overclocking support. As for laptops featuring these new video cards, NVIDIA’s partners will be shipping laptops with them this quarter – some as early as this week – and already several have been announced here at CES. Final pricing will vary by laptop, but broadly speaking, NVIDIA expects GTX 1050 for Laptops equipped systems to start at $699.

Source: NVIDIA

POST A COMMENT

20 Comments

View All Comments

  • Lonyo - Wednesday, January 4, 2017 - link

    Does Pascal even support DDR3? Reply
  • DanNeely - Wednesday, January 4, 2017 - link

    we'll probably have to wait for GP108 to show up to be sure. DDR3 (or presumably DDR4 now) only shows up in the lowest xx7 and xx8 GPU dies. So far there hasn't been any indication of it for GP107. OTOH if it was going to show up anywhere it'd be on laptop parts that've just started showing up, and potentially not until they had enough dies that binned with a failed GDDR5 controller that it wasn't cutting into sales of better variants and the initial 1050 mobile hype train is ended. OTOH GP108 leaked in benchmark tools at the same time as the rest of pascal last June; and xx8 GPUs are almost exclusively standard DDR; with the last exception being a 730 desktop variant. At this point though with the DDR3/4 price gap basically gone, I'd be shocked if they don't go with the latter because less ram starvation is free speedup. Reply
  • BrokenCrayons - Wednesday, January 4, 2017 - link

    There was an increase in electrical energy needed when moving from DDR3 to GDDR5. In power constrained situations, manufacturers may opt for DDR. After all, midrange GPUs for laptops are demanding a lot more electrical energy than they have over the previous decade (if you go by TDP anyway...which we all know isn't the best way to determine power consumption). For instance, the GeForce 8600M GT in my ancient Vostro 1500 was rated at 20W and it was a firmly midrange graphics card in its heyday. The 1050 is probably a LOT higher and you're still, on a relative scale, only getting midrange performance. Given how much technology has improved, that's really terrible from a mobility perspective. Reply
  • DanNeely - Wednesday, January 4, 2017 - link

    The speed v power tradeoff's always been there. The cut point in the lineup above which DDR isn't used at all has been steadily moving down the line though. Looking at mobile parts because they're far more power constrained, it's GM107, GK107, GF106, GT215. I stopped there because the naming convention changed for older nVidia gpu dies. This leaves if GP107 will support DDR like Maxwell and Kepler or leave it behind the way the xx6 and xx5 parts have in previous generations. I'm not sure which way this will end up swinging, on the one hand DDR4 offers a major bandwidth/power boost over DDR3 which increases the gpu performance ceiling before memory bandwidth becomes a hard wall. On the other hand, Pascal mobile parts are probably running at higher power and performance levels than older equivalents; which argues the other direction. Until a DDR4 card launches or nVidia says something one way or the other it's an open question.

    I've seen a little speculation that GP108 might end up GDDR only; I'm more skeptical about that though. With the power creep in pascal mobile, I wouldn't be surprised if there's a GDDR5 1040 at the ~40W level, but if they do that I'd expect lower power (or just cheaper) DDR4 versions below it.
    Reply
  • Hitokage_Tamashi - Wednesday, January 4, 2017 - link

    Actually, the 1060 was a 970M replacement, the 1070 was a 980M replacement. The 1050 is a direct 960M replacement, the 1050ti seems to be another 970M replacement, or perhaps a clone- it's faster than the 970M by a fair bit (sits between the 970M and 980M performance wise, hovers around 960 (non-M) performance levels iirc), while still being in sub $1000 machines. The 1050ti is definitely more exciting than the mobile 1050- the non-ti seems relatively meh at best, and 2 GB of VRAM isn't doing it any favors. Definitely can't go wrong for the price point though- $800 for this is a steal considering it's in a laptop. Reply
  • lazarpandar - Wednesday, January 4, 2017 - link

    It's a replacement in name only. In no technical way does the 1050 replace the 950m. The two are specced completely differently and have very different thermal outputs/power requirements. Reply
  • digiguy - Wednesday, January 4, 2017 - link

    From the first tests GTX 1050 is about 40-50% faster than the GTX 960M but 10-15% slower than GTX 965M (in the surface book with performance base) and than the desktop variant, while GTX 1050Ti is roughly on par with the desktop variant and with the GTX 970m. Reply
  • Hitokage_Tamashi - Wednesday, January 4, 2017 - link

    Did the Surface Book get a special 965M? The 965M was only ~20% (maybe 30%) faster than the 960M, whereas the 1050 soundly beats it, the 1050ti destroys it. The 960M was essentially an overclocked 750ti anyways Reply
  • fanofanand - Thursday, January 5, 2017 - link

    It's incredible that they are still shipping GPUs with 2GB of VRAM. Reply
  • mrcaffeinex - Friday, January 6, 2017 - link

    In the class of device that they are installing these mGPUs, 2GB of VRAM is probably sufficient for the majority of titles people will be playing at the resolutions these devices support. The mGPUs would be strained to perform in a situation where 4GB of VRAM was necessary outside of caching data so that it does not need to be read from system RAM as often. These are budget parts with 2GB of VRAM to keep costs down. Reply

Log in

Don't have an account? Sign up now