Nvidia’s GeForce GTX 1050 and GTX 1050 Ti graphics cards unveiled

Nvidia’s GeForce GTX 750 Ti has long enjoyed a reputation as the go-to card for the gamer on a budget. The mini-Maxwell offered solid performance for a modest price, and it could run off nothing more than PCI Express slot power thanks to its lightweight TDP. All was well until the Radeon RX 460 came along. AMD’s pint-size Polaris card offers higher performance than the GTX 750 Ti does, and it’s also capable of droppping into prebuilt systems without a PCIe power connector. Ante, upped.

With the RX 460’s arrival, it was only a matter of time before Nvidia fired back with a Pascal-powered budget brawler of its own. To counter Polaris 11, Nvidia is unveiling not one, but two cards this morning: the GeForce GTX 1050 and the GeForce GTX 1050 Ti. These cards use a new Pascal GPU, GP107, that promises to cause quite the rumble in the sub-$150 price bracket. Many immensely popular games like Dota 2, Rocket League, Overwatch, and Counter-Strike: Global Offensive need only modest graphics cards to run well, and the huge audiences for those e-sports titles have graphics card makers clamoring for mind and market share at that affordable price point.

EVGA’s take on a GTX 1050 Ti

As you’ve probably already guessed, gamers building their first PCs or adding a graphics card to dad’s Dell won’t need to shell out a lot for the privilege of owning a GTX 1050 or GTX 1050 Ti. The GTX 1050 rings in for the same $110 price tag as the Radeon RX 460 2GB card did at launch, and the GTX 1050 Ti will go for $140 and up.

Sadly, we don’t have the usual Chiclet-filled block diagram of GP107 to share with you today, but we do have some basic numbers to talk about, at least. Let’s consider how GP107’s key specs stack up against past and present budget-friendly options from Nvidia and AMD:

  ROP

pixels/

clock

Texels

filtered/

clock

(int/fp16)

Stream

processors

Rasterized

triangles/

clock

Memory

interface

width (bits)

Estimated

transistor

count

(Millions)

Die size

(mm²)

Fab

process

GM107 16 40/40 640 1 128 1870 148 28 nm
GP107 32 48/48 768 2 128 3300 132 14 nm
GM206 32 64/64 1024 2 128 2940 227 28 nm
Polaris 11 16 56/28 896 2 128 14 nm
Bonaire 16 56/28 896 2 128 2080 214 28 nm

GP107’s resource allocation puts it somewhere between the GM107 chip on board the GTX 750 Ti and the GM206 part that powers the GTX 950 and GTX 960. This Pascal chip has more muscle in most every regard than GM107, but it fits nearly double the transistors in less die area. Moore’s Law marches on, folks. By the way, that’s not a mistake in the table above. Every Pascal part we’ve seen so far has been fabricated on TSMC’s 16-nm FinFET process, but GP107 is built on an unspecified 14-nm process. Samsung and GlobalFoundries come to mind as potential sources, but Nvidia hasn’t confirmed anything to that effect. We’ll have to wait and see.

The fully-fledged GP107 chip and its 768 stream processors will take flight aboard the GTX 1050 Ti. Nvidia pairs that chip with 4GB of GDDR5 RAM running at 7GT/s, and it specifies 1290MHz base and 1392MHz boost speeds for the smallest Pascal in this form. The GTX 1050 comes with one of its Pascal streaming multiprocessor units disabled, dropping the SP count to 640 and the texturing unit count to 40. It also boasts 2GB of GDDR5 RAM. Perhaps to compensate a bit for those cuts, the GTX 1050 runs at 1354MHz base and 1455MHz boost speeds. Both cards offer 32 ROPs and a 128-bit memory interface, and both will dissipate 75W in operation.

The handsome Asus Dual GTX 1050 Ti

Thanks to that modest TDP, neither GTX 1050 needs an external PCIe power plug to do its thing. Some partner cards will come with beefy-looking twin-fan coolers on board, to be sure, and those cards might have external power inputs, but at least builders will have an option of a drop-in graphics card with both models.

Much like the GTX 750 Ti before them, the GTX 1050s should bring a substantial performance boost to the crucial plug-and-play graphics card market. To see just how much of a performance boost we should expect, we did some simple math to see how each card might perform in a few theoretical measures of graphics prowess. Have a look:

  Peak pixel

fill rate

(Gpixels/s)

Peak

bilinear

filtering

int8/fp16

(Gtexels/s)

Peak

rasterization

rate

(Gtris/s)

Peak

shader

arithmetic

rate

(tflops)

Memory

bandwidth

(GB/s)

GeForce GTX 1050 47 58/58 2.9 1.9 112
GeForce GTX 1050 Ti 45 67/67 2.8 2.1 112
GeForce GTX 750 Ti 17 43/43 1.1 1.4 86
GeForce GTX 950 24 57/57 2.4 1.8 106
GeForce GTX 960 38 75/75 2.4 2.4 112
Radeon RX 460 19 67/34 2.4 2.2 112

On paper, the GTX 1050 and GTX 1050 Ti clobber the GTX 750 Ti in pixel fill rates, texture filtering rates, and polygon throughput. In fact, the GeForce GTX 950 and GTX 960 are the more sensible points of comparison. The GTX 1050 duo can move many more pixels per second than those GM206-powered cards can, and they’re pretty closely matched in every other regard. Impressive stuff, considering that the GTX 950 used to sell for around $160 and that GTX 960 4GB cards moved for north of $200. The RX 460’s theoretical figures keep it in the running with those GeForces in most every regard save pixel fill rate, so it’ll be interesting to see how Polaris 11 and GP107 stack up.

Nvidia didn’t send GTX 1050 or GTX 1050 Ti cards to the press for review ahead of today’s launch, so we’re mostly left wondering how those theoretical figures translate into real-world performance. We do have some internal numbers from Nvidia for the GTX 1050, at least. That card is purportedly capable of running Grand Theft Auto V at 62 FPS and Gears of War 4 at 65 FPS on a 1920×1080 display at medium settings, while esports titles like Dota 2 and Overwatch should run above 60 FPS at high settings. Our sense of those figures is that the GTX 1050 and the RX 460 2GB will be close competitors.

Gigabyte’s nifty twin-fan GTX 1050 Ti

We don’t have a similar set of internal performance data to pigeonhole the GTX 1050 Ti with, but our friends at Asus say the card is capable of delivering GTX 960-class performance for 1920×1080 gaming. If that’s the case in practice, Nvidia could have a real winner on its hands. AMD’s $140 RX 460 4GB couldn’t quite catch a GeForce GTX 950 in our review, and the GTX 1050 Ti appears to offer even more performance than that card does.

Despite showing off some illustrations that star an adorably stubby GTX 1050 reference card, Nvidia won’t be releasing a Founders Edition version of either GTX 1050. Instead, it’s up to the company’s board partners to bring these cards to life. Nvidia’s partners already have a veritable fleet of custom cards prepared to join battle with the RX 460, and many of those don’t seem much longer than the PCI Express slot itself. Keep an eye on our news page throughout the day as we learn more about those products. We’ll be working to get one or both of Nvidia’s latest into TR’s labs for a full review as soon as we can, too—stay tuned.

Comments closed
    • novv
    • 3 years ago

    RX460 4GB it’s better than a standard GTX960 2GB on most of the newest DX12 titles. And I’m saying this not by looking at the reviews, but testing it. So please give me a break with that “RX460 when tested was slower than GTX950”. Please be fair and retest the cards with the latest drivers and LATEST GAMES!!!

      • Jeff Kampman
      • 3 years ago

      Setting up a test suite that’s favorable for AMD cards shows favorable results for AMD cards? I’m shocked.

      • Freon
      • 3 years ago

      Is it the fact a game is either Gaming Evolved or TWIMTBP, or is it the DX version they target?

      I’m not yet convinced DX11/DX12 matters as much as who the developer partnered with.

    • Ninjitsu
    • 3 years ago

    Well, Vega hasn’t materialized so far, either.

    • AnotherReader
    • 3 years ago

    Overclocking results on these should give a fair indication of the performance advantage of TSMC’s first generation finfet process over that of Samsung’s.

    • LoneWolf15
    • 3 years ago

    Jeff,

    When you guys test the 1050 and 1050Ti, can you test 10-bit H.265 HEVC decoding?

    Looking for a low-power card that will do this in my HTPC. Intel’s Braswell which built in Broadwell graphics doesn’t have all of the features in its driver of the i-Series CPUs (thanks Intel for waiting six months to tell me that after going back and forth, and telling no-one else so I can’t make it scale to match my TV screen), and while I was interested in the AMD RX 460, I’ve just heard about too many issues involving flickering to feel comfortable.

      • RtFusion
      • 3 years ago

      I second this!

    • Chrispy_
    • 3 years ago

    Huzzah!

    Now the desktop niche for these is tiny because people on a budget will be picking up older GCN cards and GTX 900-series for an absolute steal – it’s literally aimed at the performance/Watt crowd.

    By which I mean laptop people.
    Bring on the 1050 laptops plx!

      • ImSpartacus
      • 3 years ago

      What about the crowd that wants as much performance as possible without any power connectors?

        • Chrispy_
        • 3 years ago

        That’s exactly the tiny niche I was talking about, the performance/Watt crowd. “as much perfomance at 75W as possible”.

        What did you think I meant in the context of a 75W Pascal launch? 🙂

    • DragonDaddyBear
    • 3 years ago

    Here’s to hoping that rumor of the RX465 isn’t just a rumor.

      • BurntMyBacon
      • 3 years ago

      There is pretty big gap between the RX460 and the RX470 prices that it could easily slot into.

      • watzupken
      • 3 years ago

      The cut down RX 470 is likely going to be a stop gap solution for AMD since the chip is likely more pricey to produce than a GTX 1050Ti.

    • deruberhanyok
    • 3 years ago

    These look like the 3gb/6gb 1060 PCBs and fans, which looked like the 950 and 960s.

    With a lower TDP, how about some single slot designs? or low-profile? or both?

      • I.S.T.
      • 3 years ago

      DUnno about you, but I prefer the bigger coolers because they keep the chip cooler. That results in a longer lasting card. I’ve been using my card for three years, and if I hadn’t overspecced the cooling on it(I went for literally the best aircooler of the bunch according to several tests I read), who knows if it’d last as long as it does. I will likely be using it for at least two more years.

        • UberGerbil
        • 3 years ago

        Which is great if it fits in the case, but the people asking for low-profile and/or single-slot designs generally aren’t doing so for some arbitrary aesthetic reason: in most cases, they’re asking because such designs open up a world of smaller form factor systems to (relatively) powerful dGPUs. The thermal complications are a given in those situations and are accepted (and, in some cases with heroic measures, mitigated) as part of the trade-off for having a pocket rocket.

      • Shobai
      • 3 years ago

      I came to ask the same question…

    • chuckula
    • 3 years ago

    Judging by the beefiness of the chips and not necessarily based on the pros/cons of the different GPU architectures here’s a rough shakeout of this segment:

    Rx 480 ~= full GTX-1060 > Rx 470 ~= 3GB GTX-1060 > GTX-1050Ti > Rx 460 ~= GTX-1050.

    Right now the best moves for AMD would be to get most of the 4GB Rx 460 model’s prices down to the roughly $100 range to show some advantage over the GTX-1050 and do a smaller price cut to the Rx 470 in hopes of attracting buyers up-market from the 1050Ti.

      • Generic
      • 3 years ago

      (GTX 1060 6 GB ≈ RX 480) > (GTX 1060 3 GB ≈ RX 470) > (GTX 1050 Ti) > (GTX 1050 ≈ RX 460)

      Double tilde (or [i<]swung dash[/i<] to you hipsters) accomplished.

      • Freon
      • 3 years ago

      Besides Hitman, I’d be surprised if the 460 can keep up with the 1050.

        • AnotherReader
        • 3 years ago

        The 1050 has lower boost clocks than the 1060, 1070 and 1080. It may end up being closer to the 460 than we might expect.

    • Hattig
    • 3 years ago

    I wonder how AMD will respond in terms of 460 pricing.

    That presupposes that AMD has a plan and is aware of the 1050, which given it’s AMD, might not be the case. They know the product details, they know the suggested prices, and they should know the reality of the competitive landscape and what the cards will really be priced at.

    However, if they can get the 2GB 460 to $89 – $99 in actual retail, then that’s a strong position against the 1050.

    The 4GB 460 needs to drop to around $119 – $129 to be viable against the $139 1050 Ti. A 4GB 465 with 1024 shaders at 1300MHz+ would also be nice, but I think they’re holding back the xx5 series to bump alongside Vega in Q1.

    The 4GB 470 should have a price drop too, as a means of being the obvious upsell from the $139 1050 Ti. Why get that if you can drop an extra $30 for 60-100% more power? Again, this requires actual product selling at that price, rather than >$200 Nitros.

      • MathMan
      • 3 years ago

      > However, if they can get the 2GB 460 to $89 – $99 in actual retail, then that’s a strong position against the 1050.

      Using this kind of reasoning, they’d be in an even stronger position if they lower the price even further. 🙂

      I haven’t seen any benchmarks yet, but it is to be expected that even a 1050 will blow the 460 out of the water. With a die size that’s just 10% smaller, that’s not a great position to be in, no matter how you price it.

    • NTMBK
    • 3 years ago

    Interesting that the clock speed is so much lower than for other Pascal parts- and much closer to Polaris clocks. Looks like Samsung/GloFo can’t hit the same speeds as TSMC.

      • ImSpartacus
      • 3 years ago

      I think it’s partly that, but also partly the issue of being limited to 75W.

      I bet Nvidia smartly knew they wouldn’t be able to use high clocks anyway on 75W parts, so they went with glofo/sammy to reserve capacity at tsmc.

      Unfortunately, this probably means that factory overclocked parts with 6 pin connectors won’t have quite as much headroom to work with had it been on tsmc.

        • watzupken
        • 3 years ago

        I feel the power constraint is likely the main reason. Maintaining a high clockspeed likely will require quite a fair bit of power, which I feel is the reason why the GTX 1060 and the GTX 1070 have pretty close power requirements.

      • DPete27
      • 3 years ago

      We saw that with AMD on GloFo 14nm. Although we don’t know the fab for sure, as mentioned in the article these are 14nm parts as opposed to the other Pascal parts being 16nm. So it shouldn’t be a surprise.

    • TwoEars
    • 3 years ago

    I wonder if Nvidia is turning a profit on these or if they’re dumping.

    Edit: Did I say something? It’s a legitimate question you know. And we do want Scott to still have a job don’t we?

      • MathMan
      • 3 years ago

      Well, they say there aren’t stupid questions, only stupid answers.

      But you’re really pushing it.

      • chuckula
      • 3 years ago

      [quote<]It's a legitimate question you know.[/quote<] It's exactly as legitimate as asking the same question about AMD's profit/dumping policy on the Rx 460, which includes a similar sized GPU die from a foundry (GloFo) that is doing such a stellar job that [url=https://techreport.com/news/30600/amd-takes-a-335m-one-time-charge-for-more-sourcing-flexibility<]AMD just paid them another $335 million to [b<]not[/b<] make chips for AMD[/url<]. Now here's the real question you should be asking yourself: If your question had been posted by somebody else in the Rx 460 review, would you have been upthumbing or downthumbing the post?

      • K-L-Waster
      • 3 years ago

      Yeahhh, the company that is consistently profitable and that is constantly accused of gouging and over charging *must* be dumping. Right.

      • Klimax
      • 3 years ago

      Sorry, but did you really ask that question about Nvidia????????????????????????????

      • Freon
      • 3 years ago

      Nvidia’s parts are significantly more efficient and require a cheaper BOM because of it, so I don’t see why you would think that. AMD has to put up a card with 30-50% higher power and 30-50% more memory bandwidth to compete. Surprise, Nvidia can undercut them on price while maintaining a profit. They also seem to move enough volume to disperse the ongoing and upfront R&D costs to achieve that higher efficiency.

      The writing has been on the wall now for years. You can’t just keep writing off the fact they take last year’s cards, rebadge and drop the price while maintaining the same BOM.

        • NTMBK
        • 3 years ago

        The 460 has a smaller die than the 1050. What are you talking about.

          • Klimax
          • 3 years ago

          Likely exception of 9mm^2
          Note: Die size taken from [url<]https://www.techpowerup.com/gpudb/2849/radeon-rx-460[/url<] (Its missing in this article)

          • Freon
          • 3 years ago

          Die sizes have remained similar for similar performance, so I didn’t even bring it up, but since you did here are several more–turns out it’s a pretty small gap, particularly for the 1050/460 you mentioned.

          290/390 series: 438 mm^2
          970/980: 398

          1060: 200
          480: 232

          1050: 135
          460: 123

          My expectation here is the 1050 will beat the 460 outside Gaming Evolved titles based on what we see with the 1060/480 here, and other various specs. The drop from the 470/480 to the 460 is larger on paper than the drop from the 1060 to 1050. As another thread mentions, it could put AMD in a position where they need to produce a “465” to compete effectively.

    • DPete27
    • 3 years ago

    And now that Nvidia has released their full lineup, AMD is again in 2nd place across the board.

    Keep pimping FreeSync AMD, that alone got you my purchase.

      • Chrispy_
      • 3 years ago

      FreeSync is kinda ridiculously good given how expensive G-Sync is tho.

        • stefem
        • 3 years ago

        Many FreeSync monitors still have important shortcoming like not supporting Low Framerate Compensation (LFC) unless the maximum refresh of the panel is equal or more than 2.5x the minimum.

        And before someone label this as a thing of the past, the latest FreeSync monitor, which was announced in a news here at TR, does not support LFC.

    • derFunkenstein
    • 3 years ago

    RIP RX 460.

      • rudimentary_lathe
      • 3 years ago

      Would be nice if this brought the RX 470 down a notch or two. There’s hardly any difference in price between the RX 470 and RX 480 anyways, at least where I live.

        • derFunkenstein
        • 3 years ago

        Seems to be the case in the US. $220 for a 4GB RX 480 vs $200 for the cheapest 470. The 470 has never (to my knowledge) been available at its recommended $180.

          • JustAnEngineer
          • 3 years ago

          The Radeon RX470 was within $5 of that price last week. You should read those special sales announcements that Newegg e-mails to you.

            • derFunkenstein
            • 3 years ago

            $5 over recommended price is such a steal.

      • I.S.T.
      • 3 years ago

      Seriously. It’s about to have its’ *** whipped, much like the llamas of old.

        • derFunkenstein
        • 3 years ago

        </3 WinAmp.

    • chuckula
    • 3 years ago

    [quote<]AMD's $140 RX 460 4GB couldn't quite catch a GeForce GTX 950 in our review, and the GTX 1050 Ti appears to offer even more performance than that card does.[/quote<] And that's the problem right there. At $140, the Rx 460 was only marginally in the game against the old 950, but those prices aren't sustainable now.

      • NTMBK
      • 3 years ago

      Bring on the price drops 🙂

        • Klimax
        • 3 years ago

        Yeah, fairly common refrain with AMD….

Pin It on Pinterest

Share This