Geforce GTX 1050 / 1050 Ti Launch Thread ($109 / $139 - October 25th)

[mohit9206]

So today we will get the reviews of both the cards right? Also this 1050Ti vs RX470 will be 750ti vs 270/270X all over again.

 

[Sweepr]

NDA is up at 05:00h PT (2 hours and 10 minutes to go).

So today we will get the reviews of both the cards right?

It is. Any last minute predictions?

 

[mohit9206]

NDA is up at 05:00h PT (2 hours and 10 minutes to go).




It is. Any last minute predictions?

Prediction is 1050 2gb faster than rx460 2gb and 4gb in dx11 but same or slower in dx12.
1050ti faster than rx460 in both dx11 and dx12 and with overclocking almost match rx470 in dx11.

 

[hojnikb]

I highly doubt 1050ti can match 470. It is made on glofo 14nm, so oc headroom could be slim.

 

[mohit9206]

I highly doubt 1050ti can match 470. It is made on glofo 14nm, so oc headroom could be slim.

Am not an expert on these things but care to explain why 14nm glofo/samsung makes it have less oc headroom than 16nm tsmc?

 

[Glo.]

The reviews are up. Good performance, temperatures, and efficiency, however... the pricing is again very problematic.

 

[R0H1T]

The reviews are up. Good performance, temperatures, and efficiency, however... the pricing is again very problematic.

The OC headroom is almost certainly confirmed to be lower on 14nm GF/Samsung, so is the efficiency btw a bit surprised to see stock RX 460 this close to the 1050 & 1050Ti.

 

[railven]

Really interesting that NV can hit 1900+ on Samsung. Makes you wonder if the copy+paste job over at GloFlo is more botched then originally speculated.

 

[f2bnp]

Was just checking this and Guru3D's review. 1050 makes no sense IMO, but if you can get the 1050Ti for the same or 10-20$ more than the RX 460, it is an attractive option.
RX 470 blows them out of the water though.

I honestly did expect the 1050/1050 Ti to fare a little better.

 

[R0H1T]

Really interesting that NV can hit 1900+ on Samsung. Makes you wonder if the copy+paste job over at GloFlo is more botched then originally speculated.

Where did you see 1050/Ti hitting 1900Mhz (boost?) clocks

 

[railven]

Where did you see 1050/Ti hitting 1900Mhz (boost?) clocks

Well TechPowerUP OC'ed theirs up to 1911mz and claim it would go higher:

GPU overclocking is capped too, at 1911 MHz actual frequency after GPU Boost, but the implementation seems buggy. Even though it shows 1911 MHz, the actual clock can still be increased, because beyond 1911 MHz displayed the card will run at higher performance and at even higher clocks get increasingly unstable. Maximum overclock of our sample was +201 MHz GPU clock, a 15% increase

 

[Arachnotronic]

Really interesting that NV can hit 1900+ on Samsung. Makes you wonder if the copy+paste job over at GloFlo is more botched then originally speculated.

It's not the process, it's the circuit implementation/design. Zen is hitting >3GHz on that same process at GloFo.

 

[Arachnotronic]

Was just checking this and Guru3D's review. 1050 makes no sense IMO, but if you can get the 1050Ti for the same or 10-20$ more than the RX 460, it is an attractive option.
RX 470 blows them out of the water though.

I honestly did expect the 1050/1050 Ti to fare a little better.

GTX 1050 Ti -> $139

RX 470 -> $170

Different price points.

 

[R0H1T]

Well TechPowerUP OC'ed theirs up to 1911mz and claim it would go higher:

Strange considering GPUz doesn't show the said clocks in their review. Unless I'm missing something, that or GPUz is off by some ~250MHz.

 

[nvgpu]

http://www.eurogamer.net/articles/digitalfoundry-2016-nvidia-geforce-gtx-1050-ti-review

The Zotac version of the GTX 1050 Ti is very small and pretty basic but it requires no PCI Express power input. It still manages to boost up to 1650MHz straight out of the box.

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1050-ti,4787.html

This time around, we're using the values from our highest sustainable overclock to represent peak power consumption (we went as high as 1911 MHz). Nvidia's 75 W power target didn't seem to be a limiting factor; we increased the target by 25% and didn't see consumption rise at all. In other words, the 1050 Ti hit its limit at that frequency.

Pascal just flies on any process tech, even inferior Samsung 14nm.

 

[Sweepr]

Overall much better than I was expecting.

- Power consumption is ridiculously low, only 67W typical gaming:

- Performance per watt 50% above Polaris even with Samsung 14nm - beats TSCM based GTX 1060 as well:

- From a technical point of view the 1050 Ti completely mops the floor with RX 460 at equal/lower power levels. 80% RX 470 performance with a much smaller die and 67W vs 121W is also impressive:

 

[Arachnotronic]

Pascal just flies on any process tech, even inferior Samsung 14nm.

This. People downplayed the serious work that NVIDIA put into allowing Pascal to clock high at good power consumption, with some people calling it "Paxwell" to try to disparage it.

The reality is that performance is some combination of clock speed and work done per clock, and Pascal delivered a very nice boost in performance mainly through higher clocks.

 

[Sweepr]

Edit: Nevermind.

 

[USER8000]

So looking at those figures,the GTX960 2GB used was most likely the Nvidia reference one running at stock clockspeeds. That means the GTX960 4GB model which was usually sold as AIB models and was pre-overclocked would still be close to the performance as the AIB GTX1050TI 4GB Techpowerup used. No wonder they said to buy a RX470 instead:

MSI's pricing for the GTX 1050 Ti Gaming X is set to be around $165, which is $25 more than the "starting at $140" price provided by NVIDIA. In my opinion this increase is way too big and brings the card in a territory where RX 470 is the better option due to higher performance at the same price point. GTX 1050 Ti has definitely better power/heat/noise over RX 470, but I'd consider it as an option because performance trumps everything in this segment where every dollar counts. Radeon R9 380 is completely obsolete now, with the spotlight being on RX 470 and GTX 1050 Ti. Even at reference pricing of $140, I feel slightly tempted to recommend RX 470 for its higher performance; with the price of AAA games these days, the difference is basically to skip one game and spend that money on a better graphics card which will bring you more joy in all the other games you play. Another option could be GTX 1060 3 GB, which is slightly more expensive than RX 470 (after its price drops), but offers a bit better performance, especially in older titles.

 

[nvgpu]

The Zotac board we were provided with happily boosts to 1650MHz out of the box without breaking a sweat, and overclocking taps out at 1850MHz, without exceeding that hard 75W limit.

As expected, the external 6-pin does nothing useful, buying the no power connector card is the way to go.

GP107 absolutely destroys the competition in power efficiency for HTPC usage.

https://tpucdn.com/reviews/MSI/GTX_1050_Ti_Gaming_X/images/power_bluray.png

https://tpucdn.com/reviews/MSI/GTX_1050_Gaming_X/images/power_bluray.png

https://tpucdn.com/reviews/ASUS/RX_460_STRIX_OC/images/power_bluray.png

 

[Arachnotronic]

As expected, the external 6-pin does nothing useful, buying the no power connector card is the way to go.

Great for turning crap pre-built OEM systems into reasonable gaming systems. That's the main use for this class of card.

 

[R0H1T]

So looking at those figures,the GTX960 2GB used was most likely the Nvidia reference one running at stock clockspeeds. That means the GTX960 4GB model which was usually sold as AIB models and was pre-overclocked would still be close to the performance as the AIB GTX1050TI 4GB Techpowerup used. No wonder they said to buy a RX470 instead:

That's a rarity from Wizz & TPU, certainly as the consensus on this forum would suggest.

 

[dark zero]

Still... No fanless card from NVIDIA...
Seems that temps could be higher that not to use it.

Also that confirms that without going FP64 is the way to go for nVIDIA. Double presicion isn't that good than expecting.

 

[USER8000]

That's a rarity from Wizz & TPU, certainly as the consensus on this forum would suggest.

The thing is people need to remember he only has a reference GTX960 2GB to hand(probably),my GTX960 4GB is pre-overclocked out of the box,so add another 10% to the performance figures for the GTX960 2GB he used due to the combination of the higher clockspeeds and increased VRAM. The GTX1050TI he tested is pre-overclocked out of the box and has 4GB of VRAM.

So this looks like a slightly faster slot powered GTX960 4GB. Now that is not a bad thing for certain pre-built systems,but Overclockers UK had the Powercolor RX470 Devil 4GB for £165 and that is apparently close to a reference RX480 4GB/8GB.

I do think AMD and Nvidia are quietly trying to upsell the RX470 and GTX1060 3GB in some ways.

 

[railven]

It's not the process, it's the circuit implementation/design. Zen is hitting >3GHz on that same process at GloFo.

I understand CPU and GPU have different ICs.

I'd wager money NV wouldn't hit these clocks if they went with GloFlo. Just a hunch.

Strange considering GPUz doesn't show the said clocks in their review. Unless I'm missing something, that or GPUz is off by some ~250MHz.

Tom's is reporting similar. 1911mhz through overclocking.