Geforce GTX 1050 / 1050 Ti Thread

[alcoholbob]

Hmm shoulda waited for this instead of a getting a GTX 1060 Mini. Ended up being far more performance than I needed for a simple HTPC.

 

[Sweepr]

September 4th Update: Specs Out

https://benchlife.info/nvidia-geforce-gtx-1050-4gb-gp107-400-09042016

60% the number of CUDA cores of GP106 (GTX 1060 6GB), compared to 38.8% for Polaris 11 (RX 460) vs Polaris 10 (RX 480). Clocks are a little conservative for Pascal, probably a lot of headroom left for custom cards / future refreshes.

 

[SPBHM]

expected specs apart from the low clocks, maybe it's because of the no power plug requirement, but hopefully it overclocks a lot (considering where the other Pascals are)

can be great if it overclocks well and well price, should be easily a lot faster than the 460.

 

[Hi-Fi Man]

I expect the 4GiB version to be more future proof than the 1060 3GiB. 3GiB is just barely enough right now for a card in that performance class.

If priced right, this could be a really good card.

 

[Shivansps]

If they price it at $150 its gona be nice, and i could imagine a cut down GP107-200 with 2GB at $100 as well. It whould be VERY competive to the RX460s. And im petty sure some crazy OEM will want to put 8GB on those as well, not sure if possible with 128bit.

It remains to be seen what GP108 will bring in the lowest end on the sprectrum after this, Nvidia needs to bring NVENC and HEVC/VP9 hardware decoders to sub-$100, im tired of seeing Fermi and Kepler rebrands. And AMD is even worse with those 5450 and 6450 rebrands.

 

[DamZe]

The price will have to be right, if they can get that magical 750Ti 150$ MSRP, it will fly off the shelves.

 

[tviceman]

September 4th Update: Specs Out

https://benchlife.info/nvidia-geforce-gtx-1050-4gb-gp107-400-09042016

60% the number of CUDA cores of GP106 (GTX 1060 6GB), compared to 38.8% for Polaris 11 (RX 460) vs Polaris 10 (RX 480). Clocks are a little conservative for Pascal, probably a lot of headroom left for custom cards / future refreshes.

Nvidia is going with a different GPC configuration with GP107 like they did with GM107. I had predicted 640 cuda cores (half GP106). For its TDP, it should be a beast. It will probably best the GTX 960 by 15%.

 

[Arachnotronic]

Perfect GPU for people trying to upgrade OEM systems without PCIe connector. And people on budgets who still want to game.

Looks good.

 

[Rezist]

Kinda disappointed about the clocks, with the architecture these should be capable of like 1.8GHz at decent power consumption should they not, possible justify a 6-pin version.

 

[whm1974]

Kinda disappointed about the clocks, with the architecture these should be capable of like 1.8GHz at decent power consumption should they not, possible justify a 6-pin version.

I would prefer a 1050 without a power pin so it can replace the 750 Ti I've been recommending to folks who buy "Walmart" specials.

 

[boozzer]

In all truth, you weren't far off. It should be the 1050. XD

a very subtle jab indeed

 

[Samwell]

Kinda disappointed about the clocks, with the architecture these should be capable of like 1.8GHz at decent power consumption should they not, possible justify a 6-pin version.

Seems Samsungs process is really way worse in high clocks compared to TSMC in the other pascals. Might actually be a big help for AMD with Vega if they change to TSMC:

 

[jpiniero]

Not going to be faster than the 960, although aftermarket cards with a plug should let it clock high enough to pass it. At least it is 768 cores.

 

[dark zero]

This card seems to be great and is the rival of the 470. Now AMD will have some problems on there.

Only 1 card is missing and is the GT 1040 for budget markets, being the true rival of the 460 and needs to be pinless and fanless in order to bring an unique card to the market. With that AMD would be totally defeated.

 

[antihelten]

Nvidia is going with a different GPC configuration with GP107 like they did with GM107. I had predicted 640 cuda cores (half GP106). For its TDP, it should be a beast. It will probably best the GTX 960 by 15%.

This thing definitely isn't going to beat the GTX 960 with that boost clock. In fact it will probably be about 15% slower (and thus compete directly with the RX 460 4GB).

This card seems to be great and is the rival of the 470. Now AMD will have some problems on there.

Only 1 card is missing and is the GT 1040 for budget markets, being the true rival of the 460 and needs to be pinless and fanless in order to bring an unique card to the market. With that AMD would be totally defeated.

With those specs it isn't going to come anywhere near an RX 470. It will be roughly 40% slower.

So by extension it obviously goes without saying that the GT 1040 isn't going to be the true rival of the 460, since that job is clearly taken by the GTX 1050. The GT 1040 might end up competing with some cutdown version of Polaris 11 (i.e. Radeon 450)

 

[Sweepr]

This thing definitely isn't going to beat the GTX 960 with that boost clock. In fact it will probably be about 15% slower (and thus compete directly with the RX 460 4GB).

Pascal can easily go above the rated Turbo, the same should apply to GP107 (unless power limited). Specs wouldn't suggest a GTX 1060 matches a GTX 980, but it does. GTX 1050 vs GTX 960 should be similar, and that already puts it 20% above a RX 460:

https://tpucdn.com/reviews/ASUS/RX_460_STRIX_OC/images/perfrel_1920_1080.png

Anyway, the real show here will come from custom models.

 

[whm1974]

If I was Nvidia, I would be releasing the 1050 now at a reasonable price point.

 

[Shivansps]

This thing definitely isn't going to beat the GTX 960 with that boost clock. In fact it will probably be about 15% slower (and thus compete directly with the RX 460 4GB).



With those specs it isn't going to come anywhere near an RX 470. It will be roughly 40% slower.

So by extension it obviously goes without saying that the GT 1040 isn't going to be the true rival of the 460, since that job is clearly taken by the GTX 1050. The GT 1040 might end up competing with some cutdown version of Polaris 11 (i.e. Radeon 450)

It remains to be seeing how fast GP108 will be, but anyway the main point of GP108 will be bringing multimedia capabilities and NVENC to sub $100.

If i had to bet, it whould be

GTX1050 4GB = GTX960 or more

GT1040 GP108 = GTX750TI.

 

[antihelten]

Pascal can easily go above the rated Turbo, the same should apply to GP107 (unless power limited). Specs wouldn't suggest a GTX 1060 matches a GTX 980, but it does. GTX 1050 vs GTX 960 should be similar, and that already puts it 20% above a RX 460:

https://tpucdn.com/reviews/ASUS/RX_460_STRIX_OC/images/perfrel_1920_1080.png

Anyway, the real show here will come from custom models.

I know that Pascal can go above the rated Boost, but I already accounted for that in my estimates. So unless the GTX 1050 goes above its boost clock by a higher relative margin than the GTX 1060 does, the estimate should be accurate.

Looking at specs the 1050 has 50% less shading performance than the 1060 (768 cores at 1380 MHz vs. 1280 cores at 1708 MHz is 51.5% less). If both cards go above their rated boost clock by a similar amount, then the 1050 obviously still only has 50% the shading performance of the 1060.

If 50% less shading performance leads to 50% less game performance (which would assume that shading performance is the main bottleneck), then the 1050 will end up at roughly 10-15% slower than the GTX 960, and roughly equal to the RX 460 4GB.

 

[Sweepr]

If 50% less shading performance leads to 50% less game performance (which would assume that shading performance is the main bottleneck), then the 1050 will end up at roughly 10-15% slower than the GTX 960, and roughly equal to the RX 460 4GB.

Doesn't work like this. 1060 6GB has 2/3 the shader performance of 1070, but 72.5% the gaming performance according to TPU. Partners will probably have factory OCed cards at launch, especially those with 6-pin power connectors - and we know Pascal can clock high.

Ps: Reference GTX 960 is 21.2% faster than RX 460 4GB Nitro in your link, same as what I posted.

 

[antihelten]

Doesn't work like this. 1060 6GB has 2/3 the shader performance of GTX 1070, but 72.5% the gaming performance according to TPU. Partners will probably have factory OCed cards at launch, especially those with 6-pin power connectors - and we know Pascal can clock high.

Ps: Reference GTX 960 is 21.2% faster than RX 460 4GB Nitro in your link, same as what I posted.

Actually it works exactly like this, I guess you missed the part where I wrote "which would assume that shading performance is the main bottleneck". This clearly isn't the case with the 1060 vs the 1070, which can most likely be chalked up to the 1070 being bandwidth limited (the 1060 only has 68% the shading performance of the 1070, but it has 75% of the bandwidth).

The 1050 has 51.5% less shading performance and 41.7% less bandwidth. If the bandwidth is the main bottleneck instead of shading performance then the 1050 could potentially be neck and neck with the GTX 960.

At this point it's anyone's guess whether shading performance or bandwidth is the main bottleneck. But either way I think it's safe to say that the 1050 will not beat the 960 by 15% with those specs, nor will it get anywhere near the RX 470.

 

[Sweepr]

Actually it works exactly like this, I guess you missed the part where I wrote "which would assume that shading performance is the main bottleneck".

Actually I didn't, I just don't agree that it it will be 15% slower than a 960 based on this, when it isn't the case with 1060 6GB vs 1070. And as you already pointed in the rest of your post, 1050 has almost 60% the bandwidth of a 1060.

But either way I think it's safe to say that the 1050 will not beat the 960 by 15% with those specs

Agreed.

 

[antihelten]

Actually I didn't, I just don't agree that it it will be 15% slower than a 960 based on this, when it isn't the case with 1060 6GB vs 1070.

The shading performance is not the main bottleneck with regards to the 1060 vs the 1070, so that example holds no relevance here (since I specifically said that my example was based on shading performance being the main bottleneck). And if you don't agree that gaming performance follows shading performance when shading performance is the main bottleneck, then I would love to hear how in the world that would work.

 

[Sweepr]

The shading performance is not the main bottleneck with regards to the 1060 vs the 1070

Doesn't matter, this is just an example where shading performance doesn't translate to overall gaming performance. And you don't have data to prove that shading (or bandwidth) will be the bottleneck regarding 1050 vs 1060, as you carefully stated here:

At this point it's anyone's guess whether shading performance or bandwidth is the main bottleneck.

You're just assuming the former is the case, and doing some pessimistic performance predictions based on it. I think it will be closer to a 960 than you expect.

 

[VirtualLarry]

Nvidia needs to bring NVENC and HEVC/VP9 hardware decoders to sub-$100

This!!!

PS. Thanks for this thread, Sweepr.

Edit: For the 1050 to truly "replace" the GTX950, it would have to support VGA monitors. Does it?