Nvidia GeForce GTX 1050 Specifications Leaked

Megalith

24-bit/48kHz
Staff member
Joined
Aug 20, 2006
Messages
13,000
Here is what to expect for the next release in the Pascal lineup. Apart from clocks and memory size, the card is comparable to the GTX 950. We can also assume that the card may be purely powered through PCI-e.

According to the latest report, upcoming mid-range solution based on Pascal GP107 silicon will feature 768 CUDA cores. That’s 512 CUDA cores less than full fat GP106. It means that GP104 will host 12 Streaming Multiprocessors Pascal. Benchlife posted a GPU-Z screenshot with driver name, BIOS version, Device ID and Subvendor information erased, but showing some very important information we didn’t know yet. It’s worth adding that we don’t know if this is a reference card, or if there’s even going to be ‘founders edition’ of this model, but it’s a good starting point for our speculation.
 
And the closeness to the release of the RX 460 is a complete and utter coincidence... Here's hoping that they release a fanless version.
 
Be nice if they can land these close to the $100. The 750/750Ti held on for a very long time because the 950 was just a bit too far over $100 to make it an easy drop in dedicated for budget PCs.
 
I'm still rockin' a GTX560Ti. Will this be the first nVidia x50 series card to beat it since?
 
If this is faster than a 7850 and uses less power, its going in my HTPC.
 
Probably be about as fast as a RX470 for [EDIT Due to typo]$225[/edit]. Each has [edit]4 gigs[/edit] memory. So, that would be a tough call because between the two. The 1050 would overclock really well on low power. But it would be significantly slower on DX12 / vulkan apps compared to the RX470

Sorry I was tired when I typed that reply
 
Last edited by a moderator:
Hoping for a low profile version to mess with! Heck even a low pro Rx 460 would be decent.
 
Probably be about as fast as a RX470 for $125. Each has 4 megs memory. So, that would be a tough call because between the two. The 1050 would overclock really well on low power. But it would be significantly slower on DX12 / vulkan apps compared to the RX470
4 megs? :eek: :p
 
That could be a great budget gaming card if it sells for $150 or less.
 
I'm still rockin' a GTX560Ti. Will this be the first nVidia x50 series card to beat it since?

isn't that 750ti already ahead of 560Ti? i remember looking bench 750ti matching the performance of 480. and 480 is comparable to 570.
 
Probably be about as fast as a RX470 for $125. Each has 4 megs memory. So, that would be a tough call because between the two. The 1050 would overclock really well on low power. But it would be significantly slower on DX12 / vulkan apps compared to the RX470

i don't think 1050 will touch RX470 performance. maybe more like 960/380 at 75w
 
I want a mobile version of this. The 1060 is still too much heat/power for the mobile space, and I want to get a laptop before too long.
 
I want a mobile version of this. The 1060 is still too much heat/power for the mobile space, and I want to get a laptop before too long.
You do know they came out with mobile versions of these chips right? The MGTX1060 comsumes about 80 Watts.
 
I guess AMD is going to have to release the full Polaris 10 chip with all 16 CU's enabled, RX465 inbound? maybe that was the plan all along...
 
Let's see, 768 shaders @ 1380 for the 1050 vs 1280 @ 1506 for the 1060 gives us roughly 55% of the power of the 1060 on paper.

Let's be generous and say that it is actually 60% since adding more cores doesn't scale linealy with effective performance. This would be above 960, and slightly above a 380 for 1080p "In Theory".


Time will tell if it actually performs in that bracket, and at what price.
 
I guess AMD is going to have to release the full Polaris 10 chip with all 16 CU's enabled, RX465 inbound? maybe that was the plan all along...

Doesn't seem to matter what AMD does, what move they try to make - Jen-Hsun will just slap it down with the 2x4 in his left hand, garden burger in his right.
 
Last edited:
Let's see, 768 shaders @ 1380 for the 1050 vs 1280 @ 1506 for the 1060 gives us roughly 55% of the power of the 1060 on paper.

Let's be generous and say that it is actually 60% since adding more cores doesn't scale linealy with effective performance. This would be above 960, and slightly above a 380 for 1080p "In Theory".


Time will tell if it actually performs in that bracket, and at what price.

Seriously, there's nothing to suggest this won't clock just as high as everything else. 2GHz EZ
 
I guess AMD is going to have to release the full Polaris 10 chip with all 16 CU's enabled, RX465 inbound? maybe that was the plan all along...

No such chip exist. Also you mean Polaris 11.

Both Polaris 10 and 11 have been released in all they got. No secret unused CUs or anything else.
 
Turns out this is manufactured at Samsung 14nm lpp so clocks could be very different from the other Pascal parts.

Will be nice to compare this rx480 and figure out what happened there
 
Turns out this is manufactured at Samsung 14nm lpp so clocks could be very different from the other Pascal parts.
14nm LPE was a less mature, and more expensive, process intended for "first to market" 14nm designs. The E in that acronym stands for Early. I'm not sure why you think GP107 on 14nm LPP is a bad thing (the second P in that acronym stands for Plus). Basically all new (for interesting PC parts, at least) designs going forward will be on LPP if it's on Samsung's 14nm process (including GLF).

The chosen clock speed on the GP107 is almost certainly not a limitation of the process, although it may have been designed to work at lower clocks than the higher end parts*. As usual, clock speed is one of the ways of segmenting price strata.

* there are trade-offs that can make this choice even more power and performance efficient per W or clock vs a simple clock reduction of the GP106/GP104/GP102 designs.
 
Last edited:
Yeah, can't wait to see how this performs without a power connector. Should be GTX 960 performance or higher! This is the true replacement for the GTX 750 Ti :D

The RX 460 CAN be bus powered, but takes a massive performance hit without the 6-pin (ext power card is 20% faster in most games):

AMD Radeon RX 460 im Test (Seite 3)

Which is why the vast majority of sites were shipped 6-pin models for review, INCLUDING HardOCP. Nvidia didn't pull that shit with the 750 Ti.

And while some tests are obviously VRAM-limited on the 2GB card, you're still going to have to deal with that because most of the 4GB cards have a 6-pin.

The RX 480 is a good card, but the high performance of the RX 460 comes at a high cost. The performance/watt of Polaris 11 is no better than the GTX 960, making it a poor choice for laptops (in light of the coming GP107).
 
Last edited:
That would be a 50% overclock. I wouldn't expect that "EZ".

The gtx 1060 3gb is 1500mhz boost on paper, clocks comfortable to 2ghz like all the other Pascal cards. 33% isn't EZ but it's doable all the time.
14nm LPE was a less mature, and more expensive, process intended for "first to market" 14nm designs. The E in that acronym stands for Early. I'm not sure why you think GP107 on 14nm LPP is a bad thing (the second P in that acronym stands for Plus). Basically all new (for interesting PC parts, at least) designs going forward will be on LPP if it's on Samsung's 14nm process (including GLF).

The chosen clock speed on the GP107 is almost certainly not a limitation of the process, although it may have been designed to work at lower clocks than the higher end parts*. As usual, clock speed is one of the ways of segmenting price strata.

* there are trade-offs that can make this choice even more power and performance efficient per W or clock vs a simple clock reduction of the GP106/GP104/GP102 designs.

Polaris uses LPP, I know LpE is older. Anyway clock reduction (if any) will have a lot to do with process I highly doubt they change the design so it clocks lower and they make it only 6 SMs at the same time

I've literally said nothing to suggest I think LPP is worse than LPE...
 
I know, but I still can't help be curious.

I think LinusTechTips did a video recently where they tested a bunch of different combos of high end and low end cards mixed with SLI results to compare the PhysX advantage. Might be worth a watch.
 
Back
Top