Geforce GTX 1060 Thread: faster than RX 480, 120W, $249

[3DVagabond]

Because for the umpteenth time, there is no generally agreed standard upon which one can easily categorize good reviews versus bad reviews.

The may be a few cases here and there, where we can easily conclude that a subpar method was used, like for instance when Computerbase.de uses DX12 in Tomb Raider for AMD and DX11 for Nvidia, even though DX11 is the best API for both (at least until they start testing with the new patch). TPU using an open case test setup is also problematic, albeit arguably to a lesser degree (since there may actually be consumers out there who also uses open cases).

Problem is that every review out there probably suffers from issues like the above one to a lesser or greater degree, and unless someone is willing to trawl through every single one of them and identify every single potential issue and then develop a weighting system based on this, the safest thing is to simply go with something like 3DCenters analysis.

Which is why I'd argue that you can't simply lump them all together. You actually have to take the time and make the effort to look at them individually.

The way TPU does it has it's value. Is it definitive? No. The way [H] does it has it's value. I don't believe that you can average their results though and have a better solution. Each tells you something different about the hardware.

And if you think you get frustrated having to repeat yourself, imagine how hard it is to read you repost the same thing over and over like if you say it enough it'll make it right.

 

[zinfamous]

I've noticed some sites test Witcher 3 and they find the 1060 beat the 480 by a big margin, I wonder what's going on when side by side comparison with video show them really close:

https://www.youtube.com/watch?v=uzyUVQHzDwk

https://www.youtube.com/watch?v=lBlCUV6TwiY

^ These aren't nobody reviewers either, the 2nd is Digital Foundry, testing in the toughest area in the city. And the above is from the expansion, again in the city.

Do these reviews which show a big gap run with HairWorks & HBAO+ on or something?

Just some observations and I don't really know what they mean:
AMD is using more system memory and nVidia uses a bit more vRAM. nVidia seems to be more efficient at using all CPU cores

FPS is pretty much equal with only 1200Mhz for AMD and 1800Mhz for nVidia...how is AMD vastly more efficient in pushing performance per clock? Is that the right way to say that? With those unvetted leaks for the Nitro showing insane performance gains (20-30%) at only 100-200Mhz OC on the 480, is this what we are seeing? Other reviewers have said that 1060 OCs well while some of said it doesn't, and that gains are minimum. Still other reviewers have claimed that AMD OC gains are minimum.

I don't know enough (anything, really), about processor architecture to extract meaningful information from this clear difference in performance per clock--is a lot of that AMD's focus on the low-level processing and CPU utilization (but in that first link, 1060 seems to be better at loading the CPU cores)?

 

[zinfamous]

Because for the umpteenth time, there is no generally agreed standard upon which one can easily categorize good reviews versus bad reviews.

The may be a few cases here and there, where we can easily conclude that a subpar method was used, like for instance when Computerbase.de uses DX12 in Tomb Raider for AMD and DX11 for Nvidia, even though DX11 is the best API for both (at least until they start testing with the new patch). TPU using an open case test setup is also problematic, albeit arguably to a lesser degree (since there may actually be consumers out there who also uses open cases).

Problem is that every review out there probably suffers from issues like the above one to a lesser or greater degree, and unless someone is willing to trawl through every single one of them and identify every single potential issue and then develop a weighting system based on this, the safest thing is to simply go with something like 3DCenters analysis.

Then metanalysis is no better than simply picking the better turd in a pile of turds, if they are all turds. That's really the only rational way to approach things here.

The goal is to find or develop a better testbed, not average out the results from a pile of bad data.

 

[selni]

Just some observations and I don't really know what they mean:
AMD is using more system memory and nVidia uses a bit more vRAM. nVidia seems to be more efficient at using all CPU cores

FPS is pretty much equal with only 1200Mhz for AMD and 1800Mhz for nVidia...how is AMD vastly more efficient in pushing performance per clock? Is that the right way to say that? With those unvetted leaks for the Nitro showing insane performance gains (20-30%) at only 100-200Mhz OC on the 480, is this what we are seeing? Other reviewers have said that 1060 OCs well while some of said it doesn't, and that gains are minimum. Still other reviewers have claimed that AMD OC gains are minimum.

I don't know enough (anything, really), about processor architecture to extract meaningful information from this clear difference in performance per clock--is a lot of that AMD's focus on the low-level processing and CPU utilization (but in that first link, 1060 seems to be better at loading the CPU cores)?

It's because the GPUs are different "sizes", not a single unit - GPUs are among other things a very large array of vector arithmetic units. The size of this array varies by GPU and the number of cores is of course very important for overall performance (it's really clockspeed x cores that matters, not just raw clockspeed).

Recently AMD has tended to favor wider designs while NV has preferred higher clocks and the 480 is no exception - it has a considerably larger number of cores than the 1060 which is why the clockspeed doesn't tell the whole story.

 

[zinfamous]

It's because the GPUs are different "sizes", not a single unit - GPUs are among other things a very large array of vector arithmetic units. The size of this array varies by GPU and the number of cores is of course very important for overall performance (it's really clockspeed x cores that matters, not just raw clockspeed).

Recently AMD has tended to favor wider designs while NV has preferred higher clocks and the 480 is no exception - it has a considerably larger number of cores than the 1060 which is why the clockspeed doesn't tell the whole story.

Ah I see, thanks. The ROPs and Shader units and whatever?

 

[selni]

Ah I see, thanks. The ROPs and Shader units and whatever?

Shaders/various other names (CUDA cores for nv, stream processors for amd). ROPs are something different entirely.

 

[antihelten]

Which is why I'd argue that you can't simply lump them all together. You actually have to take the time and make the effort to look at them individually.

The way TPU does it has it's value. Is it definitive? No. The way [H] does it has it's value. I don't believe that you can average their results though and have a better solution. Each tells you something different about the hardware.

And if you think you get frustrated having to repeat yourself, imagine how hard it is to read you repost the same thing over and over like if you say it enough it'll make it right.

Yes looking at them and evaluating them individually is arguably better, but the point is that noone is going to do that, and as such while a meta review like the one 3DCenter does is not optimal it is still the best we got.

The thing people apparently don't understand is that I never claimed that doing a meta review was the perfect approach, merely that it is the best approach that anyone has actually bothered carrying out.

Then metanalysis is no better than simply picking the better turd in a pile of turds, if they are all turds. That's really the only rational way to approach things here.

The goal is to find or develop a better testbed, not average out the results from a pile of bad data.

The underlying assumption is that the individual reviews are all flawed in different ways, i.e. one review may be biased towards Nvidia and another may be biased towards AMD. Doing a meta review then ensures that these errors evens out.

And yes it would be awesome if someone out there developed better testbeds / testing methodologies, but I don't see anyone doing it.

 

[railven]

And yes it would be awesome if someone out there developed better testbeds / testing methodologies, but I don't see anyone doing it.

Easier to complain on forums and accuse everyone of being in someone's pocket. Until said site/reviewer makes a glowing endorsement and suddenly they become the only legitimate source of information.

 

[Sweepr]

MSI Gaming X 1060 vs Sapphire Nitro+ RX 480

4GB model

8GB model

Overclock Comparison

4.8% (RX 480) vs 8.0% (GTX 1060). 4GB model still draws an extra 85W under load, 100W for the 8GB model.

 

[coercitiv]

4.8% (RX 480) vs 8.0% (GTX 1060). 4GB model still draws an extra 85W under load, 100W for the 8GB model.

Really curious to see how mobile chips end up, since within 35-60W limits any significant power advantage turns into pure performance advantage, and by the looks of it Nvidia chips will maintain at least some advantage even there.

 

[nerp]

Reviews can be thrown off in millions of ways. I don't doubt some review somewhere was tainted by a background process chugging away. Who knows. The system could suddenly start reindexing a huge network volume with 50mb tiffs and the "runtime broker" could be going ape during the benchmarking. Stuff happens.

In six months, the difference in performance between these cards will be much more clear. Until the next driver/game update.

 

[crisium]

Really curious to see how mobile chips end up, since within 35-60W limits any significant power advantage turns into pure performance advantage, and by the looks of it Nvidia chips will maintain at least some advantage even there.

I'm interested in the best they can squeeze out of PCI-e only. GTX 950 which has a pinless version at full speed, for example. GTX 1060 Green Edition with lower clocks (if such a thing existed, basically mobile 1060 on desktop) or perhaps a full speed GTX 1050 may be able to do so as well.

 

[watek]

I moved away from big form factor gaming to SFF and this EVGA 1060 SC mini is the best card i've touched so far in awhile! Really impressed with the performance, so power efficient, low noise quiet and its the size of a persons hand lol. And overclocks to 2ghz game stable on this little beast! Nvidia did right with this GTX 1060 release big thumbs up.

 

[Sweepr]

NotebookCheck added some preliminar 3DMark 11 results for mobile GTX 1060 (13690 pts). If the leaked specs are correct, they're offering ~90% the performance of the desktop variant (15290 pts) at nearly half the TDP - 65W vs 120W.

 

[antihelten]

NotebookCheck added some preliminar 3DMark 11 results for mobile GTX 1060 (13690 pts). If the leaked specs are correct, they're offering ~90% the performance of the desktop variant (15290 pts) at nearly half the TDP - 65W vs 120W.

To put this in context of what is currently available at comparable TDP:

960M - 65W - 5293.6 (1060M is 159% faster)
965M - 60W - 7619.5 (1060M is 80% faster)
970M - 75W - 9950.2 (1060M is 38% faster)

And the fastest mobile GPU Nvidia currently offers:

980M - 100W - 12459.2 (1060M is 10% faster)

 

[Keysplayr]

Easier to complain on forums and accuse everyone of being in someone's pocket. Until said site/reviewer makes a glowing endorsement and suddenly they become the only legitimate source of information.

Yeah, that is rampant.

 

[sontin]

NotebookCheck added some preliminar 3DMark 11 results for mobile GTX 1060 (13690 pts). If the leaked specs are correct, they're offering ~90% the performance of the desktop variant (15290 pts) at nearly half the TDP - 65W vs 120W.

Purepc.pl has released a Firestrike score two weeks ago: http://www.purepc.pl/notebooki/geforce_gtx_1060_mobile_pelny_uklad_gp106_trafi_do_laptopow

GP106 will be a force in the notebook market. Offering nearly twice the performance over the competition in the same power range changes the whole market.

The first real "VR Premium" notebook GPU.

 

[godihatework]

Purepc.pl has released a Firestrike score two weeks ago: http://www.purepc.pl/notebooki/geforce_gtx_1060_mobile_pelny_uklad_gp106_trafi_do_laptopow

GP106 will be a force in the notebook market. Offering nearly twice the performance over the competition in the same power range changes the whole market.

The first real "VR Premium" notebook GPU.

this is what i'm waiting for.

i'm too old for the desktop + dedicated gaming area, and everything that entails.

But a Full HD laptop with a mobile 1060 that can run any game i could throw at it with all the bells and whistles that weighs 3 pounds and is under an inch thick?

sign. me. up.

 

[hsjj3]

To put this in context of what is currently available at comparable TDP:

960M - 65W - 5293.6 (1060M is 159% faster)
965M - 60W - 7619.5 (1060M is 80% faster)
970M - 75W - 9950.2 (1060M is 38% faster)

And the fastest mobile GPU Nvidia currently offers:

980M - 100W - 12459.2 (1060M is 10% faster)

Dang, would I love to have bought such a card for my PC! Why don't they sell these mobile versions for PCIE-slot variants?

 

[Flapdrol1337]

Dang, would I love to have bought such a card for my PC! Why don't they sell these mobile versions for PCIE-slot variants?

If you undervolt and downclock you can probably get quite close.

They don't sell them like that because heat and power use isn't a big problem in desktops, a bit more performance sells better.

 

[BSim500]

Dang, would I love to have bought such a card for my PC! Why don't they sell these mobile versions for PCIE-slot variants?

You can actually tweak a 1060 all the way down to 60w, with surprisingly little drop in performance:-
http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1060-pascal,4679-7.html

 

[boozzer]

Easier to complain on forums and accuse everyone of being in someone's pocket. Until said site/reviewer makes a glowing endorsement and suddenly they become the only legitimate source of information.

will not happen because the current way of doing reviews is the perfect ammo for marketing.

with a legit, traceable review methodology, the spin will become very hard.

 

[boozzer]

To put this in context of what is currently available at comparable TDP:

960M - 65W - 5293.6 (1060M is 159% faster)
965M - 60W - 7619.5 (1060M is 80% faster)
970M - 75W - 9950.2 (1060M is 38% faster)

And the fastest mobile GPU Nvidia currently offers:

980M - 100W - 12459.2 (1060M is 10% faster)

and my 980m laptop just dropped in value by 800$ time for me to sell it before 1060m drops. hehehe.

 

[Flapdrol1337]

You can actually tweak a 1060 all the way down to 60w, with surprisingly little drop in performance:-
http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1060-pascal,4679-7.html

Wow, that is quite amazing.

half the power and still 80% of the performance.

 

[IEC]

Wow, that is quite amazing.

half the power and still 80% of the performance.

That will for sure be my next laptop GPU. 1060M should be a fantastic replacement for my 960M.