MSI GTX 1060 Gaming X 3 GB

[etayorius]

Thanks for the review Wizz, Finally low level API benchmarks, great news.

But I think the RX480 should be faster in Doom Vulkan, most other benchmarks i seen at 1080p Ultra Settings shows the RX480 AVG of 110-115 FPS, 90 avg fps seems somewhat low.

 

[jabbadap]

A bit of off topic, but: Heh had to chuckle a bit, confess W1zzard you could not resist a buy that monitor of yours with that kind of name on it. I kind of dislike acer as a brand, but is it any good?

 

[Artas1984]

You should stop testing BF4, the game is older than my 2-year-old cousin.

Fallout 4 seems broken in 1080p. No Man Sky (the super stupid hyped game) is broken in all resolution. Deus Ex MD has a stupidly low FPS, it's broken as well. I hope DX12 is going to fix it.

You should stop talking garbage. Techpowerup has been testing some of the older titles since beginning of time. Quake 4, Crysis, Call of Duty 4 have been tested for over 5 or 6 years in a row and there was no problem with that, as long as newer games like Crysis 2 or Call of Duty Advanced Warfare have been included together with them later on, it was interesting to compare the same older vs newer game in the past.. So expect Battlefield 4 to stay for another 2 years, as long some new Battlefield title will be included in comparison. It's trend that is common on TP.

Perhaps you ave forgotten that out of all Battlefield games, Battlefield 4 is still the most popular multiplayer game? It will stay as long as it will have a validation point in community.

 

[W1zzard]

A bit of off topic, but: Heh had to chuckle a bit, confess W1zzard you could not resist a buy that monitor of yours with that kind of name on it. I kind of dislike acer as a brand, but is it any good?

back then it was the cheapest monitor with multiple inputs and proper support for 1600x900 and 2560x1440, besides 4k. it is somewhat sensitive to thinking bios boot screen = monitor off, but other than that no issues.

that's not the monitor i use for every day work, it's only for the vga test system (my work system monitors are dell u3011 2560x1600 and a 1280x1024 eizo to the right of it)

 

[AC BEE]

Nice review.
If amd the nvidia card both are not REFERENCE.
RX480 beats GTX1060 6G a little bit.
RX470 beats GTX1060 3G a little bit and equal to GTX 1060 6G.

RX480 REF is totally disaster. It is even worse than RX470 nitro and red devil.

 

[xorbe]

... first post, bashing AMD in NV review thread ...

Welcome, you'll fit right in here!!!

 

[Ungari]

Reviews are not consistent across the board so far. For some reason the benchmarks for the 3GB vary wildly between various review sites, and way outside the margin of error.

If 3.5 VRAM on the GTX 970 was a limitation in several games 2 years ago, why isn't 3GB VRAM a bad idea in 2016?
As to the value conclusion, for the price of this MSI custom card one could get the full 8GB RX 480, never-mind that many reference 4GB 480's actually have 8GB on the board that can be unlocked, a value that cannot be beat.
I cannot understand how anyone could recommend purchasing this card.

I am dismayed that many reviewers are not chastising Nvidia on what is a deceptive practice by not renaming the cutdown version of this card to GTX 1050 or 1050Ti, something to let the consumer know it is NOT a true 1060.

 

[rtwjunkie]

I am dismayed that many reviewers are not chastising Nvidia on what is a deceptive practice by not renaming the cutdown version of this card to GTX 1050 or 1050Ti, something to let the consumer know it is NOT a true 1060.

It's actually normal for them. All the way back to the 9000 series they have had variations of the same model, distinguished only by the different VRAM amount in the name.

If customers can't read, then maybe they have no business being customers?

 

[sutyi]

Pretty useless having a 3GB VRAM with this performance class of GPU especially in this price range. New games pretty much eat up to 3.5GB at 1080p.
Unless you fancy videomemory related constant micro-stuttering in new AAA titles go 4GB+ if you get my drift...

 

[bug]

Let's pretend we would score 100 fps with 1920x1080. If we were to apply your logic we would gain 30 fps for a total of 130 fps when rendering in 1600x900. Your math showed that (1600*900)/(1920*1080)=0.69.
130 fps * 0.69 = 90 fps → Oops!
It's rather you being stubborn.

Classic mistake. That .7 factor is determined when going from 1920x1080 to 1600x900. You can't apply the same factor when going from 1600x900 back to 1920x1080 and expect to get back where you started.
It's like you earning $1,000 a months and your boss comes and tells you: due to financial difficulties I'm going to cut your paycheck by 50% (multiply by .5), but I'm going to give you 50% more (multiply by .5 again) after 6 months.

If 3.5 VRAM on the GTX 970 was a limitation in several games 2 years ago, why isn't 3GB VRAM a bad idea in 2016?

3.5 GB VRAM was never an actual limitation. Just something people liked to argue about.
The only way 3.5GB could be a limitation is if a game spent a significant amount of time precisely between 3.5 and 4 GB. And even if such a game existed, you'd lower texture quality a bit and be in the clear. Most games offer only minimal differences beyond Medium texture quality and once you're past High quality, you can usually only spot differences in screen shots.

 

[troelses]

Classic mistake. That .7 factor is determined when going from 1920x1080 to 1600x900. You can't apply the same factor when going from 1600x900 back to 1920x1080 and expect to get back where you started.
It's like you earning $1,000 a months and your boss comes and tells you: due to financial difficulties I'm going to cut your paycheck by 50% (multiply by .5), but I'm going to give you 50% more (multiply by .5 again) after 6 months.

Actually you're the one that made a classic mistake in original post and Dethroy was just trying to point this out to you by using your logic. You said that 900P gained 30% more performance due to 30% fewer pixels, however with that statement you are making the mistake of mixing terms like "more" and "fewer" without correcting for the denominator. The correct statement would have that 900P gains 44% more performance than 1080P, since 1080P is 44% more pixels, and vise versa that 1080P gets 30% lower performance since 900P is 30% fewer pixels.

In other words you basically said a reduction of 30% (in pixels) would be equalled by an increase of 30% (in performance), which is exactly the same as saying that cutting your paycheck by 50% is equalled by an increase of 50%.

 

[RaviSS]

So the new performance hierarchy taking GTX 1070 = 100 becomes (all reference cards):

GTX 1070 8GB - 100
GTX 1060 6GB - 71
RX 480 8GB - 67

GTX 1060 3GB - 65 (reference should be ~3% lower than the card tested, as cores are overclocked 4% and memory is overclocked 0%. So overall 8-9% slower than 6GB version).
RX 480 4GB - 63 (on avg. 6% slower than 8GB version)
RX 470 4GB - 54

GTX 1060 3GB easily has the best price-performance ratio at reference price.

 

[Ungari]

It's actually normal for them. All the way back to the 9000 series they have had variations of the same model, distinguished only by the different VRAM amount in the name.

If customers can't read, then maybe they have no business being customers?

There is nothing stated on the box that tells the consumer it has less CUDA cores from the 6GB version, and so many will buy this thinking the only difference is VRAM.
It is deceptive to call this card a 1060, and Nvidia should be censured for it.

 

[Ungari]

3.5 GB VRAM was never an actual limitation. Just something people liked to argue about.
The only way 3.5GB could be a limitation is if a game spent a significant amount of time precisely between 3.5 and 4 GB. And even if such a game existed, you'd lower texture quality a bit and be in the clear. Most games offer only minimal differences beyond Medium texture quality and once you're past High quality, you can usually only spot differences in screen shots.

GTA V, and Shadow Of Mordor, R6 Siege, and Skyrim with HD Texture Packs exceed 3GB VRAM, ---the limitations are real.

 

[rtwjunkie]

There is nothing stated on the box that tells the consumer it has less CUDA cores from the 6GB version, and so many will buy this thinking the only difference is VRAM.
It is deceptive to call this card a 1060, and Nvidia should be censured for it.

Again, they've done this now for at least ten years, and many times the difference is not just with the VRAM amount, and no one has called for their censure. It seems like you are just noticing this?

I'm not saying it's not a little bit crappy, but it's never actually been something I have seen people getting upset over.

 

[Nordic]

For all those who can't make sense of why someone would buy this card when new AAA games will use more than 3gb of memory. I don't play new AAA games. The most intensive game I play is star citizen, and that gets bad performance even on the best of hardware.

As I said before:

As far as I can tell the 1060 3gb gives about 6% less performance than the 1060 6gb at 1080p. The price difference between the cheapest option of both is $60 in the USA when I just looked 5 minutes ago. $60 cheaper for a 6% performance hit seems worth it to me.
The 480 4gb does not seem as good of a deal to me. It is the same price of $199.99 but uses a 30% more power. The 470 just doesn't match the performance.

I would rather have 4gb to 8gb vram for star citizen, but by the time that game comes out I will probably have upgraded 2 more times.

It is the best value card right now by a lot. I bought and EVGA mini itx one to go in my ITX build. It should be here today.

 

[bug]

GTA V, and Shadow Of Mordor, R6 Siege, and Skyrim with HD Texture Packs exceed 3GB VRAM, ---the limitations are real.

That's not even coherent.
The card has 3.5GB VRAM that run at full speed and an additional .5GB that run at reduced speed. The only scenario it would suffer compared to a card that has 4GB of VRAM running at full speed is when a game spends a significant amount of time exactly between 3.5 and 4 GB VRAM. If a game stay over 3GB, but under 3.5 GB VRAM, the 970 is fine. If the game goes over 4GB, the 970 would have been in trouble even with 4GB full-speed VRAM.

*and I'm not even sure what HD texture packs are. do they come standard with those titles?

 

[Ungari]

HD Textures are available to be enabled the last three games I listed.
The R6 Siege has a 6GB VRAM Texture Pack.

 

[rtwjunkie]

GTA V, and Shadow Of Mordor, R6 Siege, and Skyrim with HD Texture Packs exceed 3GB VRAM, ---the limitations are real.

Skyrim is completely unable to exceed 3GB of VRAM. The reason has to to with textures being replicated in system RAM. Since it is a 32 bit game, you can see that it presents a problem for game stability. Even the LAA patch only allows for 4GB of RAM use.

Since game files and any mods that aren't texture mods also have to be in RAM, this makes it impossible, without setting up ENB, to have an extraordinary amount of texture mods. ENB skirts this by keeping the textures only in the VRAM.

 

[Ungari]

Again, they've done this now for at least ten years, and many times the difference is not just with the VRAM amount, and no one has called for their censure. It seems like you are just noticing this?

I'm not saying it's not a little bit crappy, but it's never actually been something I have seen people getting upset over.

Can you cite a previous product where they cut CUDA but sold the card using the same name as the full CUDA model?

 

[Ungari]

It is the best value card right now by a lot. I bought and EVGA mini itx one to go in my ITX build. It should be here today.

If you are shopping for value and wish to ignore the RX 480, just wait a few months until Nvidia finally releases a new architecture Volta.
There will be a Fire Sale on Paxwell cards, especially used ones that people will be dumping on EBAY for over 50% off what they paid for them.

 

[kaspar737]

@W1zzard Why didn't you downclock the card to reference clocks like you have done previously when you don't have a reference card? And any chance to get a Palit/Gainward/PNY 1060 for a review?

 

[xorbe]

@W1zzard Why didn't you downclock the card to reference clocks like you have done previously when you don't have a reference card? And any chance to get a Palit/Gainward/PNY 1060 for a review?

With the new boost stuff, it's better to flash the card to stock speeds most likely.

 

[W1zzard]

@W1zzard Why didn't you downclock the card to reference clocks like you have done previously when you don't have a reference card?

Not reliable due to boost and other parameters in BIOS. I'm probably just gonna buy a card that's very close to stock

 

[kaspar737]

Not reliable due to boost and other parameters in BIOS. I'm probably just gonna buy a card that's very close to stock

https://www.computerbase.de/2016-08/geforce-gtx-1060-279-euro-test/

The Gainward/Palit/PNY (all the same card) should be a close match to reference and would be a good chance to kill 2 birds with one stone (seen others here asking for a review of this card as well).