First GTX 1080 Review is HERE!

[Notional]

2 minutes ago, LAwLz said:

So far? 2 people. Both of which have failed basic reading comprehensions and math and instead jumped and attacked Nvidia on complete misinformation.

 

Maybe you missed it in the other thread but I have already said I think the price is too high and I won't be getting it (will probably get a Polaris 10 card).

The price makes sense when you do the math. Nvidia known what the card performs likes and charges according to that. Sucks for us consumers that we don't get a huge leap in price:performance, but that's what happens when there isn't any competition.

As for the Swedish price, Nvidia re just fucking us over. The prices are far more reasonable in the US.

 

It's not "amazing", but it is a good card nonetheless.

Also, you should not think of it as a "double node jump" since the jump isn't that much bigger than a regular "single jump". For example 25nm to 28nm was 38% smaller. 28nm to 16nm is 43%. So it's more like "1.2 node jumps" rather than two.

Still, there's no point. I thought we had finally moved passed that retarded term in LLT forums.

 

I understand it because AMD is not out with their cards yet. But let's say for the argument that a Polaris 10 chip is as fast, but is 150-200$ cheaper. That would make the 1080 a blatant ripoff. I guess we can blame HBM2 for not being mass produced yet for GP100 and Vega, but I honestly think both would postpone both cards anyways. I dont like this new idea of launching these "mid end" cards at such an insane price only to launch a faster chip of the same generation 4-6 months later at an even more insane price.

 

It's all of EU, Canada and probably most other countries too, not just Sweden. Even the US price is quite high though.

 

It's not just a node shrink, but also a tech switch from planar to FinFet's. That counts too.

[LAwLz]

1 minute ago, Notional said:

Still, there's no point. I thought we had finally moved passed that retarded term in LLT forums.

I think there is a point to using words like that when appropriate. It highlights how a certain person thinks in a single word. I think it would be more "retarded" to write several words that essentially mean the same thing as the word "fanboy", just because "we shouldn't say the F word".

 

 

11 minutes ago, Notional said:

I understand it because AMD is not out with their cards yet. But let's say for the argument that a Polaris 10 chip is as fast, but is 150-200$ cheaper. That would make the 1080 a blatant ripoff. I guess we can blame HBM2 for not being mass produced yet for GP100 and Vega, but I honestly think both would postpone both cards anyways. I dont like this new idea of launching these "mid end" cards at such an insane price only to launch a faster chip of the same generation 4-6 months later at an even more insane price.

Last time I checked, Polaris will not compete with the 1080. AMD have said that it is a mainstream card so I am expecting it to be something like a 380X replacement or lower. It doesn't really make sense to paint a very unlikely scenario and go "if this plays out the the 1080 would not be worth the price!". I mean sure if that scenario did happen then you would be right, but what are the odds of that happening? All evidence we got right now points at the 1080 keeping the single GPU crown for quite some time. Despite the high price the price:performance (in the US) is very solid.

 

 

15 minutes ago, Notional said:

It's all of EU, Canada and probably most other countries too, not just Sweden. Even the US price is quite high though.

Well, let's see how supply and demand handles it. If people don't think it will be worth the price then they should fall. The price will go down when non-reference cards gets released as well. Nvidia have already said that the MSRP for founders cards (the initial version) will be higher than other versions.

 

 

17 minutes ago, Notional said:

It's not just a node shrink, but also a tech switch from planar to FinFet's. That counts too.

Sure, but I doubt going to FinFet would have the same effect as 0.8 "traditional node shrinks". I think it would be a stretch to even compare this to one and a half node shrinks. The previous shrinks were just so much more drastic.

[Biecher]

so, what do you guys think about the performance, especially when it comes to overclocking?

I'm new to all this and would appreciate learning from your experience.

 

Let's assume custom cards and FE edition will cost the same. Worth waiting for like a month because of performance and temperatures?

I got a brad new PC without a dedicated GPU, thats right. It is literally standing here, doing nothing but office stuff.  

[Scitesh]

9 hours ago, don_svetlio said:

20-30% above a Titan X and stock 980 Ti. Definitely not "double the performance" as everyone was expecting

These are done without drivers so upon driver launch we would see it go up a bit more

[awesomeness10120]

23 minutes ago, Scitesh said:

These are done without drivers so upon driver launch we would see it go up a bit more

Fully Optimized drivers that is. Pascal has been supported by drivers since last fall.

[Big-G]

1 hour ago, awesomeness10120 said:

Fully Optimized drivers that is. Pascal has been supported by drivers since last fall.

Beta support yes...

[awesomeness10120]

29 minutes ago, GidonsClaw said:

Beta support yes...

Hey, support is support

 

[Mew]

5 hours ago, Simon771 said:

 

 

 

Aaaah that makes sense. But what normal person want to have GPU fan at 100%  

Unless you are using thunderbolt and have your PC in another room ... they you probbably just don't care  

Well, pretty much every guide I read on this forum tells you to spin your fan up to 100% before even doing anything.

 

[Godlygamer23]

A correction must be made about the link for Anandtech. They did a preview, not a full blown review as indicated by the length of the article, and their title.

[Kinda Bottlenecked]

7 hours ago, Bouzoo said:

And we have a test with the first aftermarket cooler, the Arctic Accelero Xtreme IV.

 

 

He managed to get 2.1 GHz while sitting at 50 C on an open bench.

What he says is that the card has a different problem, and that is capping the power limit of 1 8-pin connector in heavy OCs (if you understand german he says that at the end of the video). Seems like the custom cards might (or rather have to) come with more pins to solve this (speculations). Source

 

 

See I fuckin knew it. That single 8 pin will starve the card. Looks like I'll be mostly recommending custom PCB card. Even more so if the aim is to water cool and overclock

[Notional]

10 minutes ago, Pohernori said:

 

See I fuckin knew it. That single 8 pin will starve the card. Looks like I'll be mostly recommending custom PCB card. Even more so if the aim is to water cool and overclock

Yup, the exact same thing as the Nano: power throttling, not thermo throttling.

[TheKDub]

13 hours ago, don_svetlio said:

20-30% above a Titan X and stock 980 Ti. Definitely not "double the performance" as everyone was expecting

 

Depends on usage, and most benchmarks out there for it currently are stock speeds, not overclocked.

[Limepanda911]

Haha, suck it zmeul. Lol.

[PerfectTemplar]

13 hours ago, don_svetlio said:

You'd be surprised how many people were running around screaming "2x performance KEK"

I didn't see any of that and I've been following this for ages. They said it will beat SLI 980s, which it kind of did. It's not strictly the case though. 

 

I believe 2x/3x performance stuff was in relation to power usage and VR.

[Kinda Bottlenecked]

6 minutes ago, Notional said:

Yup, the exact same thing as the Nano: power throttling, not thermo throttling.

 

And that single 8 pin Strix 970. Atleast the produo now has 3x8pin rather than the 2x8pin in the 295x2. 

 

High end cards aren't really meant to starve like that

[Bouzoo]

3 hours ago, Notional said:

Yup, the exact same thing as the Nano: power throttling, not thermo throttling.

For a card that was thermal throttling quite often, more power would've been equal to suicide. 

[Bouzoo]

7 hours ago, LAwLz said:

Last time I checked, Polaris will not compete with the 1080. AMD have said that it is a mainstream card so I am expecting it to be something like a 380X replacement or lower. 

I've said this and I will say it again, I'll bet that AMD has a different definition of the word mainstream (or are miss using it to appeal to the masses). Either that or something equal to Fury is the new "mainstream". I highly doubt the latter. 

 

[Carclis]

12 hours ago, Energycore said:

What architectural improvement?

This is more like Ivy Bridge was back in '12. Just the node shrink and a couple features.

Well Pascal was supposed to be a big jump, I mean it took them 18 months too.

[Spectrez]

11 hours ago, LAwLz said:

I feel sorry for you. It must be hard being blind. The 1080 is performing ~60% higher on average in the benchmarks reviewers are posting.

 

It must be hard being an AMD fanboy. Blind and bad at math...

wouldnt quite call myself a AMD fanboy tbh.

 

im comparing the 1080 to the 980 ti, simply because its now in the same price range, and thus i dont see those numbers.

[Carclis]

12 hours ago, Badly Browned said:

The 1080 also has the GP104 chip with the smaller die, and it's still seeing ~30% performance improvement over the 980ti.

 

I still don't think the msrp of the 1080 is bad, but I do want to wait for what the consumer release of the GP100 chip can provide.

 

At $700 it's not that bad if you look at it as being part of the current generation. My point is that a 300mm chip should not be $700. What price do you expect a 600mm GP100 card to launch at? Certainly can't be $650 if this is anything to go by.

[Badly Browned]

7 hours ago, Carclis said:

At $700 it's not that bad if you look at it as being part of the current generation. My point is that a 300mm chip should not be $700.

 

Well, I do agree that $700 doesn't appeal much to me. Like I said, the Founders looks like an early adapter tax to me. I'd like to see how the 3rd party prices really shake out, considering they are supposed to have an MSRP of $600.

 

Quote

What price do you expect a 600mm GP100 card to launch at? Certainly can't be $650 if this is anything to go by.

 

I expect the Titan P to launch at $1000. The Titan line has been stable at that price point and I do not expect it to change.

 

Since the 1080ti will use big Pascal, I expect it to maintain the $100 price difference between the x80 and the x80ti that we saw in the 900 series. So I expect a 3rd party MSRP of $700 for the 1080ti. To note, the 780ti also released at $700.

 

That said, perhaps Nvidia will decide that going to HBM2 in addition to GP100 justifies bumping up the price on both the Titan and Ti. If Nvidia sticks to their release schedule, the Titan P should come first, and if that one is coming with a $1100+ price point, the 1080ti will likely match the price increase as well.

 

That's why I think it's important for AMD's Polaris to come out swinging and keep Nvidia competitive on the prices.

 

 

 

[Bhav]

Looks like its universally over double the performance of a 970.

 

But I just found out that the 1070 wont just have fewer cuda cores, but also regular GDDR5, so Im looking forward to reviews of those.

 

Im going to try and hold of buying until either the GP100 or even wait for Volta, unless the 1070 isnt too far behind the 1080 results.

[Valentyn]

So Jayz review is out and he also had Thermal throttling, where it dropped down to its base clock. He countered this by upping Power Limit and the fan to 80%
 

 

[Prysin]

18 hours ago, LAwLz said:

Sure, but I doubt going to FinFet would have the same effect as 0.8 "traditional node shrinks". I think it would be a stretch to even compare this to one and a half node shrinks. The previous shrinks were just so much more drastic.

in terms of leakage, it is HUGE. 20nm and below is realistically and practically impossible without it. Hell even RTGs Raja Kodouri and other people in the industry has said so (in one form or another).

[Energycore]

10 hours ago, Carclis said:

Well Pascal was supposed to be a big jump, I mean it took them 18 months too.

That's what I'm wondering. If they only shrunk the die and added a couple features to the polymorph engine, did it really take them 18 months and however many man-hours to just perform a die shrink and add Simultaneous Multi Projection? Seems kind of suspect.

 

But numbers don't lie, Pascal's CUDA cores can perform less instructions per clock (I know that's a CPU-only metric since I'm not sure GPUs have their own instruction sets but whatever) than Maxwell could. There's just more of them and they're clocked higher.