Anybody else unimpressed with new midrange Nvidia GPUs, and much higher MSRP?

[boozzer]

if amd can have a 750 ti equivalent card this gen and a gpu that beats 1070, they could gain, a hell of alot of marketshare. these 2 cards still needs to be back up with marketing, which amd doesn't have. will they have it with the launch?

 

[maddie]

They can also be incredibly loud - but given that they're cooling 275 watts, I can't say I'm surprised.

I had Gigabyte cards; they're probably on the louder side of that scale. If AMD/ATI has something good to offer this time around, I'll definitely consider it...but the 1080 has my attention.

Do you know the difference between reference and aftermarket in the context of video cards?

Your statement on the Gigabyte cards leads me to think you don't know the difference.

 

[Mopetar]

I also believe GP100 will never be in a Geforce card.
But I also believe that GP102 is real and will become the new Ti and Titan to fight Vega

What causes you to think that? P100 is shipping with 4 of the 60 SMs disabled, and I imagine that they can get some chips that could function with 8 or 12 SMs cut instead of just tossing them completely. Eventually they get enough of those and they can ship a 1080 Ti that provides a decent performance bump over the 1080. I just don't think we'll see that card anytime soon.

GP102 doesn't seem to make as much sense to me since it faces many of the same problems as GP100 in terms of lower yields while not having the upside of being able to subsidize the low yields by selling at high prices to the HPC market.

 

[Leyawiin]

I haven't bought a reference card since the GTX 260. Don't plan on starting now. Maybe the holiday season for an upgrade. There should be a number third party solutions by then.

 

[Timmah!]

What causes you to think that? P100 is shipping with 4 of the 60 SMs disabled, and I imagine that they can get some chips that could function with 8 or 12 SMs cut instead of just tossing them completely. Eventually they get enough of those and they can ship a 1080 Ti that provides a decent performance bump over the 1080. I just don't think we'll see that card anytime soon.

GP102 doesn't seem to make as much sense to me since it faces many of the same problems as GP100 in terms of lower yields while not having the upside of being able to subsidize the low yields by selling at high prices to the HPC market.

I think he expects the GP102 to be pretty much GP100 minus the 1920 FP64 units and NVlink - things not needed to gamers. Resulting chip could be then possibly in 400-500 mm range.

 

[xpea]

What causes you to think that? P100 is shipping with 4 of the 60 SMs disabled, and I imagine that they can get some chips that could function with 8 or 12 SMs cut instead of just tossing them completely. Eventually they get enough of those and they can ship a 1080 Ti that provides a decent performance bump over the 1080. I just don't think we'll see that card anytime soon.

GP102 doesn't seem to make as much sense to me since it faces many of the same problems as GP100 in terms of lower yields while not having the upside of being able to subsidize the low yields by selling at high prices to the HPC market.

Don't forget that GP100 has huge number of FP64 ALUs (1/2 FP32 rate) and 8 NVLinks that are totally useless for gaming.
In consequence, 60SM GP100 is 610mm2 when 40SM GP104 is only 300~330mm2
Ice on the cake, HBM2 interposer even lower the yields a bit more...

So a GP102 with 60 same SM as GP104, without NVLinks should be around 450mm2 and will have decent yields (with some SM disabled). Use 384bit bus memory for 12GB of 480GB/s GDDR5X and you have the perfect gaming chip that will make great Ti and Titan cards

Edit: and the Pascal range will be well spaced:
GP106 with 20SM / 200mm2 die / 128bit GDDR memory
GP104 with 40SM / 330mm2 die / 256bit GDDR memory
GP102 with 60SM / 450mm2 die / 384bit GDDR memory
GP100 with 60 HPC SM / 610mm2 die / HBM2 memory on interposer

 

[brandonmatic]

Based on what we know right now, is it right that the 1080 is likely about 5%-10% faster than aftermarket 980ti cards - which right now (at 1200mhz) are about 25% faster than a stock 980ti and are available for $620 on Newegg? In that case $549 for a 1080 isn't very exciting unless it overclocks well.

http://www.techpowerup.com/reviews/ASUS/GTX_980_Ti_Matrix/23.html
http://www.newegg.com/Product/Product.aspx?Item=N82E16814125787

 

[linkgoron]

Based on what we know right now, is it right that the 1080 is likely about 5%-10% faster than aftermarket 980ti cards - which right now (at 1200mhz) are about 25% faster than a stock 980ti and are available for $620 on Newegg? In that case $549 for a 1080 isn't very exciting unless it overclocks well.

http://www.techpowerup.com/reviews/ASUS/GTX_980_Ti_Matrix/23.html
http://www.newegg.com/Product/Product.aspx?Item=N82E16814125787

Yes, that what I've been saying for the last few days.
If it doesn't monster-OC like the 980ti, it's almost a side-grade for 980ti users.

and the non-Founder 1080 have a $599 MSRP, not $549.

 

[xpea]

Based on what we know right now, is it right that the 1080 is likely about 5%-10% faster than aftermarket 980ti cards - which right now (at 1200mhz) are about 25% faster than a stock 980ti and are available for $620 on Newegg? In that case $549 for a 1080 isn't very exciting unless it overclocks well.

http://www.techpowerup.com/reviews/ASUS/GTX_980_Ti_Matrix/23.html
http://www.newegg.com/Product/Product.aspx?Item=N82E16814125787

why people compare apples and oranges always :\
like the 1080 can't be OC ...

So yes, 1080 can be OC, and above 2GHz guaranteed

Once for all:
reference 1080 is ~25% faster than reference 980Ti
and 1080 OC will also be 25% faster than 980Ti OC (2,1GHz is 25% OC above 1700MHz reference)

Yes, that what I've been saying for the last few days.
If it doesn't monster-OC like the 980ti, it's almost a side-grade for 980ti users.

and the non-Founder 1080 have a $599 MSRP, not $549.

 

[R0H1T]

why people compare apples and oranges always :\
like the 1080 can't be OC ...

So yes, 1080 can be OC, and above 2GHz guaranteed

Once for all:
reference 1080 is ~25% faster than reference 980Ti
and 1080 OC will also be 25% faster than 980Ti OC (2,1GHz is 25% OC above 1700MHz reference)

I doubt you can guarantee that, it'll still require some serious cooling, even when air cooled, & then the power consumption will also be something like 200~225W guaranteed.

 

[linkgoron]

why people compare apples and oranges always :\
like the 1080 can't be OC ...

So yes, 1080 can be OC, and above 2GHz guaranteed

Once for all:
reference 1080 is ~25% faster than reference 980Ti
and 1080 OC will also be 25% faster than 980Ti OC (2,1GHz is 25% OC above 1700MHz reference)

WOW, they showed an overclocked 1080 with custom cooling at a PR conference! It must mean that all cards overclock similarly. There's no chance that they selected a cherry picked card!

 

[Majcric]

I'm curious to know how the 1080 overclock scales to real world performance say vs. a 980ti. My 680 overclocked extremely well but it didn't seem show that power in gaming scenarios. The 980ti I currently have when overclocked you can really feel that extra grunt.

 

[Glo.]

why people compare apples and oranges always :\
like the 1080 can't be OC ...

So yes, 1080 can be OC, and above 2GHz guaranteed

Once for all:
reference 1080 is ~25% faster than reference 980Ti
and 1080 OC will also be 25% faster than 980Ti OC (2,1GHz is 25% OC above 1700MHz reference)

Wasn't the OC'ed version of the GPU compared to reference GTX 980 Ti? And that shown 25% increase in performance over it?

 

[xpea]

I doubt you can guarantee that, it'll still require some serious cooling, even when air cooled, & then the power consumption will also be something like 200~225W guaranteed.

wow, they showed an overclocked 1080 with custom cooling at a PR conference! It must mean that all cards overclock similarly. There's no chance that they selected a cherry picked card!

Have you heard what JHH said during event ?
He was so happy about that, a big spike to AMD when he said that 1080 has huge OC headroom. He said that it will OC "easily" above 2GHz

and yes, when OC, it will be above 180W (value at reference clocks), but what card doesn't increase power when OC ? it's not specific to NV, nor black magic guys

 

[xpea]

Wasn't the OC'ed version of the GPU compared to reference GTX 980 Ti? And that shown 25% increase in performance over it?

ahah so ironic... you are the one saying that Pascal is Maxwell at FF16+, so you should know that both chips will have same OC behavior ()

 

[Glo.]

Have you heard what JHH said during event ?
He was so happy about that, a huge spike to AMD when he said that 1080 has huge OC headroom. He said that it will OC "easily" above 2GHz

and yes it will be above 180W (value at reference clocks), but whiat card doesn't increase power when OC ? it's not specific to NV, nor black magic guys

You were the one who argued that increasing the TDP would be really bad from marketing point of view, for Nvidia, when I said the TDP will be around 200W.

Well, it is. And now you change the talk 180 degrees.

ahah so ironic... you are the one saying that Pascal is Maxwell at FF16+, so you should know that both chips will have same OC behavior ()

Did you understood my question, or not?

 

[xpea]

You were the one who argued that increasing the TDP would be really bad from marketing point of view, for Nvidia, when I said the TDP will be around 200W.

yes my expectation was a bit lower, yours a bit higher.
But performance is also a bit higher than what I though.
all in one, not a big deal and I also said that 160 or 200W makes no difference in power supply and connectors. so ?

 

[R0H1T]

Have you heard what JHH said during event ?
He was so happy about that, a big spike to AMD when he said that 1080 has huge OC headroom. He said that it will OC "easily" above 2GHz

and yes, when OC, it will be above 180W (value at reference clocks), but what card doesn't increase power when OC ? it's not specific to NV, nor black magic guys

No, does it matter though? Like I said, OC up to 2GHz isn't guaranteed, not even by JHH & there's no need to over stress the importance of that particular number since actual performance gains will also be different for each gamer.

Weren't you stressing on this TDP number, & 165W for 980, multiple times previously? The thing is with such high clocks, say 25% above stock, you'll never get the same amount of uplift (25%) in performance however your power consumption could skyrocket to 200~250W, depending on the ASIC quality. All this talk about efficiency then goes out of the window.

 

[Riek]

why people compare apples and oranges always :\
like the 1080 can't be OC ...

So yes, 1080 can be OC, and above 2GHz guaranteed

Once for all:
reference 1080 is ~25% faster than reference 980Ti
and 1080 OC will also be 25% faster than 980Ti OC (2,1GHz is 25% OC above 1700MHz reference)

25% over 1733 is 2.2Ghz.

And no you don't know as it requires memory bandwidth as well and you don't know how GDDR5x and the 1080 controller will handle higher frequencies.

 

[linkgoron]

Have you heard what JHH said during event ?
He was so happy about that, a big spike to AMD when he said that 1080 has huge OC headroom. He said that it will OC "easily" above 2GHz

and yes, when OC, it will be above 180W (value at reference clocks), but what card doesn't increase power when OC ? it's not specific to NV, nor black magic guys

Well, obviously everything JHH says is 100% true, we've never heard him lie, or tell half-truths before. If we could take the PR speech at face value, we wouldn't need review sites. In addition, as Majcric mentioned, We should also see how performance (and perf/watt) behaves when increasing clocks.

Also, maybe you should re-read our posts. You'll see we both clearly said "unless it overclocks well." "If it doesn't monster-OC", so I'm not sure what you want to prove.

We should wait and see how the cards perform, but either way - as brandonmatic asked - it looks like at stock it performs a touch better than after-market 980TIs, and maybe even equivalently if you OC the 980Tis to the max.

For reference, here is the performance delta between reference Titan X,980ti and a max-ocd Gigabyte xtreme-gaming in battlefield:

and the difference before OC (in all games, not just BF):

 

[tweakboy]

Thats a nice comparison. wow.

 

[brandonmatic]

why people compare apples and oranges always :\
like the 1080 can't be OC ...

So yes, 1080 can be OC, and above 2GHz guaranteed

Once for all:
reference 1080 is ~25% faster than reference 980Ti
and 1080 OC will also be 25% faster than 980Ti OC (2,1GHz is 25% OC above 1700MHz reference)

That's a fair point. Although my point was mainly that you can buy almost that level of performance now for just slightly more money, which makes the 1080 less exciting (it's not a huge leap, which is what we were all hoping for from a new lithography).

Also note that these aftermarket 980ti cards can also overclock another 10% or so.

http://www.techpowerup.com/reviews/ASUS/GTX_980_Ti_Matrix/26.html

 

[tential]

What does AMD have anything to do with it? It's a simple matter of GP100 being very limited by the fact that it's a massive die on a new node and isn't going to yield well. They can either sell every functional die for $12,000 or they can sell it for around $1,500 because that's what the performance premium could command.

Some people would gladly pay for it because they'll always buy the most powerful card. Hell they'd probably buy two of them, but it doesn't make any financial sense for NV to sell something for 1/10 of what they could get otherwise.

Eventually they will release a consumer GP100 die, but I don't see it happening until late 2017. P100 isn't shipping until early next year so it's unlikely we see a consumer version before the summer and more likely not until next fall. The person I was replying to was asking about the 1080 lasting 6 months before NV comes out with a new card that beats it. I don't see that happening for the reasons I've outlined above. The only way it happens is if they sell an even more cut down version of GP100 as a 1080 Ti, which they may well do, but it really depends on the yields.

You addressed nothing of my post, but responded to me. My point stand, the 1080Ti will do perfectly fine, and if you think the 1080 customers will be pissed off then you are completely ignoring history.
The fact that people are betting against Nvidia marketing is laughable.

 

[Actaeon]

Haven't seen this covered yet but I'm wondering what kind of OC room is left on the GDDR5X. Maxwell loved memory bandwidth and benefited a lot from memory OCs. I expect Pascal being similar in architecture to also like memory bandwidth.

On a 1080, Nvidia had it speced at around 10.5Ghz for 336GB/s but changed it right before launch settling on 10Ghz @ 320GB/s. Not sure why they bumped it down, but it'd need 12Ghz to get to 384GB/s. I know Micron has GDDR5X chips with those speeds, but does anyone know what the memory on the 1080 is speced at?

 

[provost]

If you hold onto your card for that long, then you really have no right buying Nvidia products. You just are pledging yourself to horrendous performance for long periods of time. Get the 1080Ti if you want, but if you're going to wait 2 generations inbetween high end Nvidia purchases, you're getting outperformed by cards half the price, regularly, and that's highly disappointing.

And, this is why I am lukewarm on Nvidia cards; I don't like "renting performance" while retaining the "ownership risk" associated with the resale value of these cards. I just don't like how this deal is unfavorably skewed towards me, as the buyer. Call it a philosophical aversion to getting screwed, doesn't matter if it's for a few hundred bucks or, well a lot more meaningful and material... Lol
How many new games can one get out of a new top performing card before having to upgrade because of Nvidia's short term driver optimization cycle for new cards- 2-3 at most?... Lol
I am not as confident about 1070 doing nearly as well as the 970 from a unit sales perspective, but I guess time will tell.