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

[moonbogg]

Looks like a reference card to me.


SO explain to me in detail what a founders card is.

Yes I watched it.

What is the difference between a reference card and a founders card. In detail please.

Its a way to make another $100 on a mid range die! $700 for a mid range die! OMG!

OH
MY
GOD

 

[Azix]

I think the rumours might actually be true that this is maxwell with tweaks. The resulting performance is due mostly to massively increased clocks from the new manufacturing process and a small increase in shader count over the 980.

So far nothing about new dx12. But I'm just at the VR audio thing. The couple features I have heard so far are software, something they can put on maxwell as well. And that audio thing is just catching up to AMDs trueaudio (or at least results will be similar).

 

[Silverforce11]

The founders card is a custom made design by nVidia to archive "unbelievable overclocking".

It is not the reference design which will be avaiable at 27th may gloabally.

Founders edition ...

They should have called it the K edition.

1080K, binned chips with 2.1ghz OC potential.

 

[Accord99]

What is the difference between a reference card and a founders card. In detail please.

If we go by Tesla, the Founder's Edition will be a beautifully crafted and hand-made video card that you'll get first, while many of the plebes who reserved the regular video card will still be waiting 6 months later.

 

[alcoholbob]

There's a bigger difference between x70 and x80 this gen.

Last gen there was 15% diff in CUDA cores, this gen its 25% plus the increase in memory bandwidth.

Last gen 970 SLI > Single 980 made too much sense for pure performance.

But this generation a OCed 1080 might keep up with stock SLI 1070 (being a fair comparison since SLI has higher temps and noise, so you'd probably have to stay stock for comparable noise).

 

[96Firebird]

Did I miss the core count of the 1070?

 

[topmounter]

Since I'm shopping in the 970/390 price range, I see no reason to not wait and see what AMD announces.

 

[Head1985]

There's a bigger difference between x70 and x80 this gen.

Last gen there was 15% diff in CUDA cores, this gen its 25% plus the increase in memory bandwidth.

Last gen 970 SLI > Single 980 made too much sense for pure performance.

But this generation a OCed 1080 might keep up with stock SLI 1070 (being a fair comparison since SLI has higher temps and noise, so you'd probably have to stay stock for comparable noise).

Thats why we better wait for AMD.
This card(1070) is 1070 only by name.It looks like overpriced GTX1060.
crazy... how Nv marketing working.

 

[Silverforce11]

I think the rumours might actually be true that this is maxwell with tweaks. The resulting performance is due mostly to massively increased clocks from the new manufacturing process and a small increase in shader count over the 980.

So far nothing about new dx12. But I'm just at the VR audio thing. The couple features I have heard so far are software, something they can put on maxwell as well. And that audio thing is just catching up to AMDs trueaudio.

They can't put those into Maxwell.

1. Maxwell uarch does not support instant graphics <-> compute workload switch. The entire SMX/CU either does graphics or compute. If it's doing graphics and needs to switch to compute, it has to flush and wait for full idle before starting the compute task.

2. Maxwell uarch is incapable of interrupting a graphics task in flight for a compute task on priority (fine-grained preemption). It has to wait for the graphics task to finish, then flush/idle, switch to the compute workloads. Because of this, it is incapable of async timewarp for VR, if you move your head often, the async timewarp on priority preemption will get stuck behind a current draw call and does not trigger at the time it needs to.

I said this several times for awhile now, some folks used to not believe me, but it's what NV have been saying in their programming guide & dev blog since late 2014.

The new VRWorks features are Pascal only.

Pascal is Maxwell made to be GCN-like, running at very high clocks due to TSMC's 16nm FF node. It's still missing multi-engine (ACEs), but Async Compute usage will not cause a performance loss on it due to #1. Volta I think will have a multi-engine design. Late 2018 when by then I hope advanced DX12 games will be the norm.

 

[EliteRetard]

Yeah its pretty bad paying 599USD for GTX560TI-GTX680 like card.
1070 on other hand looks Ok priced, but it will be slow vs GTX1080.Nv will force people to buy overpriced GTX1080.
From this it looks like GTX1080 is 40-50% faster than TITANX, but 1070 will be i think only on par with TITANX-There will be 40% gap between 1070 and 1080.

That image is a comparison of the two midrange cards, not the high end titan x and the new midrange card.

The 1080 is about 40% faster than it's equal sibling the old 980.
Which is about right for a node shrink and new generation.
It's maximum price should be no higher than the 980 was (which already moved the price bar higher). Nvidia just keeps finding new ways to push prices higher and higher.

Since we love car analogies, that'd be like making a Kia Rio a 30K car because they developed a new engine that was a little more powerful and more fuel efficient and added some new features like standard bluetooth and power windows.

The P100 type GPU is the replacement for the 980Ti and Titan X.

 

[EightySix Four]

That image is a comparison of the two midrange cards, not the high end titan x and the new midrange card.

Do people seriously consider the 1080/980 mid range? $300ish is mid range IMO, that is high end.

 

[Silverforce11]

Do people seriously consider the 1080/980 mid range? $300ish is mid range IMO, that is high end.

It's the new mid-range. Get used to high prices for GPUs.

$379 for 1070 is better than many of us expected. So at least NV cares about the low mid-range gamers.

And unless there's a GP100 Titan or Vega this year, the 1080 will be the fastest GPU in 2016. It's got the premium associated with that, and GDDR5X OC gives it good potential scaling.

 

[EightySix Four]

It's the new mid-range. Get used to high prices for GPUs.

$379 for 1070 is better than many of us expected. So at least NV cares about the low mid-range gamers.

Lol. Man I'm getting old.

 

[Magee_MC]

Lol. Man I'm getting old.

It beats the alternative.

 

[JDG1980]

The 1080 is about 40% faster than it's equal sibling the old 980.
Which is about right for a node shrink and new generation.

Traditionally, GPU performance gains from a node shrink (at the same die size) were 80-100%. That's why I think many people are a bit underwhelmed with what Pascal is offering. Probably the reason for this is that in past eras, there was usually only one generation of GPUs on each node. Thus, a node shrink and a new (or at least seriously revamped) architecture would go hand-in-hand. This time, Nvidia went with a version of Intel's "tick-tock" strategy. Maxwell offered the big architectural gains and some process-specific optimizations for better power usage, while Pascal is basically a slightly tweaked, node-shrunk version of Maxwell. If we compared GK104 (initial 28nm) to GP104 (initial 16nm) then the cumulative improvements would probably be closer to what was traditionally seen.

 

[Azix]

They can't put those into Maxwell.

1. Maxwell uarch does not support instant graphics <-> compute workload switch. The entire SMX/CU either does graphics or compute. If it's doing graphics and needs to switch to compute, it has to flush and wait for full idle before starting the compute task.

2. Maxwell uarch is incapable of interrupting a graphics task in flight for a compute task on priority (fine-grained preemption). It has to wait for the graphics task to finish, then flush/idle, switch to the compute workloads. Because of this, it is incapable of async timewarp for VR, if you move your head often, the async timewarp on priority preemption will get stuck behind a current draw call and does not trigger at the time it needs to.

I said this several times for awhile now, some folks used to not believe me, but it's what NV have been saying in their programming guide & dev blog since late 2014.

The new VRWorks features are Pascal only.

Pascal is Maxwell made to be GCN-like, running at very high clocks due to TSMC's 16nm FF node. It's still missing multi-engine (ACEs), but Async Compute usage will not cause a performance loss on it due to #1. Volta I think will have a multi-engine design. Late 2018 when by then I hope advanced DX12 games will be the norm.

which features are you talking about?

 

[EliteRetard]

Do people seriously consider the 1080/980 mid range? $300ish is mid range IMO, that is high end.

That's entirely the point. You can't price crap as a high end product when it is not a high end product. And just because they did price the midrange product higher than the current high end, does not make it a high end product. You are confusing price for product category. We know for a fact that the 1080/1070 are the midrange products, since they already announced the high end GPU die in the P100.

But now that the midrange cards are occupying the current high end category price, they can now crap out the new high end cards at even more ludicrous prices. The new 980ti can be $1k now because "Hey it's faster than the old Titan X", instead of $650. The new Titan X can be $1.5k+, now they can justify that Titan Z product at $3k.

 

[Silverforce11]

which features are you talking about?

VR SM Projection, it's likely compute based, and their version of True Audio, also compute based calculations for sounds.

 

[96Firebird]

That's what happens when the competition is lacking...

 

[Silverforce11]

That's entirely the point. You can't price crap as a high end product when it is not a high end product. And just because they did price the midrange product higher than the current high end, does not make it a high end product. You are confusing price for product category. We know for a fact that the 1080/1070 are the midrange products, since they already announced the high end GPU die in the P100.

But now that the midrange cards are occupying the current high end category price, they can now crap out the new high end cards at even more ludicrous prices. The new 980ti can be $1k now because "Hey it's faster than the old Titan X", instead of $650. The new Titan X can be $1.5k+, now they can justify that Titan Z product at $3k.

As long as gamers buy it, why shouldn't they?

You cannot say a company that's selling their products a high price shouldn't be doing that. It's gamers who should wise up and stop spending that much for it.

 

[swilli89]

Yeah its pretty bad paying 599USD for GTX560TI-GTX680 like card.
1070 on other hand looks Ok priced, but it will be slow vs GTX1080.Nv will force people to buy overpriced GTX1080.
From this it looks like GTX1080 is 40-50% faster than TITANX, but 1070 will be i think only on par with TITANX-There will be 40% gap between 1070 and 1080.

Yeah but man those are ultra cherry picked, with only the bottom two even meaning anything. Company PR + cherry picked, very far away from averages that will be seen in RL.

 

[Sonikku]

It's the new mid-range. Get used to high prices for GPUs.

Damnit AMD get it together. (Or at least get better marketing people)

 

[happy medium]

The 1080 is about 40% faster than it's equal sibling the old 980

More like 60% faster and even more with Tomb raider direct x 12 (look at the chart).
The gtx980ti was 33% faster than a gtx980. The 1080 is 29% faster than a gtx980ti or 25% faster than a Titan x..

 

[EightySix Four]

That's entirely the point. You can't price crap as a high end product when it is not a high end product. And just because they did price the midrange product higher than the current high end, does not make it a high end product. You are confusing price for product category. We know for a fact that the 1080/1070 are the midrange products, since they already announced the high end GPU die in the P100.

But now that the midrange cards are occupying the current high end category price, they can now crap out the new high end cards at even more ludicrous prices. The new 980ti can be $1k now because "Hey it's faster than the old Titan X", instead of $650. The new Titan X can be $1.5k+, now they can justify that Titan Z product at $3k.

I was under the impression that P100 was missing a number of features necessary to be a full GPU. This price inflation is pretty absurd - I don't mind companies chasing the performance crown outright but these kinds of things are what keep my brother and his friends (a fill generation behind me) from getting involved in PC gaming like me and my friends did, they just stick to consoles. Eventually this pricing strategy is going to hurt the GPU makes in ways that can't necessarily recover from IMO.

Could be totally wrong and parents out there are dropping $1500+ on gaming rigs but it doesn't seem like the "next generation" of buyers are going to care about all of this at this rate.

 

[Silverforce11]

Damnit AMD get it together. (Or at least get better marketing people)

AMD has signal it's plan since the start of the year.

Vega is not due til Q4 2016 or Q1 2017.

Polaris 11 is entry. Polaris 10 is mainstream. Both small chips.

They aren't going to compete with 1080 and 1070 because it's in a separate market segment in terms of price, power usage, performance profile.

NV can sell the 1070 and 1080 at whatever price they deem is best for profits until Vega arrives. Even then, who here expects Vega w/ HBM2 to be cheap? Nobody should.

This is why I said get used to the new expensive mid-range.