The 980 performance prediction poll

[BrightCandle]

Limited by the same process technology its not going to be anything dramatic if it is faster, more like it will be more efficient. With a lower bus bandwidth I think we'll see it be both faster and slower than the existing 780 in the same way we often see better processor performance and worse memory performance play out.

The big jump is still going to come with TSMC 20/16nm and not this refresh.

 

[QuantumPion]

I find it hard to believe that nvidia would release a 980 that is not significantly faster than its predecessor. If the 980 was merely on par with the 780ti they would have called it the 970 and released a better card later on. I'm guessing at least 20% faster than a 780ti at $500 or $550.

 

[Anarchist420]

i dont really appreciate differences in frame rate so im not going to make a prediction about the 980's performance (whatever that is) relative to the 780Ti.

but i do care a lot about image quality and input lag. power consumption isnt an issue to me.

i am sure that i am not the only one who cant wait until hardware rendering becomes a dead end.

 

[MathMan]

15% faster. I'm an optimist, so I chose 15%-30%.

$500 seems reasonable. That would already put pricing pressure against the $430 of R290.

More interesting question: how much will they price a 970. If it follows tradition and disables 2 cores out of 15 or 16 (15% slower) and it is priced at $350, then R290 and R285X will be in a world of hurt.

 

[Anarchist420]

i also think that subtle image quality differences are made up in each new driver and that only certain parts are effected so that performance appears to go up... it has to do with marketing. they arent all bad, but they would probably focus on drivers more if their later hardware was more different than their old hardware. remember the G80 and how they had trouble until some time in the 100 series (but then how the 100 series was relatively stable)?

All of that may have something to do with the possibility of not much radical change from GK to GM just like GK wasnt radically different from GF (which itself wasnt radically different from previous architectures). They're still making all their GPUs with hardware blending and hardware depth after all. CUDA doesnt seem like it has outlived its usefullness.

 

[Jumpem]

15% faster. I'm an optimist, so I chose 15%-30%.

$500 seems reasonable. That would already put pricing pressure against the $430 of R290.

More interesting question: how much will they price a 970. If it follows tradition and disables 2 cores out of 15 or 16 (15% slower) and it is priced at $350, then R290 and R285X will be in a world of hurt.

The AMD cards should already be in a world of hurt. I am not sure why they would sell with g-sync monitors becoming available. I am sure people will still buy them though.

 

[Keysplayr]

Nobody thinks the GM204 will be the GTX880? If you think the 8 series is only going to be reserved for mobile, anyone can tell you that an GTX880M would not be anywhere near the same as a GTX880 although it bears the same numbers. But i could be wrong. First for everything. Look at GTX 680 vs GTX 680M. Not even the same cores.

Anyway, just thinking out loud. I know.... shhhh.... hehehe.

 

[tviceman]

New leak alleges GTX 980 will be 10% faster than 780 TI @ 170 watt power use. Leak is not confirmed to be accurate.
http://wccftech.com/geforce-gtx-980-alleged-benchmark-tdp-170w/

If true, I'd eat that up. Not a huge improvement in outright performance, but the halo cards should overclock really well with a bios mod. Looking forward to seeing the big perf/watt improvements!

 

[VulgarDisplay]

I'm going with roughly the same, smaller die, less power.

 

[Silverforce11]

New leak alleges GTX 980 will be 10% faster than 780 TI @ 170 watt power use. Leak is not confirmed to be accurate.
http://wccftech.com/geforce-gtx-980-alleged-benchmark-tdp-170w/

If true, I'd eat that up. Not a huge improvement in outright performance, but the halo cards should overclock really well with a bios mod. Looking forward to seeing the big perf/watt improvements!

256 bit bus, 32 ROPs, if it can actually beat 780ti and still on 28nm, that is very impressive.

VERY IMPRESSIVE.

 

[MathMan]

i also think that subtle image quality differences are made up in each new driver and that only certain parts are effected so that performance appears to go up... it has to do with marketing. they arent all bad, but they would probably focus on drivers more if their later hardware was more different than their old hardware. remember the G80 and how they had trouble until some time in the 100 series (but then how the 100 series was relatively stable)?

All of that may have something to do with the possibility of not much radical change from GK to GM just like GK wasnt radically different from GF (which itself wasnt radically different from previous architectures). They're still making all their GPUs with hardware blending and hardware depth after all. CUDA doesnt seem like it has outlived its usefullness.

Am I the only one who has no clue what you're talking about?

 

[Elfear]

Maybe I'm looking at this too simplistically but why would Nvidia launch a $500 980 that's faster than the 780Ti? No one would buy the 780Ti if they can get a cheaper faster card. I'm sure GM204 will be cheaper to produce than GK110 but will it be cheap enough to offset the $150 lower MSRP?

My prediction is somewhere between 780 and 780Ti performance for $500-550. That would still allow the Ti to sell for $600-650 until GM110 arrives on the scene.

The other likely possibility would be 780Ti +10-15% but sell for $600-650.

Seems like the last few generations we've all been hoping for faster performance for less money and come away disappointed (from both manufacturers). Maybe Nvidia will surprise us but that hasn't been the trend lately.

 

[Anarchist420]

256 bit bus, 32 ROPs, if it can actually beat 780ti and still on 28nm, that is very impressive. VERY IMPRESSIVE.

I dont see how there wont be more drawbacks than usual. Perhaps they'll use worse cells for the extra on die cache meaning more TDRs. Perhaps they will continue to do subtle "optimizations" like compressing the depth range at least like they have in some titles and on some products. what if filtering is more selectively applied and if nvidia retains more control over it than they have now (which is already too much)? what if integer performance is a lot worse than GCN and if they also dont do fast int to double float conversions on all their products?

but then what if everyone eats it up?

I mean, i just dont see Maxwell being very impressive and even i was pretty sure bulldozer wasnt going to be very good before i knew of any benchmarks (even though i may have been fooled into having doubts, i dont remember). not that i am smart nor can i thrive independently but patents suck.

Am I the only one who has no clue what you're talking about?

i am dependent on the intelligence of others and unable to reason independently. does that answer your question?

 

[MathMan]

Maybe I'm looking at this too simplistically but why would Nvidia launch a $500 980 that's faster than the 780Ti? No one would buy the 780Ti if they can get a cheaper faster card. I'm sure GM204 will be cheaper to produce than GK110 but will it be cheap enough to offset the $150 lower MSRP?

I forgot what the latest gm204 die size speculations are, but let's say 420mm2. That's at least a whopping 25% cheaper than gk110 (probably more due to non-linear silicon cost scaling.) And it 33% fewer DRAM chips (though 4GB instead of 3GB, but gk110 was migrating towards 6GB.) And, compared to GTX780Ti 30% lower power consumption, so cheaper cooling as well. Let's say 25% reduction in cost in total. A GTX780Ti is $588. If they price it at $500, it's only 15% less. And in the process, they railroad the AMD product line.

Let me turn around the question: why would they not do this?

Seems like the last few generations we've all been hoping for faster performance for less money and come away disappointed (from both manufacturers). Maybe Nvidia will surprise us but that hasn't been the trend lately.

There is a precedent, where Nvidia released the GTX680 as a significantly lower price than the slower (at the time) 7970.

 

[Elfear]

I forgot what the latest gm204 die size speculations are, but let's say 420mm2. That's at least a whopping 25% cheaper than gk110 (probably more due to non-linear silicon cost scaling.) And it 33% fewer DRAM chips (though 4GB instead of 3GB, but gk110 was migrating towards 6GB.) And, compared to GTX780Ti 30% lower power consumption, so cheaper cooling as well. Let's say 25% reduction in cost in total. A GTX780Ti is $588. If they price it at $500, it's only 15% less. And in the process, they railroad the AMD product line.

Let me turn around the question: why would they not do this?

The lowest priced 780Ti is $588 on Newegg. The average price is more like $680 (before rebate) for the 15 models they sell. Assuming your estimate that GM204 is 25% cheaper to produce is correct, $500 would still be less profitable per unit than GK110. Maybe Nvidia could make up for that in volume by the excitement created by a new product. Still a lot of assumptions though.

There is a precedent, where Nvidia released the GTX680 as a significantly lower price than the slower (at the time) 7970.

Compared to the 7970 the 680 was $50 cheaper. Compared to the 580 though it was the same street price at the time. So better performance for the same money. That's what I expect Nvidia will do this time around. 980 will be faster than the 780Ti for the same money.

 

[alcoholbob]

I predict four GTX 980s can break the 60 fps barrier in 4k for Crysis 3 ...since 4 Titan Blacks can do 55 fps

 

[x3sphere]

Maybe I'm looking at this too simplistically but why would Nvidia launch a $500 980 that's faster than the 780Ti? No one would buy the 780Ti if they can get a cheaper faster card. I'm sure GM204 will be cheaper to produce than GK110 but will it be cheap enough to offset the $150 lower MSRP?

My prediction is somewhere between 780 and 780Ti performance for $500-550. That would still allow the Ti to sell for $600-650 until GM110 arrives on the scene.

The other likely possibility would be 780Ti +10-15% but sell for $600-650.

Seems like the last few generations we've all been hoping for faster performance for less money and come away disappointed (from both manufacturers). Maybe Nvidia will surprise us but that hasn't been the trend lately.

Way I see it, the $500 market for video cards is bigger than the $600+ market. A 980 that's 10% faster than the Ti, runs cooler with less power draw, gets the people with 780s to upgrade.

Otherwise why upgrade if you have a non-Ti 780 and this only ends up being 5-10% faster? Going from the 780 to a Ti already not worth it, they need a slightly faster card than the Ti to win over these folks.

 

[MathMan]

The lowest priced 780Ti is $588 on Newegg. The average price is more like $680 (before rebate) for the 15 models they sell. Assuming your estimate that GM204 is 25% cheaper to produce is correct, $500 would still be less profitable per unit than GK110.

$500/$680 = 27% cheaper. Those 2% aren't going to make a big difference. It's all a big first order calculation anyway, since Nvidia sells dies, not finished GeForce GPUs, so the margin of error in estimates is large on this anyway.

Maybe Nvidia could make up for that in volume by the excitement created by a new product. Still a lot of assumptions though.

Yes, that too. And, as x3sphere points out, you need to give people a reason to buy something now.

Compared to the 7970 the 680 was $50 cheaper. Compared to the 580 though it was the same street price at the time. So better performance for the same money. That's what I expect Nvidia will do this time around. 980 will be faster than the 780Ti for the same money.

But a 680 was 40%(?) faster than a GTX580. If the GTX980 is only 10% faster, asking the same price wouldn't be a great incentive.

 

[Elfear]

But a 680 was 40%(?) faster than a GTX580. If the GTX980 is only 10% faster, asking the same price wouldn't be a great incentive.

Actually the 680 was closer to 25-30% faster than the 580. The 680 also had some close competition with the 7970. Unless AMD magically comes out with the 300 series in the next month or two, the 980 would only have the 780Ti to compete with. I just don't see Nvidia giving everyone a bargain if they don't have to. Hopefully I'm proven wrong since a $500 card that's 15% faster than a 780Ti would be sweet.

 

[rgallant]

I have to agree with Elfear

if the 980 does come in 20% faster it will take the 780ti price spot-$700.00ish
having the 970 come in some where below the going price of the 780ti but more than the $500.00 of the 780's old listing or the same.
-the deal will be buying the 780ti until they go eol then you will end up with the $700.00 -650.00 gtx 980 and the $500.00 gtx970 .

20% faster cards at the current listings.

and I think the cards will be smaller but with highly tweaked boost[1200] to win those first benches that will remain in history forever.

 

[Termie]

Actually the 680 was closer to 25-30% faster than the 580. The 680 also had some close competition with the 7970. Unless AMD magically comes out with the 300 series in the next month or two, the 980 would only have the 780Ti to compete with. I just don't see Nvidia giving everyone a bargain if they don't have to. Hopefully I'm proven wrong since a $500 card that's 15% faster than a 780Ti would be sweet.

I have to agree with Elfear

if the 980 does come in 20% faster it will take the 780ti price spot-$700.00ish
having the 970 come in some where below the going price of the 780ti but more than the $500.00 of the 780's old listing or the same.
-the deal will be buying the 780ti until they go eol then you will end up with the $700.00 -650.00 gtx 980 and the $500.00 gtx970 .

20% faster cards at the current listings.

and I think the cards will be smaller but with highly tweaked boost[1200] to win those first benches that will remain in history forever.

Just to expand on the great points above, the 256-bit mid-level 680 replaced the 384-bit 580 and offered much better performance, but NOT a better price. It was in fact more expensive. It was also directly countering the HD7970.

Today's situation is different. Nvidia faces no competition. If the 980 is also 256-bit, and is thus replacing a 384-bit card (the 780, by the way, not the 780 Ti), I think we can expect it will either be cheaper or faster, but not both. No reason for Nvidia to do that.

As I stated before, I think the 980 will match the 780 Ti at $600, thus beating the 780 soundly in performance, but not price. It's been a VERY long time since we had a new card introduction that offered both higher performance and a lower price than a card it was replacing.