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

[R0H1T]

Will be very interesting to see what the 1070 brings. JHH claiming 'faster than Titan X' but it was in the same context/method in claiming the 1080 is '2x faster & 3x more efficient', which was specific to VR and not overall performance.

6.5Tflops for the 1070 vs 9Tflops for 1080. 28% less TFlops for the 1070. If 1080 is 15-20% quicker than a stock clocked 980 Ti at 9Tflops, I don't see how a 1070 with 28% less Tflops is going to be quicker than a Titan X. Not to mention it'll be choked by limited memory bandwidth with its 256 bit bus and GDDR5.

It's loaded with caveat(s) I bet, like faster than Titan X *for resolutions up to 1440p

There's no way it can beat a Titan X across the board, especially at 4K, unless of course they'll gimp Maxwell though drivers D:

Oh man, this is pure gold.

One of the 5 Marvels of Pascal:



http://nvidianews.nvidia.com/news/a-quantum-leap-in-gaming:-nvidia-introduces-geforce-gtx-1080

What did I tell you guys, they will "enable Async Compute" in their drivers soon... for Pascal only.

It's technically NOT DX12 multi-engine or async compute, it's just better able to handle graphics <-> compute switching which means DX12 games that use AC, will not cause a performance regression. But NV is going to market it as new "AC advances" anyway.

Too early to say if it's just (faked?) Async compute that'll be the difference between Pascal & Maxwell, it could well be the clock speeds that make all (or most of) the difference between the two.

When they showed the 2.1ghz clock speed, did they show what the power usage was when overclocked?

No, but 200W at the very least is what I'd say.

 

[poohbear]

So... the founders edition will be the special OCed cards and the regular ones will need to be OCed? also... that means that OCed custom cards will appear some time later... and you know guys... those cards, no matter if is nVIDIA or AMD is far better.

I fear that nVIDIA will get a supply shortage so soon.

im guessing it'll be similar to the Gigabyte G1 gaming line, supposedly cherry picked GPUs that can overclock higher for a price premium.

 

[tential]

I don't see how AMD can continue Fiji after the 1070? If it's even 5% slower than a 980Ti, it's still under $400? The only way AMD can continue Fiji is by dropping the Nano to $350, Fury X to $400, and then releasing Polaris in the under $300 range?

I'm really confused as to how AMD plans to compete

 

[escrow4]

I don't see how AMD can continue Fiji after the 1070? If it's even 5% slower than a 980Ti, it's still under $400? The only way AMD can continue Fiji is by dropping the Nano to $350, Fury X to $400, and then releasing Polaris in the under $300 range?

I'm really confused as to how AMD plans to compete

On value. Nvidia is a rip.

 

[sontin]

Fury is dead. Why would anybody buy a 300W card with 4GB when you get the same performance with half the power and twice the memory.

Actually everything from nVidia and AMD above $300 was killed with the announcement of the GTX1070.

 

[EightySix Four]

I don't see how AMD can continue Fiji after the 1070? If it's even 5% slower than a 980Ti, it's still under $400? The only way AMD can continue Fiji is by dropping the Nano to $350, Fury X to $400, and then releasing Polaris in the under $300 range?

I'm really confused as to how AMD plans to compete

While I don't think AMD will do anything white as far as removing it from the shelves, they will probably ride out Fury based products with what little stock they have and focus all marketing efforts on TCO of VR systems vs nVidia with Polaris 10.

Cheaper upfront costs, better perf/watt, and "good enough" performance to drive today's games on a Vive/Oculus. I feel like they could make some inroads that way but it certainly won't be pretty for enthusiasts.

 

[SimianR]

Yeah, I can't imagine AMD is really going to bother with production of Fiji GPU's from this point forward. I'm guessing it will be price drops to clear what inventory is left, focus on Polaris 10 when it's available, and try to compete in the high end with VEGA when it's eventually available.

I'm still surprised that AMD is going to leave the high end to NVIDIA for 8 months or more though. That is a lot of sales left on the table. Maybe the long plan of HBM2-only designs on their high end chips ended up hurting them in this situation.

 

[BlitzWulf]

I don't see how AMD can continue Fiji after the 1070? If it's even 5% slower than a 980Ti, it's still under $400? The only way AMD can continue Fiji is by dropping the Nano to $350, Fury X to $400, and then releasing Polaris in the under $300 range?

I'm really confused as to how AMD plans to compete

For them to have any presence in that segment this year Polaris 10 Needs to be close to FuryX in performance and they should retire Fiji dies to ProDuos and workstation cards, Doesn't make sense for them to sell a 600mm2 die with a clc for around/less than a 1070

 

[Bacon1]

For them to have any presence in that segment Polaris 10 Needs to be close to FuryX in performance and they should retire Fiji dies to ProDuos and workstation cards, Doesn't make sense for them to sell a 600mm2 die with a clc for around/less than a 1070

Polaris seems to be aiming at $200ish market, maybe even under for 10. I could see 290x/980 perf for $200 selling amazing for them. Hell even for $250 that still beats out 1070 price/perf. Polaris 11 will probably be $100-$125 range unless 10 is >290x perf

AMD wants the mass sales Nvidia has enjoyed in the 750 TI and 970 price range.

 

[BlitzWulf]

Polaris seems to be aiming at $200ish market, maybe even under for 10. I could see 290x/980 perf for $200 selling amazing for them. Hell even for $250 that still beats out 1070 price/perf. Polaris 11 will probably be $100-$125 range unless 10 is >290x perf

AMD wants the mass sales Nvidia has enjoyed in the 750 TI and 970 price range.

Totally agree with you but just speculating on an alternate universe in which P10 performs that well ,wouldnt it be great to see a 480x within 5-10% of a 1070 for say $250-$300?

 

[R0H1T]

I'm still surprised that AMD is going to leave the high end to NVIDIA for 8 months or more though. That is a lot of sales left on the table. Maybe the long plan of HBM2-only designs on their high end chips ended up hurting them in this situation.

Where did you get the "8 months" from? It'll be 2 quarters IMO till Vega hits & if smaller Pascal has GDDR5x then you can bet full Polaris 10 will have it as well, at least some version of it. I'm expecting AMD will further boost Polaris clocks, till it launches, to get it within 5~15% of the stock 1080, however only the GDDR5x version(s) will have a chance to be even that close.

HBM2 is the future, even Nvidia acknowledged that with GP100, & more so is the fact that future multi GPUs will need HBM & an interposer. We can expect more small GPU's being glued together, rather than another 295x2, for obvious reasons.

 

[tential]

Polaris seems to be aiming at $200ish market, maybe even under for 10. I could see 290x/980 perf for $200 selling amazing for them. Hell even for $250 that still beats out 1070 price/perf. Polaris 11 will probably be $100-$125 range unless 10 is >290x perf

AMD wants the mass sales Nvidia has enjoyed in the 750 TI and 970 price range.

Would you rather have a 290x for $250, or a GTX 980Ti for $380?

I feel like that's a no brainer.... if that's the case, Nvidia has this wrapped up.
I'm not seeing how anything AMD can offer will be better than the 1070.

 

[MagickMan]

Meh. My 980TIs are stable at >1540 all day (both are 74% ASIC) and each performs as well as a single 1080 (within 2%), taking into account their ~50% OCs. So I see no reason to even think about upgrading until the 1080TI, or maybe even the next gen... provided they don't blow up (unlikely, since I'm only using +60mV on each and they're both WC).

I will say, though, I am rather disappointed in their pricing, the 1070 and 1080 should have been closer to $300/500 (or at least $350/550).

 

[Hitman928]

I read elsewhere that the cards don't actually support dp 1.4, that they can be used on a 1.4 monitor but don't support the extra bandwidth above 1.2. Anyone else hear/read this?

 

[renderstate]

HBM might be the future but if its cost doesn't come down anytime soon it is the wrong technology for consumer cards. GDDR5X is a much incremental, less risky and cheaper step, which could be better suited for high end consumer products.

 

[R0H1T]

I read elsewhere that the cards don't actually support dp 1.4, that they can be used on a 1.4 monitor but don't support the extra bandwidth above 1.2. Anyone else hear/read this?

Not sure what to make of this :hmm:

Nvidia 1080 Specs page

1 - 7680x4320 at 60 Hz RGB 8-bit with dual DisplayPort connectors or 7680x4320 at 60 Hz YUV420 8-bit with on DisplayPort 1.3 connector.
2 - DisplayPort 1.2 Certified, DisplayPort 1.3/1.4 Ready.
3 - Recommendation is made based on PC configured with an Intel Core i7 3.2 GHz processor. Pre-built system may require less power depending on system configuration.

Source

 

[AtenRa]

So the die is 330mm2 ?? is that confirmed ??

 

[Bacon1]

I read elsewhere that the cards don't actually support dp 1.4, that they can be used on a 1.4 monitor but don't support the extra bandwidth above 1.2. Anyone else hear/read this?

They say 1.2 certified, 1.3/1.4 ready, but I don't think there are any 1.4 DP monitors out there?

Would you rather have a 290x for $250, or a GTX 980Ti for $380?

I feel like that's a no brainer.... if that's the case, Nvidia has this wrapped up.
I'm not seeing how anything AMD can offer will be better than the 1070.

Again it all depends on your budget and what kind of performance you need.

290x performs very well and is ~25% slower than a 980 TI.

$250 -> $380 is a 50% price increase. So price/perf goes to the $250 290x card all day long. Not to mention it will consume less power since apparently thats what everyone cares about

 

[msi2]

I don't see how AMD can continue Fiji after the 1070? If it's even 5% slower than a 980Ti, it's still under $400? The only way AMD can continue Fiji is by dropping the Nano to $350, Fury X to $400, and then releasing Polaris in the under $300 range?

I'm really confused as to how AMD plans to compete

They cant and actually the same question arises for the R9 390X (and non X).

 

[Hitman928]

Not sure what to make of this :hmm:


Source

That's probably where people are getting it from and then jumping to conclusions. I don't follow the display advances much but it's quite possible that a dp 1.4 cert program isn't even available yet. Thanks for sharing.

 

[SimianR]

Where did you get the "8 months" from? It'll be 2 quarters IMO till Vega hits & if smaller Pascal has GDDR5x then you can bet full Polaris 10 will have it as well, at least some version of it. I'm expecting AMD will further boost Polaris clocks, till it launches, to get it within 5~15% of the stock 1080, however only the GDDR5x version(s) will have a chance to be even that close.

HBM2 is the future, even Nvidia acknowledged that with GP100, & more so is the fact that future multi GPUs will need HBM & an interposer. We can expect more small GPU's being glued together, rather than another 295x2, for obvious reasons.

As far as I know, VEGA is listed as Q4 2016 or Q1 2017 at the moment, meaning very late 2016 or early 2017 - so about 8 months or so. I'm not sure we'll see Polaris 10 with GDDR5x. I don't think anything we've seen so far has suggested that it will have it or that it would benefit from it, I suppose it depends on how bandwidth starved it is. I agree that HBM2 is the future, but placing all your bets on it and abandoning the high end market for 8 months seems like a poor strategy. To me, using HBM1 on Fiji didn't really work out all that well because it left them in a strange place where their older cards actually had more VRAM than their newer high end GPU's. Now it seems the wait for HBM2 is leaving them without a high end contender for a long period of time. I personally think that the Nano was the best thing to come out of using HBM1 because it showed the end user the benefit of using it - that the overall card size was significantly reduced. The other changes that HBM brought with it never really materialized in a way that benefited the end user.

I do agree though that if Polaris 10 is focused on the low-midrange/value market, it might make sense to release a higher clocked SKU perhaps with more VRAM to compete with the 1070, if it can come anywhere near its performance for a significantly lower price AMD might be able to do some damage there. But I'm not sure we'll see that level of performance from Polaris 10, I think we're looking more at 390-390X performance at a lower price and obviously much lower power use.

 

[Samwell]

Just for performance speculations of Nvidias offerings and first P10 Gpus i made a guessed calculation.
GTX1080 it's 1733 mhz against GTX980TI 1075 mhz boost clock.
speculated P10 at 1200mhz with 2560 SP (a bit higher then the 1154mhz leak in driver for small P10, as the faster model normally also has higher clocks)
Vram speed i believe in 7GBps or 8Gbps for P10, as we haven't heard anything about G5X for AMD yet.

GTX1080 vs 980TI vs P10
8,9 Tflop vs 6 Tflops vs 6,1 Tflops
ROP pixel rate, probably 64 Rops 1080 and P10: 110912 Mpix/s vs 105.776 MPix/s vs 76800 Mpix/s
Bandwidth : 320Gb/s vs 336 GB/s vs (228-256 Gb/s)?

Amd is very speculative, but as long as AMD doesn't offer much higher clocks than speculated at the moment, i don't see them near the GTX980TI. The 1080 is strongly bottlenecked by ROPs and bandwidth, but compute heavy titles might see a big jump from maxwell.

 

[Flash831]

I don't see how AMD can continue Fiji after the 1070? If it's even 5% slower than a 980Ti, it's still under $400? The only way AMD can continue Fiji is by dropping the Nano to $350, Fury X to $400, and then releasing Polaris in the under $300 range?

I'm really confused as to how AMD plans to compete

I imagine this is the reason AMD decided to launch the Radeon Pro Duo chip, to get rid of all the Fiji dies. The box even contains 1 Fiji chip as a souvenir As the ASP is 1499$ they sell it at 500/chip = same as Nano.

The Nano still might hold the lead as M-itx champ though. But Fiji never was a mass product, the 390 was.

 

[PhonakV30]

I'm looking for Crazy IPC (Like Intel) not Crazy Overclock like Vishera!

 

[iiiankiii]

Totally agree with you but just speculating on an alternate universe in which P10 performs that well ,wouldnt it be great to see a 480x within 5-10% of a 1070 for say $250-$300?

If you think about, I don't see any reason why Polaris 10 would be slower than a 390X. It just doesn't make any sense. If 14nm provides 2.2X density advantage of 28nm, that would make a 14nm 232mm ~ 28nm 510mm. Hawaii is 438mm. That means, by comparison, Hawaii is actually SMALLER than Polaris 10 if they're both on the same 28nm node.

Also, Polaris 10 will use a much smaller memory bus (256 vs 512). That should free up some more die space for Polaris 10 for more cores. Rumors suggest Polaris 10 will use gddr5x. That will solve the bandwidth problem.

Lets not forget about the higher clocks achievable through 14nm FF. Most 390/x should be able to hit 1100mhz. If Polaris 10 is able to clock upwards of 1300mhz, it might actually be able to reach Nano/fury level of performance.

Basically, if you just do the math, there shouldn't be any doubt that Polaris 10 will be at the very least, faster than a 390x.