NVIDIA Announces GP102-based TITAN X with 3,584 CUDA cores

[ShiftyFella]

4 hours ago, djdwosk97 said:

I think someone is missing the fact that the Titan isn't designed to compete directly against other gaming cards. 

 

Titan ($1000, Feb 2013) = 780 ($650, May 2013) = 290 ($400, Nov 2013, 780 was $500 at this point)

Titan Black ($1000, Feb 2014) = 780Ti ($700, Nov 2013) > 290x ($550, Oct 2013)

Titan X = 980Ti ($650, June 2015) > Fury X ($650, June 2015)

 

It wasn't until around the release of Maxwell that the 290x pulled ahead of the 780Ti and that was likely to do with the fact that GCN was still receiving the full focus of AMD whereas Nvidia began focusing on Maxwell. 

 

The 490 will likely compete with the 1070, the 490x with the 1080, and the Rage X (Fury2 X) with the 1080Ti/Titan X 2016.

When it comes to gaming it's debatable but considering titan series is for professionals where compute performance is where it matters most and on that front AMD was very competitive from day one, if look at raw numbers and actual compute benchmarks. Titan had about 1/3 fp64 performance and around 4.5glops when 290x was 1/8 fp64 and 5.7glops, Titan X was 1/32 fp64 and had 6.2glops, while Fury X had 1/16 fp64 and over 8gflops. Hawaii gpus are still best bang for the buck, If you need opencl perfomance or you do lots of work with  Simulink for modeling and analysis

[Senzelian]

1 hour ago, Bouzoo said:

Iirc HBM is limited to 4,8 and 12, GDDR5X is free, like we got 6 GB GDDR5 980Ti and 1060. I'm sure someone will correct me. 

Someone told me at some point that you were only able to double the capacty of VRAM while keeping the same chip.
Because the 1080Ti should be based on a GP102 chip like the Titan it should follow the same VRAM rules.
SO 3, 6, 12, or 24GB of VRAM.

I actually don't know if any of this is true or not. I just heard something similar last time when the 980Ti launched.

[zMeul]

[kaylexmorgana]

8 hours ago, ARikozuM said:

Double precision doesn't help normal users/gamers much, but why only 12GB 5X? Might have to replace my Titans if it really is better by 60%.

Since you really need 24gbs don't you?

[DXMember]

what happened to

"Pascal will have 16GB HBM2"

[laminutederire]

8 hours ago, AresKrieger said:

 

HBM is currently pointless and hard to get, so they just didn't bother it would appear

For a 1200$ card why not go the extra mile?

[VagabondWraith]

I'm wondering if this means Volta is coming sooner than expected? Like summer 2017. That's quick succession in nVidia's cadence of graphics cards. I mean holy shit the 1080 just came out. Inb4 eBay is flooded with 1080's. lol

[i_build_nanosuits]

OMG...first the 1070/1080...then the 1060 and now a new Titan...at this rate AMD will be done with by christmas!

[AresKrieger]

10 minutes ago, laminutederire said:

For a 1200$ card why not go the extra mile?

It doesn't add anything if you can get 480GB/s out of GDDR5X, and as I said it's hard to get that doesn't just mean it's expensive, no one currently has enough supply to meet the number needed for the titan launch.

[Lethal Seraph]

Ugh,... when is HBM2.0 coming out. I ain't settling for GDDR5x yet.

[Lays]

16 minutes ago, laminutederire said:

For a 1200$ card why not go the extra mile?

If HBM2 is still extremely hard to get and in short supply, they'd probably have to raise the price even more, and it'd most likely have even worse quantity shipments vs the 1080, so it'd probably be impossible to get your hands on one. People would also be buying all the stock and re-selling it on ebay, just like they are on 1080's right now.

 

They're already having problems with 1080 development and shipment quantities, now try and make an even bigger chip with an even harder to get memory, and it'd be a nightmare. They'd either have to charge out the wazoo, or wait even longer to release it.

[i_build_nanosuits]

the name is worying me though...why TITAN X again?! why not Titan P like they first said...how many people will buy Maxwell titanX thinking it's the real thing?! i can see issues with that...especially that the cards are looking very similar and all.

[i_build_nanosuits]

2 minutes ago, Lethal Seraph said:

Ugh,... when is HBM2.0 coming out. I ain't settling for GDDR5x yet.

yeah...weak card...10 000mhz 12GB GDDR5X...PFFF who would want that?!

[AresKrieger]

22 minutes ago, DXMember said:

what happened to

"Pascal will have 16GB HBM2"

G100 line still is a mystery so my guess is that;s when we'll see HBM2 on Pascal

[laminutederire]

5 minutes ago, AresKrieger said:

It doesn't add anything if you can get 480GB/s out of GDDR5X, and as I said it's hard to get that doesn't just mean it's expensive, no one currently has enough supply to meet the number needed for the titan launch.

Like much people can afford such gpu?

[HKZeroFive]

Huh, didn't notice this earlier but this looks to be a cut down GP102 chip. Are we going back to the Kepler ways? There might not even be a GTX 1080Ti since it's gonna be 3072-3200 cores at most, which isn't that much over the GTX 1080.

 

We'll have the see how AMD responds with Vega (and hopefully, HBM2). In the meantime, we'll be stuck with a pseudo-monopoly in the high end market until 2017.

 

5 minutes ago, AresKrieger said:

G100 line still is a mystery so my guess is that;s when we'll see HBM2 on Pascal

Seeing how GP100 is focused on computing (Teslas), I don't see NVIDIA offering a GP100-based gaming card for the time being.

[i_build_nanosuits]

I'll just say this: i'm using a GTX 980ti at 2560x1440 on a 144hz Gsync monitor and i can run ANY game on ULTRA settings with AT LEAST 80FPS and usually around 100FPS...easily.

Titan X pascal: OVERKILL MUCH!

[Tom Hanks]

Buying one.

[DXMember]

7 minutes ago, i_build_nanosuits said:

yeah...weak card...10 000mhz 12GB GDDR5X...PFFF who would want that?!

Fury from 2015 has 512GB/s memory

[i_build_nanosuits]

Just now, DXMember said:

Fury from 2015 has 512GB/s memory

yeah 4GB of it and it's a stuttery mess in games...check reviews. it's trash.

[AresKrieger]

Just now, laminutederire said:

Like much people can afford such gpu?

The titanx line is often used in ai projects by numerous companies, it's not a gaming card it's a professional card and always has been, just because baller gamers buy the card doesn't mean that was its intended audience, that 44Tflop stat is while being used in ai applications below is a paper by nvidia on the technology

 

https://www.nvidia.com/content/tegra/embedded-systems/pdf/jetson_tx1_whitepaper.pdf

1 minute ago, HKZeroFive said:

Seeing how GP100 is focused on computing (Teslas), I don't see NVIDIA offering a GP100-based gaming card for the time being.

Well true I never said it was a gaming card, 

[DXMember]

Just now, i_build_nanosuits said:

yeah 4GB of it and it's a stuttery mess in games...check reviews. it's trash.

check signature

[Notional]

1 hour ago, zMeul said:

Derp

They are. It's called owning the mainstream market with the 480, 470 and 460. You think a 1200$ NVidia card, that's not even HBM2 based, will do anything useful? Nah.

 

19 minutes ago, i_build_nanosuits said:

OMG...first the 1070/1080...then the 1060 and now a new Titan...at this rate AMD will be done with by christmas!

OMG another AMD is dead by *insert arbitrary date*, because a highly priced super niche product has been paper launched. 

 

---

The titan series has turned into an e-peen card for people with more money than sense, and youtubers who get it for free for that cheap marketing. Also what happened to the GP100? Not gonna happen. HBM2 NVidia GP cards are 2017-18 Volta only.

[i_build_nanosuits]

Just now, DXMember said:

check signature

yeah then you bought into the hype and overpaid for a very bad graphics card...and you bought 2 of them...congrats son.

[DXMember]

1 minute ago, AresKrieger said:

 that 44Tflop stat is while being used in ai applications

it's not tflops it doesn't do floating point operations there

 

if it's for perfesionals, then why can't Maxwell do double precision at all and why did they disable it on Kepler?