Move Over GTX 1080, There’s A New Titan X In Town

[Kenneth Barker]

I was very much hoping for this new Titan to be HBM2 (Because I think its cool, and it is the first major advancement in chip and memory architecture in the last nearly 20 years) I would have bought 2 HMB2 Titans, even if it ran me $5,000 to do so. I have been waiting for it forever.

Wow, so how many AMD Fury cards do you have?

HBM is certainly a big jump in performance, but not more than the switch from DDR3 to GDDR5. I don't see why you care that much. Sure, they put it inside the package, and it so happens they have to stack the dies to make it fit. But it ultimately comes down to performance. It's not like the software would see it any differently. Once it's all packaged up into a product that you buy, the biggest difference is in your head (and maybe how much it costs).

If you want to talk about advancements in memory technology, 3D XPoint is a far bigger deal.

Get over yourself mate. Believe it or not, other people are allowed to be disappointed in a product. I had much higher expectations. Doesn't mean that what Nvidia is delivering here isn't a step up, its not what I expected, or wanted. I get to have those opinions as an adult, who gets to decide what he does, or does not buy.

I own one Fury X. 2 980Ti's and.... a 380X. Not sure how that matters here. But, I bought the fury just because it was cool, and an exciting advancement in NEW technology.

HBM is the next advancement from DDR, if you knew anything about memory, or computer technology advancements, you would know how big of a deal HBM really is. GDDR5X is not much different than DDR from 20+ years ago. It just clocks a lot higher, and has been enhanced over the original DDR. It is still the same tech though, and HBM is a major, MASSIVE advancement. Do some reading on it. Or don't, just don't go whining about others begin disappointed in being sold old news tech for another generation

 

[Samer1970]

HBM or no HBM , the Benchmarks are good enough to upgrade from 980 Ti to 1080 Ti for those who can afford it.

You mean GTX 1080, right? Otherwise, where did you find 1080 Ti benchmarks? Such a card has yet to be announced.

And I wouldn't replace a GTX 980 Ti with a GTX 1080. Of course it's faster, but I'll hold out for at least 1080 Ti. I guess it depends on how much $$$ you can throw around.

A GTX 1080 Ti with a little bit more cores, HBM2 maybe, and higher clock speed followed by a Titan with Full GP 100?

GP100 is supposed to lack graphics units. And, aside from HBM2, the GP102 appears to have the same shader count and higher clockspeed than GP100. So, the only reason to do it would be for HBM2.

I think they will deliver a HBM2 card to the masses, as soon as they can get costs down and production volumes up. I think they're having trouble even satisfying the existing market for the GP100's, though.

No , I mean the GTX 1080 Ti ... Which will have 12 GB RAM and faster than the new Titan X in games (as usual) ...

and since the new Titan X will be 11.x Tflops , expect that the 1080 Ti will be FAST , VERY FAST

 

[none12345]

more cores, less clock speed. Probably end up about 20% faster, maybe 25%.

I dont know if i see the point tho. It stil wont be quite fast enough to get 60fps in all games at 4k with ultra details. So, it really wont be much different then a 1080 for a lot more money /shrug.

I guess if money is no object, and you want the fastest card there is, go for it. At least until something else faster comes out.

Of course that was suppose to be the whole point of the 1080, the fastest thing you can buy even tho it wasn't really good enough to hold 60 fps in 4k. And now less the 2 months later, for anyone who bought the 1080 because it was the fastest, well better hand over another 1200 because it already lost that distinction.

 

[groundrat]

They should name it Skynet...

 

[Third-Eye]

All these people getting disappointed at the lack of HBM/HBM2 memory just makes no sense when you consider the cost added to the card for no real benefit currently. Why would Nvidia use more expensive memory when cheaper GDDR5X is good enough right now? What would even need the faster HBM memory for? If Nvidia actually used HBM memory for the GTX 1080, the cost would likely be $75-100 or more than their current MSRP and you still have the retailer mark up because of the current demand and low supply.

So the GTX 1080 could easily reach $750 US just by adding HBM and the retailer prices. HBM/ HBM2 would make far more sense in terms of cost and supply on the next GPU iteration of Pascal in 2-3 years.

 

[Aspiring techie]

*jaw drops to ground* Wha... wha... what is this monster? We wants the preciousses!!!

 

[Beholder88]

Meh, the Titan series are more built for video processing and not gaming. Not that they can't be used for gaming, but 12GB is overkill for even 4K. I just got a 1070 myself, but if I really wanted something crazy for gaming I would just wait for the 1080 ti.

 

[sera1552]

2016 is a year for graphics card. so many people leaving their old gpus even the new gpus they both are getting obsolete in just a month or year. now amd will again launch rx 490

 

[tsnor]

Wonder which government's super computer budget paid for this chip ?

 

[Samer1970]

All these people getting disappointed at the lack of HBM/HBM2 memory just makes no sense when you consider the cost added to the card for no real benefit currently. Why would Nvidia use more expensive memory when cheaper GDDR5X is good enough right now? What would even need the faster HBM memory for? If Nvidia actually used HBM memory for the GTX 1080, the cost would likely be $75-100 or more than their current MSRP and you still have the retailer mark up because of the current demand and low supply.

So the GTX 1080 could easily reach $750 US just by adding HBM and the retailer prices. HBM/ HBM2 would make far more sense in terms of cost and supply on the next GPU iteration of Pascal in 2-3 years.

Actually , in Theory HBM2 should be cheaper to make , but since it is a new tech they charge more for it thats all.

the main advantage of HBM after memory bandwidth , is the smaller Area of the PCB .. it will make wonders in Notebooks ... delaying HBM is very bad for the Mobile sector ...

 

[InvalidError]

I'm very surprised that the new Titan X has been revealed quite so early especially with how high demand is for 1080/1070/1060.

I'm not surprised: release the Titan before 1080 demand dies down and you can upsell Titans to big-walletted enthusiasts who get tired of waiting for 1080 availability. If enough chip production gets diverted to Titan X/1080Ti to meed demand to reduce 1080 availability, it may cause inflated retail prices to hold much longer.

 

[InvalidError]

Actually , in Theory HBM2 should be cheaper to make , but since it is a new tech they charge more for it thats all.

For the most part, HBM(2) is old tech (DDR DRAM) in a new TSV stack format with a 8x128bits interface. No major R&D there, mainly a die floor plan shuffle.

HBM isn't more expensive just because companies charge more for new stuff. There are all those tooling costs prior to the first sales that need to be recovered too and in this case, not many sales to spread those costs on. Simply having the mask sets for the chips made costs millions of dollars and I would not be surprised if Samsung ate a net loss on HBM1 due to insufficient volume. Unless more chip manufacturers jump on the HBM2 bandwagon soon, HBM2 may get sidelined as well - just look at Nvidia who was planning to put HBM2 on TitanX/1080Ti but ended up going with GDDR5X due to HBM2's cost and availability. That's one more strike against HBM becoming affordable any time soon.

 

[Stoneraaron1]

Any possibility for running two of these in SLI?

 

[leeb2013]

@imacow, we'll never need >8GB for 4k! Lol

 

[Au_equus]

The 780Ti had the full GK110 core while the first Titan (non-black) had one of 15 SMX disabled. The next gen 980Ti had two of 24 SMM disabled compared to the fully enabled titan x, but (in some cases, more than) made up for it with better clocks/OCing. The 2nd gen titan x (GP102) looks to be a GP100 without the HBM2 and, of course, nerfed DP. BMs will be interesting, but IMO, if trends from the previous two gens hold, I'd wait for the 1080Ti.

 

[cheesemanx]

Not that I will ever be able to afford it, but I would love to see some benchmarks with this in SLI.

 

[80-watt Hamster]

I'm very surprised that the new Titan X has been revealed quite so early especially with how high demand is for 1080/1070/1060.

I'm not surprised: release the Titan before 1080 demand dies down and you can upsell Titans to big-walletted enthusiasts who get tired of waiting for 1080 availability. If enough chip production gets diverted to Titan X/1080Ti to meed demand to reduce 1080 availability, it may cause inflated retail prices to hold much longer.

Does that even benefit Nvidia monetarily? I don't know how the relationship with GPU partners is structured. One would presume that it's like most industries, and the manufacturers buy the chips at a negotiated rate, then resell to distributors and so on down the chain. In this scenario, Nvidia would make the same amount per chip delivered regardless of retail pricing. Now, if it's somehow royalty-based, and their cut rises with sale price of cards, manipulating retailer prices would make sense. The other scenario I could see is if GP104 supply is tight and/or yields still low, and pushing some GP102s out the door keeps the cash coming in.

An other other possibility is that AMD has an upper-midrange or even high-end offering coming sooner than people expect, and Nvidia is trying to keep the spotlight on themselves. It's not all that likely, but would fit with their strategy of positioning themselves as the premium option. On the subject of launch strategies (and diverging a bit), that could be why the RX 480 seems like it wasn't quite done; it may not have been, but AMD needed to launch it ahead of the 1060 to keep Nvidia from (once again) stealing all the thunder.

 

[17seconds]

This means a whole new Pascal refresh lineup with HBM2 in 2017.

 

[EpicWarcrafter]

With no real performance competition I would'nt expect HBM2 anytime soon at a consumer level. This is just another example of Intel marketing, when you corner the market and there is no significant competition, there is no rush, and there is plenty opportunity to make money on tech that costs less at a manufacturing level.

 

[bit_user]

Believe it or not, other people are allowed to be disappointed in a product. I had much higher expectations. Doesn't mean that what Nvidia is delivering here isn't a step up, its not what I expected, or wanted.

I guess I got over that months ago, when the leaks started to indicate that GP102 would be GDDR5X.

The way I see it, they increased memory bandwidth more than compute, when compared to the GTX 1080. If that card is even bottlenecked at all, this should be less so.

HBM is the next advancement from DDR, if you knew anything about memory, or computer technology advancements, you would know how big of a deal HBM really is. GDDR5X is not much different than DDR from 20+ years ago. It just clocks a lot higher, and has been enhanced over the original DDR. It is still the same tech though, and HBM is a major, MASSIVE advancement. Do some reading on it.

I've been following HBM/HMC for years. The thing is, it's still DRAM. It's just connected via a shorter, wider connection that they can also clock lower, to save on power. All good things. It'll be a huge win, for mobile.

But even with GDDR5X, this thing has about the same bandwidth as Fury and 2/3rds as much as the GP100. So, as a practical matter, I just think it's not such a big deal.

just don't go whining about others begin disappointed in being sold old news tech for another generation

It was just something about it which sounded like entitlement. I think what Nvidia has done in recent generations is quite impressive, as compared with AMD or pretty much anyone else. I'm pleasantly surprised about how soon they're launching this GPU, and while I'd like to have seen more cores (AMD's Fiji still had more) and HBM2, it's still the biggest single-generation performance jump we've seen in quite some time.

 

[bit_user]

Wonder which government's super computer budget paid for this chip ?

Actually, the GP100 is the high-performance computing chip. This one is really aimed at graphics.

Now, speaking of the GP100, I expect they have far more private sector sales than to governments. You've no doubt heard of Big Data and Deep Learning, no? It's GPUs powering much of that.

 

[c4s2k3]

All I have to say is "Shut up, take my money!"

I've been running SLI 780 ti for 2.5 years and decided I want to drop SLI when I upgrade since fewer games are properly supporting it. But I wanted the single-card upgrade to be a performance improvement for SLI games (like BF4) before pulling the trigger. I think we're here now.

 

[InvalidError]

With no real performance competition I would'nt expect HBM2 anytime soon at a consumer level.

Once HMB enters volume production, it will become cheaper than GDDR5(X) on every front. The problem is getting that ball rolling. Once it does, it won't be long before you see HBM go all the way down to entry-level dGPUs and mid/high-end APUs. There have been rumors that some Zen-based APUs may have HBM as soon as next year.

 

[EpicWarcrafter]

I totally agree, but in the meantime lets bleed it for what its worth... if there is money to be made without competition, wouldnt you do the same?

 

[The Inquisition]

Could I feel any more like a sucker for buying 2 of the original Titan X's?
I hadn't thought so after the 1080, but here I am....