WCCF: AMD’s Flagship ‘Fiji XT GPU’ Debuts Radeon’s Titan Equivalent Branding

[Stormflux]

This is getting out of hand. The biggest blockbuster Witcher 3 just released and with no day one driver from AMD, their current flag ship card (290x) is trading blows with the 780ti and is on the heels of the 970.

Developing a whole baseline API at the request of a prominent dev that eventually is going to reshape the entire landscape of gamedev is poor support? Not the mention this could possibly alleviate the useless need for game specific invasive blackbox driver intervention fixes from both AMD and nVidia.

But despite this all, I'll go back to nVidia drivers. Lots of people playing the Witcher 3 are experiencing a tremendous amount of bugs and driver crashes, while having a day one driver.

IMO The only reason AMD is surviving is being a budget alternative.

If you want to be known as a premium brand and charge a premium price, then you have to have the driver support and Dev relations to make that happen.
And AMD simply doesn't.

It doesn't matter if you blame it on Gameworks or not, If you want to be a premium brand and charge premium prices, then you darn well better have optimized drivers available for games on release day.
Sadly AMD still doesn't know the importance of good dev relations and release day driver support after all these yrs.

Nobody is going to pay a premium price for a GPU that requires a 6month wait after game release dates for working optimized drivers.
The only reason AMD customers even put up with it now is because their GPU's are cheaper price/perf than the competition.

Its time for AMD to stop wasting time pointing the finger at Nvidia and start being more pro-active themselves with developers to ensure games run well on AMD hardware upon release also.
If you show you have proper game/driver support for your customers, then the "premuim" label will come naturally.

Exactly.

Biggest myths

AMD no dev relations
AMD no driver support

Release day driver support is important to you, but to me it only matters if the devs break the game somehow. Otherwise, if you don't need driver updates so frequently, you don't need it.

BTW AMD was being said to have bad drivers even when they were doing monthly drivers. Its more how people feel rather than how things should be. eg. witcher 3 performance not bad, still crying about drivers. A game could be perfect on AMD hardware and if there is no driver out when the game is out, some folks will cry. Its all feeling

Edit: Top o' the page:

It's going to be a tough marketing sell. I truly believe this is the only area AMD is failing (marketing). It's something they'll have to work tremendously at to push "Premium."

 

[3DVagabond]

@Serandur
What matters more for gaming? CPU or GPU?

Frankly my i5 2500K@4.6Ghz rig would still be awesome fast if I plug in a Titan X or 390X. It would make no difference if I swap out to Haswell or Skylake.

Thanks to Mantle the new API's are going to alleviate the CPU bottleneck.

FTFY

Releasing reference cards with the worse coolers ever will not help IMO

Hopefully they aren't dumb enough to release a premium product that isn't premium quality. Remember the 290's came out specifically to undercut the Titan and 780. And to do it by a fairly substantial amount. Then circumstances (mining and dramatic supply constraint) didn't help. Circumstances that I personally think they should have not only anticipated but marketed themselves on.

It doesn't seem like this is the plan with the new flagship. It's going to be promoted as the latest and greatest tech that the competition has nothing equivalent to.

If I were AMD I'd be asking people, Would you prefer 2x as much GDDR3 or GDDR5? This is the equivalent.

Bad timing for a 4GB card imho.

If it were 8GB or even 6GB, I think they could have a big massive hit. But 4GB @ $849, that's not going to move very well (to be fair Titans are small market too, correct?)

Anyhow, my personal theory with absolutely zero hard evidence is that HBM is a bitch to manufacture, so supplies will be limited. Result is trying to dress it up as exotic and high per unit price as possible while the tech really matures.

An AMD buyer would be far better off with a pair of 8GB 290s over an $849 4GB 390X, or is that off base? I mean obviously for 4K/Surround style setups. Obviously a single 1080 display should largely be fine @ 4GB, but who spends $849 for a 1080P single screen GPU?

Again, more is not necessarily better. AMD says they have been extremely wasteful with VRAM because there has been an abundance. We'll see.

 

[exar333]

Thanks to Mantle the new API's are going to alleviate the CPU bottleneck.



Hopefully they aren't dumb enough to release a premium product that isn't premium quality. Remember the 290's came out specifically to undercut the Titan and 780. And to do it by a fairly substantial amount. Then circumstances (mining and dramatic supply constraint) didn't help. Circumstances that I personally think they should have not only anticipated but marketed themselves on.

It doesn't seem like this is the plan with the new flagship. It's going to be promoted as the latest and greatest tech that the competition has nothing equivalent to.

If I were AMD I'd be asking people, Would you prefer 2x as much GDDR3 or GDDR5? This is the equivalent.



Again, more is not necessarily better. AMD says they have been extremely wasteful with VRAM because there has been an abundance. We'll see.

Kind of reminds me of the argument AMD had for Bulldozer. 'If we build it, they will come' and CMT will rule all. We know how well that went.

As others have said, we need to wait and see, but for AMD to charge close to a $1000 for a GPU (if the rumor is true) 4GB will be a hard sell. I don't see resale on this gen as very good, and that applies to the Titan X as well. Be careful what you shell out good $$$ for in 2015, because I don't see it holding it's value like some of the more recent cards in the past 3-5 years.

 

[Phynaz]

Only if you're making a loss on each sale.

It means much more than that. It means that you bring no other value to the table. Once you lose your price advantage (and you will) you have lost the ability to effectively sell your product.

 

[maddie]

It means much more than that. It means that you bring no other value to the table. Once you lose your price advantage (and you will) you have lost the ability to effectively sell your product.

Your statement is very incomplete and misleading.

FACT

AMD is the low price alternative.

If this is true then you have to be the best at low cost. Once you accept the reality that you are the present low cost alternative and cannot outspend your competition, other strategies are needed to stay in business. Automated design as much as possible, reusable logic blocs that can be recombined easily, etc. For all the critism, Rory appears to have encouraged this and we are now starting to see a nimbler AMD.

Low costs and the implied low price allows you to capture markets inaccessable to your high margin competitors.

Premium products and pricing is great for % profits but not always for total profit. The rise of Walmart for example should surely indicate that low cost can work. You just have to structure your business around this goal.

 

[thilanliyan]

People talking about launch day drivers...uhhh, for Witcher 3 I'm running Cats 14.6 (needed for mining) and my performance is pretty good...50+ fps at 1920x1200, mix of settings with hairworks on. Have foliage draw distance and shadow quality on medium, rest on high or ultra.
Soooo....launch day drivers are not always necessary. That is with a single 290 btw, not xfire. IMO, I've never felt like I was using a premium product when using nVidia, and never felt like a peasant using AMD/ATI hardware. The driver support has been about par in my case having used single cards.

 

[Gunbuster]

Honestly I'm most curious to see what the box art and marketing materials will look like. Knowing AMD Ruby will show up in a leather/latex thong, you know "high end"

 

[boozzer]

IMO The only reason AMD is surviving is being a budget alternative.

If you want to be known as a premium brand and charge a premium price, then you have to have the driver support and Dev relations to make that happen.
And AMD simply doesn't.

It doesn't matter if you blame it on Gameworks or not, If you want to be a premium brand and charge premium prices, then you darn well better have optimized drivers available for games on release day.
Sadly AMD still doesn't know the importance of good dev relations and release day driver support after all these yrs.

Nobody is going to pay a premium price for a GPU that requires a 6month wait after game release dates for working optimized drivers.
The only reason AMD customers even put up with it now is because their GPU's are cheaper price/perf than the competition.

Its time for AMD to stop wasting time pointing the finger at Nvidia and start being more pro-active themselves with developers to ensure games run well on AMD hardware upon release also.
If you show you have proper game/driver support for your customers, then the "premuim" label will come naturally.

with how gameworks works, I 100% disagree. premium branding is just public perception, can be easily change with enough marketing. nv doesn't magically have better hardware. that is all.

 

[gamervivek]

Reading some of the nonsense here, AMD should've played tit for tat instead of taking the high road during their halcyon days.

What I could have done as the Developer Relations guy at AMD is say "actually, what they're doing is a reasonable business investment and I'll do exactly the same thing for all the DirectX 11 code we are adding. We'll just go in an add it, and since I can't QA it on Fermi because all they've got still is a faked up board that they showed off recently, what I'll do is I'll lock it to our hardware." Morally I think that would be reprehensible, but from a business point of view I could argue in favour of it, but we think it's really the wrong thing to do and we've not locked a single line of DirectX 11 code. That's the difference in the way [AMD] works - we work through enablement and open standards. [Nvidia] works through closed standards and disablement, which, to me is inexcusable; it's as bad as that.

http://www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1

 

[xthetenth]

They probably should've for their own sake, but it's better this way. I really wouldn't want to have two nvidias duking it out.

 

[maddie]

Reading some of the nonsense here, AMD should've played tit for tat instead of taking the high road during their halcyon days.



http://www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1

I don't agree. That would take us back to the bad old days before DX.

Us consumers enable the BS that Nvidia is doing by supporting it through purchases. They will continue to do this until it affects the bottom line.

 

[thilanliyan]

I don't agree. That would take us back to the bad old days before DX.

Us consumers enable the BS that Nvidia is doing by supporting it through purchases. They will continue to do this until it affects the bottom line.

QFT.

 

[Headfoot]

"Premium" is completely made up hand-waving used to rationalize irrational preferences

 

[xthetenth]

A premium is something you pay, not something you get.

 

[Headfoot]

A premium is something you pay, not something you get.

Well said

 

[RussianSensation]

RS, what makes you think there will be less console ports moving forward? Also, GTA V is excellent, optimized scaling from low to high hardware very well.

I am not disagreeing that there will be a lot of console ports but with next gen. I am saying if 970/980 are anything to go by, we should have a card at $399-449 that is as fast or faster than $700+ GM200 6GB/R9 390X or $1000 Titan X, uses less power, and has more features (DP 1.3 possibly, etc.). Who wants to buy a 4GB card for $700+ to keep > 2 years? If you have already played GTA V, TW3, Dying Light, DAI, what is the point now to get a new $700 card that's a stop-gap between now and 14nm? A lot of gamers keep talking about 4K but the reality is you need 2x Titan Xs for that. That means you'd really need to spend $1200-1500 to get a far superior experience to 1440P maxed out. But compare HD7990/680 SLI vs. a single 980 and it's almost foreshadowing that a $500-550 Pascal card could be as fast as Titan X SLI at 4K.

As far as "Titan" branding for the halo effect for AMD, I am not sure it'll work at all. R9 295X2 easily beat the 980/780Ti and by a lot too, but that hardly mattered. Gamers didn't even buy it for $600 USD, nevermind its original $1500 price. Most gamers aren't really interested in $699-999 GPUs. Having a halo effect isn't about a $1K GPU but about its performance. Before NV even released the Titan X, NV clobbered AMD's market share with 705/750Ti/960/970/980s and mobile offerings. The most expensive of those NV cards cost $550. If AMD thinks they can release a $749-849 Fiji XT that's only 5% slower than a Titan X and sell a lot of them, they are in for a surprise.

Even enthusiast PC gamers know how fast GPUs drop in price. As a point of reference, one of the best R9 290X cards, Sapphire Tri-X, is going for $280 on Newegg, a card that cost $550+ November 2013. You kinda have to ask yourself is it really worth it to drop $700+ on a 28nm card when Pascal seems out of this world on paper?

This is ridiculous if true. The whole inflated GPU pricing thing would have completely enveloped the market. I'm deeply saddened and angered by this possibility and will scrap my whole 2015 rebuild plans if this nonsense comes to be true. I need more processing power, but it's seeming exceedingly clear that it would be wiser to save my money for Skylake-E/Cannonlake over Skylake and the same for Pascal/Arctic Islands over Fiji/GM200.

The market's screwed.

That's what happens when mid-range GPUs cost $500-550 and high-end is now $699-999 -- we now have to wait longer to get double the performance at the same price. There is still no GPU from NV/AMD at $499 that's double the speed of a 7970Ghz that launched June 21, 2012. NV provides such a product in the form of a Titan X for $999.

If we hypothetically call NV's large monolith die a Tick, and its mid-range next gen a Tock, then we get:

670/680 = Tock (new micro-architecture)
780/780Ti = Tick (advances on existing micro-architecture with more perfomrance)
970/980 = Tock
GM200 6GB = Tick

Seeing how older architectures become outdated and how quickly GPUs drop in price with new generations of cards, it might be better to upgrade on every Tock?

The Witcher 3 - 970 OC is 2X faster than 670 OC, which means a $330 970 released September 26, 2014 is faster than $800 GTX670 SLI released May 10, 2012 or $699 780Ti released Nov 7, 2013.

Because of this, I really feel like buying $700-1000 28nm flagship to hold on to > 2 years is a bad idea. Either architectural advancements or NV/AMD stopping to optimize drivers for older architectures will likely mean flagships of this gen will get killed by next gen mid-range Pascal $400 card. I mean we are talking 14nm shrink + HBM2 + new architecture.

Be careful what you shell out good $$$ for in 2015, because I don't see it holding it's value like some of the more recent cards in the past 3-5 years.

Well said. I still have hope that management at AMD isn't dumb enough to price Fiji at $700+ but have no cut-down/2nd tier Fiji cards between $400-600. I am pretty sure the price/performance king in AMD's line will be 390 non-X. Pretty much AMD's 2nd tier card since HD5850/5870 generation has been the one to buy. Their flagships are hardly worth the extra premium imo.

They can go right ahead and price Fiji XT with water cooling for $700-750 but if there is a 3584-3840 air cooled Fiji for $450-550, then at least we have progress from where we are today with the 290X/970/980.

 

[jpiniero]

I think you are being extremely optimistic on what Pascal's performance improvement will be; esp when the $/transistor improvement from 28 to SS14 is quite questionable.

 

[Abwx]

I still have hope that management at AMD isn't dumb enough to price Fiji at $700+ but have no cut-down/2nd tier Fiji cards between $400-600. I am pretty sure the price/performance king in AMD's line will be 390 non-X. Pretty much AMD's 2nd tier card since HD5850/5870 generation has been the one to buy. Their flagships are hardly worth the extra premium imo.

They can go right ahead and price Fiji XT with water cooling for $700-750 but if there is a 3584-3840 air cooled Fiji for $450-550, then at least we have progress from where we are today with the 290X/970/980.

Expect 850$ for the top product, 650$ for a 3584SPs version, the segments below will be filled with Hawai refreshes, if they can extract 10% more perfs than 290/290X while slightly reducing TDP it will be more than enough for their performance price points.

As already discussed i dont think that AMD will play the low pricing game, it didnt work with Hawai and there s no reasons that it would be the case with Fiji.

AMD would be much more inspired to make as much net income out of 1 Fiji chip rather than from two chips like they did with the 290, should had the market reacted positively to their initatiative that price would had become the main selling point, but as you know it happened otherwise..

 

[3DVagabond]

Kind of reminds me of the argument AMD had for Bulldozer. 'If we build it, they will come' and CMT will rule all. We know how well that went.

As others have said, we need to wait and see, but for AMD to charge close to a $1000 for a GPU (if the rumor is true) 4GB will be a hard sell. I don't see resale on this gen as very good, and that applies to the Titan X as well. Be careful what you shell out good $$$ for in 2015, because I don't see it holding it's value like some of the more recent cards in the past 3-5 years.

I'm missing the correlation with that and Bulldozer? I don't see them saying, if you build it they will come. They are talking about their own optimizations. Not relying on the industry like they did with Bulldozer.

Resale depends on how well it performs at the time and continued support. I'd say that Kepler has fallen short on those parameters, particularly due to the second part, continued support. I'm not sure why you are willing to give any benefit of the doubt that Maxwell will be supported any better once Pascal comes out.

 

[3DVagabond]

Your statement is very incomplete and misleading.

FACT

AMD is the low price alternative.

If this is true then you have to be the best at low cost. Once you accept the reality that you are the present low cost alternative and cannot outspend your competition, other strategies are needed to stay in business. Automated design as much as possible, reusable logic blocs that can be recombined easily, etc. For all the critism, Rory appears to have encouraged this and we are now starting to see a nimbler AMD.

Low costs and the implied low price allows you to capture markets inaccessable to your high margin competitors.

Premium products and pricing is great for % profits but not always for total profit. The rise of Walmart for example should surely indicate that low cost can work. You just have to structure your business around this goal.

Not totally disagreeing with you. On the subject of Walmart though, price is not the major part of their strategy. It's convenience and service. You can shop at Walmart 24/7, they will typically have what they sell in stock, and if you aren't happy they will take care of you.

 

[3DVagabond]

They probably should've for their own sake, but it's better this way. I really wouldn't want to have two nvidias duking it out.

Agreed. We'd be caught in the crossfire. Even though some people seem to advocate it.

 

[blastingcap]

AMD R9 395xtxx2 PE Rage Fury Pro All in Wonder

Fixed that for you.

I've said for a long time that I'm not going to buy another 28nm GPU. I didn't even want to buy the 750 Ti I'm running, but it was the lowest-wattage single-GPU 3-monitor solution I could find at the time.

So yeah, wake me up when it's raining 14nm discrete GPUs.

 

[RussianSensation]

I think you are being extremely optimistic on what Pascal's performance improvement will be; esp when the $/transistor improvement from 28 to SS14 is quite questionable.

The last time NV introduced a brand new GPU architecture on a new node, with much faster GDDR5 memory, performance doubled (780Ti is 2X faster than a GTX580). The mid-range next gen product of that generation, GTX680, was 35% faster than the last gen's flagship, GTX580.

Pascal brings:

1) A full node shrink
2) A brand new architecture
3) unified GPU memory + HBM2 memory that increases memory bandwidth 2.5-3X.
4) Perf/watt of 2X

Memory bandwidth will skyrocket from 336GB/sec of Titan X to 750GB/sec-1TB/sec

A 980 successor priced at $550 should beat the Titan X easily, while having 8GB of HBM2. I expect the top Pascal cards to have up 16GB or even 32GB of HBM2. As such, anyone who is trying to "future-proof" themselves with GM200 6GB/Fiji XT is wasting time, money and energy. A gamer should buy those cards if he/she needs the extra performance now, not in 18 months from now. In 18 months, 14nm GPUs will make both Fiji XT and GM200 6GB/Titan X overpriced and obsolete in terms of features. NV's driver focus will also naturally shift to Pascal. For anyone else who is content with their current GPU(s) or has no desire to go 4K right now, GM200 / Fiji are stop-gap cards imo.

AMD also has publicly claimed that its next gen products have 2X the perf/watt of its existing cards - so 2X the energy efficiency of existing products.

That means AMD is aiming for R9 490X to be at least 140% on this chart, or 40% faster than the Titan X.

If Fiji XT launches June 24, 2015 (i.e., in retail stores), just 1.65 years already takes us to mid-February 2017. I think many gamers are so conservative about future GPUs since 28nm has made us believe GPU progress has really slowed down. However, think about it: new architecture, 750GB/sec+ memory bandwidth + 14nm/16nm node. That's a recipe for one for he largest increased in performance in 1 generation, akin to what NV accomplished with the move from GTX480/580 to GTX780/780Ti.

I didn't even want to buy the 750 Ti I'm running, but it was the lowest-wattage single-GPU 3-monitor solution I could find at the time.

So yeah, wake me up when it's raining 14nm discrete GPUs.

My buddy built a rig last year and I helped him pick the parts. He is running 3x1080P monitors off i7 4790K on the MSI Gaming 3 Z97 board (has DVI, DisplayPort, VGA, and HDMI on 1 mobo). He says it works perfectly for watching YouTube videos, movies, etc. I think once Intel's GPUs support 3x4K or 5K monitors, discrete GPU for non-gaming needs is nearly dead on the desktop. Once 4K hardware encoding and 4K/5K multi-monitor support is present on Intel's CPUs, it's game over for budget discrete GPUs until 8K. I honestly can't think of any reason to ever buy a budget discrete GPU anymore with Intel's APUs onboard. The only time I think I would need that is if Intel's workstation CPUs don't have an APU in them (i.e., 5820K or Skylake-E, etc.).

 

[boxleitnerb]

As far as I have heard, the ALUs of Pascal will not be (very) different from Maxwell. What constitutes a new architecture? I doubt Pascal will be faster per SP than Maxwell. It will likely pack much more of them and clock them a bit higher. That and the increased bandwidth to feed them.

 

[Serandur]

That means AMD is aiming for R9 490X to be at least 140% on this chart, or 40% faster than the Titan X.

6144 GCN shader part incoming.