AMD Radeon Fiji XT Spotted in Shipping

[Psykomantis00]

It's one unit and it's a lot of whatever amount that money is. It's a prototype.

[Humbug]

You missed the big picture. Nvidia passed AMD's performance using a 256-bit bus vs. AMD's 512-bit, and with 700 fewer cores. We all know GM 200 and 210 are coming on 20nm or 16nm FinFET too. AMD has been curb-stomped architecturally. To deny this is to admit you haven't a clue about how to read the facts in front of you.

we haven't even seen AMD's next architecture... we only have rumours that it will be 20nm and that they are waiting for the manufacturing to be ready.

People who have been following the graphics wars long enough know that this leap frogging in terms of performance, efficiency and power consumption has been happening for the last 12+ years.

What recent developments mean is that Nvidia was able to get to market faster with their new generation and they have the single GPU performance crown and even moreso the efficiency crown. This is a credit to Nvidia and well done to them.

However there are no indications that anything will change on the long term. AMD will get to market too about 6 months later than Nvidia, the leap frogging will continue, Nvidia cuts prices, releases new models, the cycle continues...

[patrickjp93]

we haven't even seen AMD's next architecture... we only have rumours that it will be 16nm and that they are waiting for the manufacturing to be ready.

People who have been following the graphics wars long enough know that this leap frogging in terms of performance, efficiency and power consumption has been happening for the last 12+ years.

What recent developments mean is that Nvidia was able to get to market faster with their new generation and they have the single GPU performance crown and even moreso the efficiency crown. This is a credit to Nvidia and well done to them.

However there are no indications that anything will change on the long term. AMD will get to market too about 6 months later than Nvidia, the leap frogging will continue, Nvidia cuts prices, releases new models, the cycle continues...

Nope, as AMD continues to ignore power and thermal concerns (needing AIO again for now a single card) will turn off the supercomputing world even if it does have HBM (which won't affect games btw since no games today are close to saturating GDDR5). Nvidia has kicked their ass with 760 fewer cores and a half-wide bus, not to mention having price-parity with AMD's upcoming launch and having an ace in the wings with the Titan 2 having 3072 cores at 1.4 GHz boost 1.1 stock. Nvidia wins this year, no contest.

[Humbug]

Nope, as AMD continues to ignore power and thermal concerns

err you haven't even seen AMD's new generation... The only problem AMD had this gen was some crappy reference coolers. It's not like 290x was a power hungry monster next to the 780ti.

 

We've seen all this before. New cards, new architecture, better perf / watt.

It's been happening for the last 12 years from AMD and Nvidia. You sound like you have been only following the graphics wars recently.

 

Like I said Nvidia wins right now with new Maxwell compared to Hawai, but long-term there are no indications that the leap frogging will stop.

[TechFan@ic]

I have a 290x from Powercolor.

http://media.bestofmicro.com/J/B/418007/original/powercolor_pcs-r9-290x.jpg,

So I'm not about to need one, but I have to say I feel the coming AMD cars will be worth their while.

I'm waiting till the 490x to upgrade. Second gen HBM and 16nm will make those a doozy.

I had a 290, bought it for $309, sold it for $309. Now waiting for something worth my while.

[Humbug]

I had a 290, bought it for $309, sold it for $309. Now waiting for something worth my while.

Why did You sell it? I have a 290 vapor x as well and considering the performance I'm getting I was thinking this should get me by the next couple of years.

[Mew]

That kids got such a big head. I'm guessing that it's photoshopped.

Kinda looks like Jay Leno to me 

[TechFan@ic]

Why did You sell it? I have a 290 vapor x as well and considering the performance I'm getting I was thinking this should get me by the next couple of years.

I sold it because I was able to without suffering any loss. I sold it just before the 900 series came out. I can now buy the same card for $220 used but I'll wait for the next generation of graphics cards instead.

[Kingofpants]

[MEC-777]

You missed the big picture. Nvidia passed AMD's performance using a 256-bit bus vs. AMD's 512-bit, and with 700 fewer cores. We all know GM 200 and 210 are coming on 20nm or 16nm FinFET too. AMD has been curb-stomped architecturally. To deny this is to admit you haven't a clue about how to read the facts in front of you.

And you're comparing AMD's old arch (meant to directly compete with the 700 series, which is does so extremely well) with nvidia's newest arch. Of course it's going to be better. It's like comparing AMD's Hawaii to Nvidia's 600 series, just before the 700 series dropped.

 

I understand what is currently on the market is all we currently have to compare against each other, but you have to put things in proper context. Someone is always going to have the lesser parts at any given time because of the leapfrogging. That doesn't mean who ever is currently on the lesser side cannot ever surpass the other. I very much see the big picture.

 

As it stands there are still many unknowns about AMD's next big GPU chips. We don't know for sure if they will be using an AIO for their reference coolers and we don't know what kind of TDP or thermal efficiency they will be running at either. With the changes they've been making with their CPUs towards better efficiency, we can safely assume they will be making an effort to improve things in that regard on their GPUs as well. Will they be able to match Maxwell's efficiency? I honestly don't know. Will they improve on efficiency from Hawaii? I would guess, most definitely. Will they surpass GM204's performance? Easily. Hawaii isn't that far behind to begin with and still stands it's ground at higher resolutions.

 

 

GM200 and 210 are quite a ways off so that shouldn't even be considered right now. By the time GM200/210 are released, AMD will be working on their next leap, and so on. To say "AMD has been curb-stomped, architecturally" without knowing what AMD's next arch to compete with the 900's will actually be like, or how it will run and perform is being a little ignorant, don't you think?

 

GM204 was a huge step forward in efficiency, but only a small step in performance. To a lot of gamers, efficiency still takes a back seat to performance.

 

Just to be clear; I'm not denying the 970/980 are the best cards currently on the market in many ways. I'm just saying that given their rather small performance bump over the 700 series, I find it amusing when people don't think AMD can out perform the current 900's with their next big chips. That's all. 

[NoNameYet]

It's one unit and it's a lot of whatever amount that money is. It's a prototype.

75000 Indian Rupees = 1400 cad / 1200 usd / 980 eur

I also think it's a prototype.

[Mooshi]

The majority of consumers, many LTT members and myself included, don't give a shit about technical dick measuring specs. The only thing that matters is how well these cards will perform at their price point vs the competition. To deny that makes you a fanboy. Money speaks, not whether or not you support team red or green. If these cards are priced well and have nice performance, many will be pleased. My card has overheat issues in spite of changing TIM twice and guess what, it's Nvidia. Those AMD are heater jokes were stale forever ago. Both AMD and Nvidia make good and bad cards, wait and see what this gen brings before spewing nonsensical rage.

[MEC-777]

The majority of consumers, many LTT members and myself included, don't give a shit about technical dick measuring specs. The only thing that matters is how well these cards will perform at their price point vs the competition. To deny that makes you a fanboy. Money speaks, not whether or not you support team red or green. If these cards are priced well and have nice performance, many will be pleased. My card has overheat issues in spite of changing TIM twice and guess what, it's Nvidia. Those AMD are heater jokes were stale forever ago. Both AMD and Nvidia make good and bad cards, wait and see what this gen brings before spewing nonsensical rage.

+1. Well said.

[patrickjp93]

And you're comparing AMD's old arch (meant to directly compete with the 700 series, which is does so extremely well) with nvidia's newest arch. Of course it's going to be better. It's like comparing AMD's Hawaii to Nvidia's 600 series, just before the 700 series dropped.

I understand what is currently on the market is all we currently have to compare against each other, but you have to put things in proper context. Someone is always going to have the lesser parts at any given time because of the leapfrogging. That doesn't mean who ever is currently on the lesser side cannot ever surpass the other. I very much see the big picture.

As it stands there are still many unknowns about AMD's next big GPU chips. We don't know for sure if they will be using an AIO for their reference coolers and we don't know what kind of TDP or thermal efficiency they will be running at either. With the changes they've been making with their CPUs towards better efficiency, we can safely assume they will be making an effort to improve things in that regard on their GPUs as well. Will they be able to match Maxwell's efficiency? I honestly don't know. Will they improve on efficiency from Hawaii? I would guess, most definitely. Will they surpass GM204's performance? Easily. Hawaii isn't that far behind to begin with and still stands it's ground at higher resolutions.

GM200 and 210 are quite a ways off so that shouldn't even be considered right now. By the time GM200/210 are released, AMD will be working on their next leap, and so on. To say "AMD has been curb-stomped, architecturally" without knowing what AMD's next arch to compete with the 900's will actually be like, or how it will run and perform is being a little ignorant, don't you think?

GM204 was a huge step forward in efficiency, but only a small step in performance. To a lot of gamers, efficiency still takes a back seat to performance.

Just to be clear; I'm not denying the 970/980 are the best cards currently on the market in many ways. I'm just saying that given their rather small performance bump over the 700 series, I find it amusing when people don't think AMD can out perform the current 900's with their next big chips. That's all.

Look at the engineering and benchmark leaks of the Titan 2. AMD is fried. 3072 cores at 1.4GHz with 12GB of VRAM on a 384-bit bus delivering 8.84TFlops. Also, we know for a fact it's liquid cooled. Asetek doesn't design air coolers.

Also, I've been studying heterogeneous computing all this semester and optimizing game physics with OpenCL. Almost no game will benefit from HBM for the next 4 years. AMD's card prices will go up while Nvidia's have fallen and will still have to rely on thermal solutions most consumers cannot tolerate.

Also, the performance gain of the 970 and 980 is huge. We don't have the software to properly bench them. In OpenCL tasks it's a 30% jump over the Titan Black, both cards having been thoroughly tested by our class and our nerdy professor. It's like Intel providing so much computational power despite software not taking advantage, making AMD look good when it shouldn't at only 2 ALUs per core and a 4 cycle disadvantage against Haswell for floating point mathematics.

Also, AMD has not really made their chips more efficient architecturally. The 8370s are just highly-binned 8350s with some microcode revisions for lower idle clocks and snappier movement to idle. Their GPUs haven't gotten more efficient either. Simply see the 295x2.

[Tungsten_Hue]

I am pretty sure we can only say that the new AMD card will be water cooled is because they are under contract with Asetek no? This does not mean their new card is guaranteed to have high energy demands.

 

My 2 cents.

[HippY]

@patrickjp93

But as we know, the titan not a gaming card

and as we dont know what ATi will come up, just roumors

Yes, Amd cpus are bad, but not the GPUs

295x2 had higher clockrates like ref board, while being quiet

290/x are competetive card above 1080 p against 780

We have no info regarding new cards just roumors so please stop being Nostradamus

Personally i would live an aio cooled card as you could dump all the heat out off the case quetly and quickly more effitientlty then on a blower cooler

Why do people tend to forgot that amd/ATi tend to produce good gpu s and nvidia has their downs?

[patrickjp93]

@patrickjp93

But as we know, the titan not a gaming card

and as we dont know what ATi will come up, just roumors

Yes, Amd cpus are bad, but not the GPUs

295x2 had higher clockrates like ref board, while being quiet

290/x are competetive card above 1080 p against 780

We have no info regarding new cards just roumors so please stop being Nostradamus

Personally i would live an aio cooled card as you could dump all the heat out off the case quetly and quickly more effitientlty then on a blower cooler

Why do people tend to forgot that amd/ATi tend to produce good gpu s and nvidia has their downs?

AMD had a card so hot they couldn't design a reference cooler that used air. Engineering failure, architecture failure, end of story.

Who says the single-card Titans aren't good for gaming? They're also compute cards for those dabbling in scientific computing. Furthermore, if you think there won't be a Maxwell refresh using those specs, you're dreaming.

The info about new cards is staring you in the face. Also, WCCF predicted the fully unlocked Hawaii with 4200 cores a long time ago. If it happens I demand an apology from you since you seem to think no analyst can figure out AMD's roadmap and patterns. It's a pre-Lisa Su project, and therefore it will be the same failures in engineering as the 295x2, especially since we have seen the shroud and know Asetek made it (guaranteeing water cooling since Asetek doesn't do air).

AIO might be fine for you, but most gamers and prosumers still do not trust liquid cooling or their ability to build with it. And it's still just an excuse to have a 325+ watt single card.

I predict AMD makes a slightly better gaming card than the 980, picks up no adoption in the HPC world, and then gets crushed by the Titan 2/Maxwell refresh. You heard it here and I stick by the notion.

AMD only wins slightly in gaming performance, but they make almost no money because they forget bigger markets than gamers have concerns ranked higher than raw performance. If it takes extra electricity to drive, that is cost multiplied by the number of cards deployed each month. Extra heat? It takes 1.28 Joules devoted to cooling in the most state of the art data centers for each joule of heat you need to dissipate.

Cheering on AMD for the 200 series is only leading to their downfall and playing third fiddle to Intel and Nvidia in that order.

[HalGameGuru]

AMD had a card so hot they couldn't design a reference cooler that used air. Engineering failure, architecture failure, end of story.

Who says the single-card Titans aren't good for gaming? They're also compute cards for those dabbling in scientific computing. Furthermore, if you think there won't be a Maxwell refresh using those specs, you're dreaming.

The info about new cards is staring you in the face. Also, WCCF predicted the fully unlocked Hawaii with 4200 cores a long time ago. If it happens I demand an apology from you since you seem to think no analyst can figure out AMD's roadmap and patterns. It's a pre-Lisa Su project, and therefore it will be the same failures in engineering as the 295x2, especially since we have seen the shroud and know Asetek made it (guaranteeing water cooling since Asetek doesn't do air).

AIO might be fine for you, but most gamers and prosumers still do not trust liquid cooling or their ability to build with it. And it's still just an excuse to have a 325+ watt single card.

I predict AMD makes a slightly better gaming card than the 980, picks up no adoption in the HPC world, and then gets crushed by the Titan 2/Maxwell refresh. You heard it here and I stick by the notion.

AMD only wins slightly in performance, but they make almost no money because they forget bigger markets than gamers have concerns ranked higher than raw performance. If it takes extra electricity to drive, that is cost multiplied by the number of cards deployed each month. Extra heat? It takes 1.28 Joules devoted to cooling in the most state of the art data centers for each joule of heat you need to dissipate.

Cheering on AMD for the 200 series is only leading to their downfall and playing third fiddle to Intel and Nvidia in that order.

 

Are we making the argument a dual chipped 780 would have been fine under a reference cooler? 

 

AMD does have a genuine issue with power consumption, its not like they don't know that. But, for AMD, constantly sparring with nVidia, they have to focus where they can get the best bang for their R&D buck, and IF, a BIG IF, in this iteration they found a better improvement chasing the lower TDP that's what we will see, I would assume that would be their focus as they had the edge on top down performance their hangup was power, but if they found that they had an opportunity with the new build process to get big gains in actual performance with no real improvement to power, I cannot fault them for that. I do believe with the jump to a new process they will use that opportunity to reoptimize for improvements in efficiency AND aim for lower TDP at the ground level.

 

But, when you get down to it, AMD is designing and building for a market, they know their market is competitive and saturated, they know they have to make improvements in big chunks. And so long as it sells well, whether any of us likes the way they went or not, so long as its profitable they did good, if they do exactly what we want, and we sing their praises and they still lose money that's not good. In the end making the sales and generating profit is the only criteria for whether what they did was the right direction or not.

[patrickjp93]

Are we making the argument a dual chipped 780 would have been fine under a reference cooler? 

 

AMD does have a genuine issue with power consumption, its not like they don't know that. But, for AMD, constantly sparring with nVidia, they have to focus where they can get the best bang for their R&D buck, and IF, a BIG IF, in this iteration they found a better improvement chasing the lower TDP that's what we will see, I would assume that would be their focus as they had the edge on top down performance their hangup was power, but if they found that they had an opportunity with the new build process to get big gains in actual performance with no real improvement to power, I cannot fault them for that. I do believe with the jump to a new process they will use that opportunity to reoptimize for improvements in efficiency AND aim for lower TDP at the ground level.

 

But, when you get down to it, AMD is designing and building for a market, they know their market is competitive and saturated, they know they have to make improvements in big chunks. And so long as it sells well, whether any of us likes the way they went or not, so long as its profitable they did good, if they do exactly what we want, and we sing their praises and they still lose money that's not good. In the end making the sales and generating profit is the only criteria for whether what they did was the right direction or not.

It was. GTX Titan Z. Amazing? No. Fine? Yes.

 

The best bang for their R&D buck is to beat Nvidia at power and heat. They already have had the OpenCL performance advantage for years, but HPC people won't buy because the extra power eats away at money made by having the extra performance. This is why Nvidia Teslas are the single most used GPU-based accelerator. Of course, both of them are worse than the Xeon Phi from Intel for compute and power, but the Xeon Phi has its own awkward flavor of OpenCL, and some IO-intense stuff sucks on it, so it's not a perfect accelerator by any means.

 

It's not what I alone want them to do. Their shareholders want it! The HPC world outspends gamers, consumers, academic institutions, and business environments combined 16000 : 1 on GPUs. 16000 times the money spent! If AMD had even a 5% chunk of that, they could get out from under their debt from buying up ATI!

[mr moose]

It's not what I alone want them to do. Their shareholders want it! The HPC world outspends gamers, consumers, academic institutions, and business environments combined 16000 : 1 on GPUs. 16000 times the money spent! If AMD had even a 5% chunk of that, they could get out from under their debt from buying up ATI!

 

citation?

[wpirobotbuilder]

AMD only wins slightly in gaming performance, but they make almost no money because they forget bigger markets than gamers have concerns ranked higher than raw performance. If it takes extra electricity to drive, that is cost multiplied by the number of cards deployed each month. Extra heat? It takes 1.28 Joules devoted to cooling in the most state of the art data centers for each joule of heat you need to dissipate.

Who buys a gaming GPU to stick in a real data center? They buy specially binned GPUs with lower thermal output for the same performance, with better QA for software compatibility.

 

I can't speak for the uses of AMD GPUs over NVidia GPUs in computing, but if a data center had a genuine reason to use Hawaii over GK110, then they would, regardless of the extra cost of cooling. A large portion of data center equipment is still from the 1366 days, where power efficiency was only just starting to become a concern, and where memory drew 10 watts per DIMM. They keep it around partially because it doesn't make sense to replace them with more power efficient equipment. Power and heat aren't always the first concerns of a data center.

[patrickjp93]

citation?

Look at just the top 500 supercomputers list, pull down the specs from each, multiply card counts by their current prices (lower than original buying prices in their respective years, going back to 2007 at the earliest). Compare that sum of cards*respective prices against 2 years of consumer GPU purchases from AMD and Nvidia which you can find pretty trivially (not perfect estimate, but we all know there are many smaller clusters and supercomputers built each year by the hundreds and thousands). 

 

It's about 15794:1.

[patrickjp93]

Who buys a gaming GPU to stick in a real data center? They buy specially binned GPUs with lower thermal output for the same performance, with better QA for software compatibility.

 

I can't speak for the uses of AMD GPUs over NVidia GPUs in computing, but if a data center had a genuine reason to use Hawaii over GK110, then they would, regardless of the extra cost of cooling. A large portion of data center equipment is still from the 1366 days, where power efficiency was only just starting to become a concern, and where memory drew 10 watts per DIMM. They keep it around partially because it doesn't make sense to replace them with more power efficient equipment. Power and heat aren't always the first concerns of a data center.

They're the first concerns in newly built large data centers specifically because they're going to be around a LOOOONG time generating electrical costs. Now with dataceneters going modular, all the cooling is built into a storage container which houses a chunk of a datacenter with quick disconnects for piping, electricity, and such.

 

Also, yes, the GPUs are binned more, but it's the same cores on a different PCB (to support ECC) and slightly different drivers. Also, no one has a legitimate reason to use Hawaii over Kepler. It doesn't have even remotely comparable electrical or thermal performance, despite the raw compute, but if that's your argument, you may as well get a Xeon Phi for the same price and 2 TFlops more, rendering it the perf/BTU and perf/Watt king of accelerators.

[worldwin]

Look at the engineering and benchmark leaks of the Titan 2. AMD is fried. 3072 cores at 1.4GHz with 12GB of VRAM on a 384-bit bus delivering 8.84TFlops. Also, we know for a fact it's liquid cooled. Asetek doesn't design air coolers.

Also, I've been studying heterogeneous computing all this semester and optimizing game physics with OpenCL. Almost no game will benefit from HBM for the next 4 years. AMD's card prices will go up while Nvidia's have fallen and will still have to rely on thermal solutions most consumers cannot tolerate.

Also, the performance gain of the 970 and 980 is huge. We don't have the software to properly bench them. In OpenCL tasks it's a 30% jump over the Titan Black, both cards having been thoroughly tested by our class and our nerdy professor. It's like Intel providing so much computational power despite software not taking advantage, making AMD look good when it shouldn't at only 2 ALUs per core and a 4 cycle disadvantage against Haswell for floating point mathematics.

Also, AMD has not really made their chips more efficient architecturally. The 8370s are just highly-binned 8350s with some microcode revisions for lower idle clocks and snappier movement to idle. Their GPUs haven't gotten more efficient either. Simply see the 295x2.

What are these "benchmarks" you speak of. Do they come from a reliable source? Or is it all speculation like I am sure it is.

[patrickjp93]

What are these "benchmarks" you speak of. Do they come from a reliable source? Or is it all speculation like I am sure it is.

http://www.sisoftware.eu/rank2011d/show_run.php?q=c2ffccfddbbadbe6deeadbeedbfd8fb282a4c1a499a98ffcc1f9&l=en

 

Given it's the EU's authoritative source for benchmarks, yeah, pretty reliable.