AMD Radeon Pro Duo up to 51 Percent Faster than GeForce GTX TITAN Z

[rtwjunkie]

The definition of single card is one card regardless of how many GPUs it has on it. The Voodoo 5 6000 has 4 GPUs on it, does that mean it is 4 cards? No.

When you open up the box for a 295x2 are you digging out one card or two? Only one. Single card solution.

Ok. But see my question above.

 

[MxPhenom 216]

Just for comparison.

http://www.hardwareluxx.com/index.p...cf-vs-gtx-980-ti-sli-overclocked.html?start=3

Using last generation dual card from NV is all they can do because NV doesn't have one for Maxwell. However, when you compare generation to generation, it is easy to see the massive similarities (rightly) between both camps.

Notably the Radeon Pro Duo scores lower than Crossfired Fury X's in each test.

There must be extensive drops in clocks then for the Duo.

 

[arbiter]

Sorry for double posting but it's a reply so it's not heresy! The TDP is rated at 350W so it's possible the card will throttle. Even though the power pins and PCI can deliver >500W the deign may be limiting it.

I guess you forgot that the 295x2 which they used to compare to was 5-600+watt card using only 2x8pin. So really AMD doesn't care what power pins were designed for. They will draw what ever they want.

There must be extensive drops in clocks then for the Duo.

Even slightly overclocked 980's to 1400mhz which all them can do can almost hit 25000 score

 

[RCoon]

Ok. But see my question above.

A single card solution is essentially the same thing as a dual card solution, only incredibly crippled in every shape and form. Power delivery is almost always worse than having two cards, and the methods for removing heat from dual-GPU single-card solutions are always so horribly borderline that your clocks always get throttled in everyday situations. The only reason my 295x2 doesn't get throttled is because of a full cover waterblock and powerful pump. They're also an ass to deal with when it comes to PCI-E lanes, making 16 x PCI-E 2.0 (8 x PCI-E 3.0) kinda necessary.

You still have to pray for SLI/Crossfire profiles, and yes, you need to make sure Crossfire is enabled in the first place. Dual GPU single-card solutions may have the chips on the same board, but they still rely on those technologies. It also allows companies like AMD and NVidia to claim that their card has 8GB of VRAM, when in actual fact they only have 4GB usable. Sure, there's 8GB sat on the board, but it's split between two GPUs which have to duplicate the same data. Dual-GPU single-card solutions are halo tier products, most sane people wanting to go multi-GPU will just buy two cards.

 

[PP Mguire]

Ok. But see my question above.

Irrelevant as Crossfire is software.

It's like trying to teach people the difference between Quad-SLI and 4-Way SLI. One is 2 dual GPU cards and one is 4 single GPU cards.

Here's one better. I hand you a 980ti how many video cards do you have in your hand? 1? Ok, I hand you a 295x2 how many video cards do you have in your hand? 1? Single card.

I put 2 980tis in my machine, how many video cards are in my machine? 2. I put the 295x2 in my machine, how many video cards are in it? 1. GPU count doesn't depict card count. 1 PCB, 1 card.
We could also wrap it up as a whole as "single slot solution" as you're putting more than one on a single PCI-E slot if people want to argue the weird 7950GX2 and 295 from Nvidia. Still one PCB with a daughterboard though.

 

[rtwjunkie]

@RCoon, thanks for that in-depth explanation to differentiate the pros and cons.

So, it sounds like in those situations, since SLI and Crossfire are to be enabled, then proper comparisons should always be made to dual single GPU cards running in SLI or Crossfire.

I understand Crossfire is software, but to compare to a single GPU solution is like comparing apples and oranges.

Oh well, it's not important, I was just trying to understand, since I've only ever used two single GPU cards together.

 

[RCoon]

@RCoon, thanks for that in-depth explanation to differentiate the pros and cons.

So, it sou ds like in those situations, since SLI and Crossfire are to be enabled, then proper comparisons should always be made to dual single GPU cards running in SLI or Crossfire.

Comparing 2 x 290Xs versus a 295x2 would be painful to AMD, if a little unfair you have to bare in mind that you're not really comparing the same technology. Single card solutions have to deal with different power and heat constraints, while dual card solutions are a little more free. They're explicitly designed to solve the same problem with two different sets of specifications. Somebody running a 295x2 instead of 2 x 290Xs is probably doing so because of a lack of space, a smaller motherboard, etc. Either that or they're just plain stupid.

I'm running one for fun (which grows old fast), and because I got a sweet deal

 

[PP Mguire]

@RCoon, thanks for that in-depth explanation to differentiate the pros and cons.

So, it sounds like in those situations, since SLI and Crossfire are to be enabled, then proper comparisons should always be made to dual single GPU cards running in SLI or Crossfire.

I understand Crossfire is software, but to compare to a single GPU solution is like comparing apples and oranges.

Oh well, it's not important, I was just trying to understand, since I've only ever used two single GPU cards together.

Not really. Just defining a single card isn't as difficult as some want to make it out to be. These halo cards usually aren't made besides to top charts and be talked about. Very niche, more so than even my Titan X's. So in that sense it doesn't really matter. Just wanted to be clear on the definition of "single card".

Comparing 2 x 290Xs versus a 295x2 would be painful to AMD, if a little unfair you have to bare in mind that you're not really comparing the same technology. Single card solutions have to deal with different power and heat constraints, while dual card solutions are a little more free. They're explicitly designed to solve the same problem with two different sets of specifications. Somebody running a 295x2 instead of 2 x 290Xs is probably doing so because of a lack of space, a smaller motherboard, etc.

I'm running one for fun (which grows old fast), and because I got a sweet deal

Buddy of mine with Fury X's actually ditched his 290x Crossfire setup for a 295x2 to reduce heat and fan noise. The performance difference was actually moot believe it or not. He went Fury X which is how I wound up with the card. The 295x2 was actually a well done card and the AIO made it better. A bad unfair comparison is Titan Black SLI to TItan Z.

 

[the54thvoid]

Not really. Just defining a single card isn't as difficult as some want to make it out to be. These halo cards usually aren't made besides to top charts and be talked about. Very niche, more so than even my Titan X's. So in that sense it doesn't really matter. Just wanted to be clear on the definition of "single card".

Buddy of mine with Fury X's actually ditched his 290x Crossfire setup for a 295x2 to reduce heat and fan noise. The performance difference was actually moot believe it or not. He went Fury X which is how I wound up with the card. The 295x2 was actually a well done card and the AIO made it better. A bad unfair comparison is Titan Black SLI to TItan Z.

There is a semantic error being made here. A dual card versus single card debate on the merits of naming is missing the point. It's more technically classified as a dual GPU card. One PCB with two chips but even more, one pcb with two discrete gpu units linked by a PLX chip. The thing is, a dual gpu single card design is in essence two cards on one PCB, with one cooling solution.

Best example is that a 295x2 has 8GB of total memory but effectively, it actually only has 4GB per chip. Likewise, each chip has it's own VRM's. Really, what we talk about when we say single card is single gpu. We have to clarify that a single card with two gpu's in in fact, a dual gpu single card.

It is two graphical processing units, bastardised together on a single PCB but each handling it's own share of the processing load. What is worse is that without developer and driver support that dual gpu single card will act exactly as a single card.

So semantics again rule. It's disingenuous to call a Titan Z or a 295x2 a 'single card' without reference to it's dual gpu design. In physical outward appearance, yes - it looks like a single card but in engineering it is actually two 'cards' implemented on a single board.

So, in summary, Radeon Pro Duo (like all before it) is a a dual gpu single card design. The single card bit is almost irrelevant except for the physical parameters of it. In engineering, development and driver terms, it has to be treated as two cards (or gpus).

But I know what you mean. lol.

Is a siamese twin one child or one body? Two children, one body.

 

[medi01]

Why not compare to 980 ti sli instead of Titan Z.

Why not compare it to 960?

Comparing top card to top card... how dare they...
/facepalm

 

[newtekie1]

Why not compare it to 960?

Comparing top card to top card... how dare they...
/facepalm

Titan Z isn't nVidia's top end card. It is a generation old dual-GPU card. Basically two 780Ti's.(it was also incredibly stupid and pointless for the consumer)

The Titan X is nVidia's top end card and the 980Ti is the same card with half the memory.

 

[PP Mguire]

There is a semantic error being made here. A dual card versus single card debate on the merits of naming is missing the point. It's more technically classified as a dual GPU card. One PCB with two chips but even more, one pcb with two discrete gpu units linked by a PLX chip. The thing is, a dual gpu single card design is in essence two cards on one PCB, with one cooling solution.

Best example is that a 295x2 has 8GB of total memory but effectively, it actually only has 4GB per chip. Likewise, each chip has it's own VRM's. Really, what we talk about when we say single card is single gpu. We have to clarify that a single card with two gpu's in in fact, a dual gpu single card.

It is two graphical processing units, bastardised together on a single PCB but each handling it's own share of the processing load. What is worse is that without developer and driver support that dual gpu single card will act exactly as a single card.

So semantics again rule. It's disingenuous to call a Titan Z or a 295x2 a 'single card' without reference to it's dual gpu design. In physical outward appearance, yes - it looks like a single card but in engineering it is actually two 'cards' implemented on a single board.

So, in summary, Radeon Pro Duo (like all before it) is a a dual gpu single card design. The single card bit is almost irrelevant except for the physical parameters of it. In engineering, development and driver terms, it has to be treated as two cards (or gpus).

But I know what you mean. lol.

Is a siamese twin one child or one body? Two children, one body.

You basically said what I was saying in your first paragraph. He was trying to say it wasn't a single card solution when it is. Single card, 2 GPUs. I was only correcting that.

 

[cwilczewski]

Titan Z = generation old

Radeon Pro Duo = current/next gen.

This is a great apples to grape fruit comparison.

One would only hope that it is faster.

That's not actually a fair thing to say, Maxwell and Fiji aren't that different. Fiji is the same as the previous architecture but simply had a change with power consumption and use of GDDR5. AMD hasn't really had a "new" gpu architecture in several years, with the next one slated to release in 2017.

 

[Prima.Vera]

A single card solution is essentially the same thing as a dual card solution, only incredibly crippled in every shape and form. Power delivery is almost always worse than having two cards, and the methods for removing heat from dual-GPU single-card solutions are always so horribly borderline that your clocks always get throttled in everyday situations. The only reason my 295x2 doesn't get throttled is because of a full cover waterblock and powerful pump. They're also an ass to deal with when it comes to PCI-E lanes, making 16 x PCI-E 2.0 (8 x PCI-E 3.0) kinda necessary.

You still have to pray for SLI/Crossfire profiles, and yes, you need to make sure Crossfire is enabled in the first place. Dual GPU single-card solutions may have the chips on the same board, but they still rely on those technologies. It also allows companies like AMD and NVidia to claim that their card has 8GB of VRAM, when in actual fact they only have 4GB usable. Sure, there's 8GB sat on the board, but it's split between two GPUs which have to duplicate the same data. Dual-GPU single-card solutions are halo tier products, most sane people wanting to go multi-GPU will just buy two cards.

Is interesting how the guys from 3DFX were able to use 2 or for 4 GPUs on their cards but on the same time time the whole memory was share, not divided between GPUs...

 

[newtekie1]

That's not actually a fair thing to say, Maxwell and Fiji aren't that different. Fiji is the same as the previous architecture but simply had a change with power consumption and use of GDDR5. AMD hasn't really had a "new" gpu architecture in several years, with the next one slated to release in 2017.

And the Titan Z isn't Maxwell, it's Kepler.

 

[Relayer]

Titan Z = generation old

Radeon Pro Duo = current/next gen.

This is a great apples to grape fruit comparison.

One would only hope that it is faster.

Quit defending the Z. The 295x2 beat it for 1/2 the price.

 

[Renald]

And most of all, points curve is not linear, so 51% more points doesn't means 51% faster.
It's an exponentiation curve which is made to distinguish high end card.

If you re-bench all old cards (3 or 4 generations backwards), you won't be able to tell the difference between a 560 Ti and a ... (Edit : 460 not a Ti ) which should be 20% faster I think.


And as said earlier by someone, no matter what score you do in raw computing, real test is the only one that matters.
We don't care if we have a 100% higher score in 3DMark but 3fps more on our game... It's just a non-sense to take this score alone. You need to cross benches over different situations to assume that "AMD Radeon Pro Duo up to 51 percent Faster than GeForce GTX TITAN Z".

PS : and I'm just saying this not taking any side. For now it's just assumption, not reality. You need to add "Could be up to 51 percent faster" to your title to be accurate.

 

[rtwjunkie]

you won't be able to tell the difference between a 560 Ti and a ... 460 Ti

I can tell the difference, because if you compare those cards, the 560Ti will be benching against.....a non-existent card. I don't recall ever hearing about or seeing a 460Ti.

 

[PP Mguire]

I don't think there is a doubt in anybody's mind that Dual Fiji will be faster than Titan Z. I take dumps faster than Titan Z

 

[Renald]

I can tell the difference, because if you compare those cards, the 560Ti will be benching against.....a non-existent card. I don't recall ever hearing about or seeing a 460Ti.

You definitively right !! I didn't want to use too recent cards, but my point sticks and I think you'll agree
Anyway , take a 460, and you'll have the same "problem".

 

[arbiter]

Titan Z isn't nVidia's top end card. It is a generation old dual-GPU card. Basically two 780Ti's.(it was also incredibly stupid and pointless for the consumer)
The Titan X is nVidia's top end card and the 980Ti is the same card with half the memory.

Sadly it is AMD we are talking about, Their use of questionable benchmarks has been a source of problem.

Quit defending the Z. The 295x2 beat it for 1/2 the price.

I doubt he was defending price of the Z more so the problem of using 780ti which when you compare to current maxwell cards is a little slower then a gtx980. So that 51% besides the questions of settings they used to get that "51%" claim but a gpu that is a good 20-30% slower then 980ti alone.

 

[newtekie1]

Quit defending the Z. The 295x2 beat it for 1/2 the price.

No one is defending the Titan Z, we are stating that comparing this new card to the Titan Z is stupid. The Titan Z has been discontinued for a while now.

So basically what AMD is saying is "Our new card is 51% faster than a card that is so old it has been discontinued!"

 

[the54thvoid]

So basically what AMD is saying is "Our new card is 51% faster than a card that is so old it has been discontinued!"

Which in card genealogy is the same as Nvidia saying:

GTX 980ti, 50% faster than R9 290X.

 

[PP Mguire]

Which in card genealogy is the same as Nvidia saying:

GTX 980ti, 50% faster than R9 290X.

When 980ti came out they didn't have Fury out, but I know where you were going with it.

 

[arbiter]

Which in card genealogy is the same as Nvidia saying:
GTX 980ti, 50% faster than R9 290X.

When 980ti came out they didn't have Fury out, but I know where you were going with it.

Its one thing when card was released it was current card, its another thing when the card was released its previous gen and pretty much EOL'ed at that point. But in reality a 290x is still being made now as its just been renamed as the 390x.

How offen do you see Nvidia comparing their card to AMD cards? I can't remember any time recently that they have done that. Its always been vs their last gen card. Its always reviews that do the comparing as it should be.