Gtx 1080

[Evildeffy]

Where are you all getting your information from? The chips are identical in layout, with the 104 obviously being a smaller variant.

Geforce 1080:

http://images.anandtech.com/doci/103...gram_FINAL.png

GP100:

http://images.anandtech.com/doci/102..._diagram-1.png

Look at those differences! Why you can count all... 0 of them.

But what's this, can Pascal just be a rebranded Maxwell? Look at this block diagram, just a few minor differences:

http://images.anandtech.com/doci/905...gram_FINAL.png

But what's this, can Maxwell just be a rebranded Kepler? Look at this block diagram, just a few minor differences:

http://images.anandtech.com/doci/569...gram_FINAL.png

But what's this, can Kepler just be a rebranded Fermi? Look at this block diagram, just a few minor differences:

http://images.anandtech.com/reviews/...F100/GF100.png

Going by block diagram logic, AMD hasn't had a new architecture in almost 5 years.

The devil is in the details.
If you make a comparison between the GP100 and GP104 there are more differences and construction than the differences between GM200/GM204 and GP104.

There are differences, you just have to pay attention to where they are which make key differences in architectures.

AMD has had architecture cycle differences just as much as nVidia has.
The HD7900 series was a new architecture that went into R9 280 series.
The R9 290 series was a new architecture that went into R9 390 series.
The R9 285 series was a new architecture that went into R9 380 series.
The R9 Fury (X) and R9 Nano were new architectures and they will end where they started.

Just as they are introduced this year as Polaris 10 and 11 as well as Vega 10 and 11.

That's 4 architectures, based on the same line but actual separate architectures.
Just like Fermi to Kepler to Maxwell to Pascal, the difference being that the jumps between all those architectures were larger than GM200/GM204 to GP104.

It's very hard creating a new architecture when you've been pushing that to the limits for 5 years so there's 0 point in creating new architectures.

[Zenny]

The devil is in the details.
If you make a comparison between the GP100 and GP104 there are more differences and construction than the differences between GM200/GM204 and GP104.

There are differences, you just have to pay attention to where they are which make key differences in architectures.

Not to act like a idiot, but can you point out those key differences? The block diagrams appear identical. (Other then size)

Edit: Another difference is the lack of Double Precision vs the GP100, but that won't effect games.

[Evildeffy]

Not to act like a idiot, but can you point out those key differences? The block diagrams appear identical. (Other then size)

Edit: Another difference is the lack of Double Precision vs the GP100, but that won't effect games.

Very easy to spot the differences:
01. The GPC has a Raster Engine on the Gp104, none on the GP100
02. On the GP104 every TPC has a Polymorph engine 1 larger SM cluster, the GP100 does not and has 2 separate smaller SM clusters along with an instruction cache for each cluster.
03. The GP100 has Double Precision cores which the GP104 does not have
04. In the GP104 next to the L2 cache you have I believe the memory blocks which are GDDR5X/GDDR5 (GDDR5 memory controller is compatible with GDDR5X), the GP100 has an entirely different memory controller and location for HBM2 memory on the sides rather than the pathways spreading to each cluster from the centre of the die, this caused the infamous GTX 970 memory controversy.
05. The GP104 has, due to above memory distinction, 256 bit memory bus total divided 8 controllers giving them 32 bit control each where GP100 has 4096 bit memory bus which consists of 8 memory controllers each providing 512 bit vs. the 32 bits of GP104.
06. There are 6 GPCs for the GP100 vs. the 4 for GP104
07. The GP104 does not have a high speed hub and NVLink for high speed interlinking for HPC designs.

Those are just superficial shots of what you showed alone, there's plenty more details but no... they are not identical.

[Zenny]

Very easy to spot the differences:
01. The GPC has a Raster Engine on the Gp104, none on the GP100
02. On the GP104 every TPC has a Polymorph engine 1 larger SM cluster, the GP100 does not and has 2 separate smaller SM clusters along with an instruction cache for each cluster.
03. The GP100 has Double Precision cores which the GP104 does not have
04. In the GP104 next to the L2 cache you have I believe the memory blocks which are GDDR5X/GDDR5 (GDDR5 memory controller is compatible with GDDR5X), the GP100 has an entirely different memory controller and location for HBM2 memory on the sides rather than the pathways spreading to each cluster from the centre of the die, this caused the infamous GTX 970 memory controversy.
05. The GP104 has, due to above memory distinction, 256 bit memory bus total divided 8 controllers giving them 32 bit control each where GP100 has 4096 bit memory bus which consists of 8 memory controllers each providing 512 bit vs. the 32 bits of GP104.
06. There are 6 GPCs for the GP100 vs. the 4 for GP104
07. The GP104 does not have a high speed hub and NVLink for high speed interlinking for HPC designs.

Those are just superficial shots of what you showed alone, there's plenty more details but no... they are not identical.

1. The Geforce 1080 has a raster engine, it's right at the top of the block diagram. Not sure why the GP100 doesn't, maybe because it's not the gaming orientated chip? I'm not exactly seeing how this is supposed to be a negative.
2. The GP100 has the polymorph engine as well? Each SM looks the same, other then the missing DP units.
3. Zero effect on gaming
4. Different memory architecture doesn't mean a different graphics core architecture. HBM2 wouldn't have made the graphics architecture magically faster. (Obviously would have more memory throughput though).
5. Once again, memory, nothing to do with with core itself.
6. Yes, the extra 2 are for double precision compute. Which is left out on the consumer level chip has it has zero effect on what the card is used for.
7. Once again, what does this have to do with gaming performance?

SM units are the same, other then the lack of DP units:

http://core0.staticworld.net/images/...283-medium.png
http://cdn.wccftech.com/wp-content/u...SM_diagram.png

[Evildeffy]

1. The Geforce 1080 has a raster engine, it's right at the top of the block diagram. Not sure why the GP100 doesn't, maybe because it's not the gaming orientated chip? I'm not exactly seeing how this is supposed to be a negative.

That's the GigaThread Engine, not the raster engine embedded in the GPC, this is a definite negative as the GP100 has no use for it as it does not drive monitors, even if it did it would still be different to GP104.

2. The GP100 has the polymorph engine as well? Each SM looks the same, other then the missing DP units.

Nope, no polymorph engine anywhere, it's quite clear as it's a yellow block in each SM cluster.

3. Zero effect on gaming

Irrelevant to gaming performance, very relevant to architectural differences which was the point.

4. Different memory architecture doesn't mean a different graphics core architecture. HBM2 wouldn't have made the graphics architecture magically faster. (Obviously would have more memory throughput though).

Incorrect as it does, including positioning of said memory reducing or increasing latency where a GDDR5(X) controller is also a different piece of hardware to an HBM2 controller, entirely different design.

5. Once again, memory, nothing to do with with core itself.

Wrong again, see above. You cannot simply "convert" from 1 design to another as it would require radical restructuring.

6. Yes, the extra 2 are for double precision compute. Which is left out on the consumer level chip has it has zero effect on what the card is used for.

Wrong again, we are talking about actual physically larger GPC clusters vs. a row of DP shaders, if you were to convert those DP shaders to the same SP ones it would be a monster of a graphics card, we are talking an added 50% of units that need to be connected and controlled as well as communicate with HBM2 memory, this is actually a massive difference.

7. Once again, what does this have to do with gaming performance?

Again... the point was "Architectural differences" which this is a rather large part of, it's irrelevant if it's good for gaming or not, they are by far not identical as the physical size and make-up alone is different along with the design as intercommunication. That would be exactly like saying a VW Golf is exactly the same in every single way to a VW Passat, just bigger, so they are identical.
There's no contest here.

SM units are the same, other then the lack of DP units:

http://core0.staticworld.net/images/...283-medium.png
http://cdn.wccftech.com/wp-content/u...SM_diagram.png

And this doesn't give you the idea that the architecture is radically different to one and another? Ask anyone who works in chip design if they are the same, if you were to cut it out it would be identical.
You yourself requested the architectural differences, I gave them to you, if you want gaming performance differences than you should've asked the difference in performance of card A and card B but that would entirely invalidate your first point of:

Where are you all getting your information from? The chips are identical in layout, with the 104 obviously being a smaller variant.

Which means that if it were a smaller variant the GP104 would have DP shaders as well as a high speed hub and NVLink option along with HBM2.
Along with missing raster engines, polymorph engines, have 2 separate SM clusters PER TPC instead of 1 larger cluster etc. etc.

Do you know the complexity added from adding another set of GPC to an existing engine? It's not something you can do in a short time.
It's a rather complex and VERY expensive procedure.

You simply want to define it as gaming performance differences from 1 series to the next, this does NOT define architecture in the least.
The GP100 is a far more radical difference to GM200 than a GP104 to a GM204.

[Master Guns]

What the fuck are you all even arguing about and who the fuck cares?

[Evildeffy]

What the fuck are you all even arguing about and who the fuck cares?

Reading it would tell you what it is and if you don't like it you're welcome to ignore it.
Your comment does not add anything to the thread other than flaming so either add something to the discussion worth a damn or be silent and ignore it.

As for your "who the fuck cares" comment... apparently quite a few do as the subject is being broached.

[Zenny]

What the fuck are you all even arguing about and who the fuck cares?

Well, we do obviously, otherwise we wouldn't have been having the argument in the first place. If you don't care about it why bother with a meaningless comment? Why don't you just ignore the posts and/or thread?

[Frozenbeef]

What about the below? :s Not worried about the price and intend to buy a new motherboard/ ram anyway as my current motherboard is locked to 1 slot >.<


https://www.scan.co.uk/products/giga...n-in-stock)plu

[Evildeffy]

What about the below? :s Not worried about the price and intend to buy a new motherboard/ ram anyway as my current motherboard is locked to 1 slot >.<


https://www.scan.co.uk/products/giga...n-in-stock)plu

It is slightly more expensive than if you would buy em seperately ... but having said that they are in stock with Scan. (30 GBP give or take)
So if you want it with any meaningful speed... that's not a bad combo.

All components are good.

[Remilia]

P100 has 64CC per SM compared to Maxwell /GP104 which is 128. That in itself tells you that they're different.

[Artorius]

What the fuck are you all even arguing about and who the fuck cares?

If you don't care about hardware then just don't read the posts, they surely aren't intended for normal customers.

[MonsieuRoberts]

This is annoying...

Exactly what/who are you replying to? This is an 88 page thread... You need to make it more obvious what you're responding to. XD

Whoop, I didn't quote!

That was in reference to Asus & MSI shipping their review sample cards in OC mode.

[Evildeffy]

Whoop, I didn't quote!

That was in reference to Asus & MSI shipping their review sample cards in OC mode.

MSI have actually responded by releasing the reviewer's BIOS up for download which can be flashed on the Gaming X series 1080 and 1070.

Available here: http://wccftech.com/msi-issues-state...u-clock-speed/

[Bigvizz]

If you don't care about hardware then just don't read the posts, they surely aren't intended for normal customers.

You're right and thus why the market shares won't change in AMD favor anytime soon. Because consumers don't pay attention to the details. /shrug

[mrgreenthump]

Because consumers don't pay attention to the details. /shrug

Most consumers also don't pay >$400 for a graphics card. Nvidia currently has nothing at sub $300, which just happens to be the biggest market by far. So it's quite likely the RX 480 will sell like mad till Nvidia can put out a competitor to it.

Now one can only speculate when will the 1060 come, but it ain't even announced yet.. So yeah.

[Bigvizz]

Most consumers also don't pay >$400 for a graphics card. Nvidia currently has nothing at sub $300, which just happens to be the biggest market by far. So it's quite likely the RX 480 will sell like mad till Nvidia can put out a competitor to it.

Now one can only speculate when will the 1060 come, but it ain't even announced yet.. So yeah.

/chuckle, we'll see.

[Vash The Stampede]

/chuckle, we'll see.

$200 GTX 980. I'm calling it.

[Bigvizz]

$200 GTX 980. I'm calling it.

Wouldn't be surprised.

[Deleted]

$200 GTX 980. I'm calling it.

but didn't they say a while ago they stopped making those? i imagine that factory space has been reassigned and they can't just switch back overnight.