AMD Fury X - Is it an extra large (combined) Tonga chip?

J

[jF]

Weaksauce
Joined
Feb 3, 2002
Messages
75
Hi all, long time lurker here from the UK ;) :D

Something struck me with the specs of the Fury X, is it a double full-fat Tonga chip with HBM bolted on? The fact that it has the existing GCN 1.2 architecture instead of AMD creating a new one for the new series (which I find a bit odd tbh). Plus the HDMI 1.4a limitation along with the rumoured DX 12_0 support (instead of the absolute fullest 12_1) makes me think this.

What do you guys think? Tonga was rumoured to be the original candidate chip for HBM but that was changed before release...

Cost cutting? AMD certainly don't seem to have the resources that they used to
 
The Tonga die is 359mm2 and Fiji looks to be ~560mm2. They're both 28nm.

I imagine some Tonga tech is under Fiji's hood (like that spiffy color compression stuff) but I suspect it's not as cut and dry as "(Tonga x 2)"
 
I've suspected this as well. They haven't really touted anything special about the Fury architecture itself, and instead have pounded HBM into us.

The layout may be different (more SPs, etc) but the tech is pretty much Tonga + HBM.
 
It isn't a double Tonga, it doesn't double all of Tonga's resources...
You can't change an ASIC design "right before release."

AMD did the same thing with Bonaire - Hawaii as they did with Tonga - Fiji.
It just makes more sense to KISS. You don't need to waste resources to push a new GCN iteration every time you release a new GPU.
 
It's disappointing if true, but it would explain the need for a water cooler and the lack of HDMI 2.0
 
You know who has to troll every AMD thread, cause he doesn't get paid if he misses one;)
 
LordEC911 said:
It isn't a double Tonga, it doesn't double all of Tonga's resources...
You can't change an ASIC design "right before release."

AMD did the same thing with Bonaire - Hawaii as they did with Tonga - Fiji.
It just makes more sense to KISS. You don't need to waste resources to push a new GCN iteration every time you release a new GPU.
Click to expand...

I guess I'm just a little disappointed that it just seems to be 'more' of what we already have with HBM bolted on!! I guess we'll have to wait until the 14/16nm shrinks to see some truly innovative stuff from both sides... :(
 
[jF];1041676236 said:
I guess I'm just a little disappointed that it just seems to be 'more' of what we already have with HBM bolted on!! I guess we'll have to wait until the 14/16nm shrinks to see some truly innovative stuff from both sides... :(
Click to expand...

So you are disappointed that AMD is matching a DP-light GM200 with a smaller GPU that offers just as much performance, if not more, in the same price range?

From an engineering standpoint, they increased performance 50-60% while only increasing die size, over Hawaii, ~36% and keeping the average TDP roughly the same.
 
LordEC911 said:
So you are disappointed that AMD is matching a DP-light GM200 with a smaller GPU that offers just as much performance, if not more, in the same price range?

From an engineering standpoint, they increased performance 50-60% while only increasing die size, over Hawaii, ~36% and keeping the average TDP roughly the same.
Click to expand...

From the figures seen so far it seems to only just eclipse a 980Ti, which they've done by only using a next-gem memory interface! Yes, I'm a little disappointed in AMD for being so 'poor' as it were
 
I look forward to a review that uses a heavily overclocked watercooled 980ti. Apples to apples.
 
PRIME1 said:
I look forward to a review that uses a heavily overclocked watercooled 980ti. Apples to apples.
Click to expand...
I think a clock for clock comparison like the 2/3-90x review would be neat. Find the highest common no throttle/no boost clock rate between nV and AMD, lock to that clock and test.
 
Heavily watercooled 980ti...LOL. That's shifting goal posts, considering that would cost at least $150 more.
 
[jF];1041676604 said:
From the figures seen so far it seems to only just eclipse a 980Ti, which they've done by only using a next-gem memory interface! Yes, I'm a little disappointed in AMD for being so 'poor' as it were
Click to expand...

I'm guessing 750ti and Maxwell was ok for you, right?
 
Remon said:
I'm guessing 750ti and Maxwell was ok for you, right?
Click to expand...

Not really, both have deficiencies!! I was just hoping for more than a year old architecture scaled up (hence the GCN 1.2/HDMI 1.4a/DP 1.2a. 2014 standards!)

Tonga was heavily rumoured to be the first HBM chip for a while, guess we know now why that was skipped and moved along :)
 
[jF];1041677637 said:
Not really, both have deficiencies!! I was just hoping for more than a year old architecture scaled up (hence the GCN 1.2/HDMI 1.4a/DP 1.2a. 2014 standards!)

Tonga was heavily rumoured to be the first HBM chip for a while, guess we know now why that was skipped and moved along :)
Click to expand...

The 750ti was the introductory gpu to Maxwell. Criticizing Fiji about being a bigger Tonga chip is hypocritical, unless you criticized 980 for being a bigger 750ti.

Also, these weren't the first times that both of these companies did this sort of thing.
 
I think most of us with common sense would enjoy reading benchmarks that compare the same price point to equal price point just as are favorite [H] website always does. In the end what, we can get for a certain price matters a lot, since most of us don't have disposable income.
 
Nolan7689 said:
Mmmhmmm.

Please enlighten all of us how that is Apples to Apples?
Click to expand...

It isn't, but people with Agenda's will do anything to keep the product they want in a proper light.

I mean he wants to compare a non-reference massively overclocked video card to a non overclock reference video card.

More like Apples to Kumquats
 
PRIME1 said:
I look forward to a review that uses a heavily overclocked watercooled 980ti. Apples to apples.
Click to expand...

I'll cross party lines and disagree here. Comparison should be whatever standard equipment you get out of the box at the same $650 price point. Because throwing a CLC on a 980 Ti shifts the price point. If Nvidia decides to start throwing CLC's onto ref cards as standard equipment and keeps price at $650 then that'd be a different story.

That said, there will be no shortage of benches featuring both cards at their max overclocks under any method of cooling. It'll be an orgy of benchmarks and stats next week like you've never seen. Blood will be running in the streets.

Now get out of my office.
 
Last edited:
LordEC911 said:
Because he thinks it is. But then again he is in his own deluded reality.
Click to expand...

DASHlT said:
It isn't, but people with Agenda's will do anything to keep the product they want in a proper light.

I mean he wants to compare a non-reference massively overclocked video card to a non overclock reference video card.

More like Apples to Kumquats
Click to expand...

Oh I know I just wanted to see if he would reply, and wondered about his 'logic'

DPI said:
I'll cross party lines and disagree here. Comparison should be whatever standard equipment you get out of the box at the same $650 price point. Because throwing a CLC on a 980 Ti shifts the price point. If Nvidia decides to start throwing CLC's onto ref cards as standard equipment and keeps price at $650 then that'd be a different story.

That said, there will be no shortage of benches featuring both cards at their max overclocks under any method of cooling. It'll be an orgy of benchmarks and stats next week like you've never seen. Blood will be running in the streets.

Now get out of my office.
Click to expand...

It will also be interesting when we see see nonreference factory OCd fiji hitting the benches eventually.
 
DASHlT said:
It isn't, but people with Agenda's will do anything to keep the product they want in a proper light.
Click to expand...

People should view your posts and see how ironic that statement is.

Several companies are making retail water cooled 980ti's

It's best to compare retail vs retail, than using one of AMDs golden sample cards.
 
People should view your posts and see how ironic that statement is.
roflmao u need a mirror dude, but again i got a good chuckle from your post so no ignore list for u
 
spaceace420 said:
People should view your posts and see how ironic that statement is.
roflmao u need a mirror dude, but again i got a good chuckle from your post so no ignore list for u
Click to expand...

I fully admit that I am not fan of AMD and the way the conduct business.

Now feel free to look at his posts and decide if he has an agenda or not.

Saying that you should not compare retail cards, especially after AMD got caught sending review sites "golden sample" cards is a prime example of trying to show a product in a "better light' Just like AMD not wanting to give cards to sites that won't give them a review in their favor.

This is the kind of shady behavior that no consumer should support.
 
PRIME1 said:
I fully admit that I am not fan of AMD and the way the conduct business.

Now feel free to look at his posts and decide if he has an agenda or not.

Saying that you should not compare retail cards, especially after AMD got caught sending review sites "golden sample" cards is a prime example of trying to show a product in a "better light' Just like AMD not wanting to give cards to sites that won't give them a review in their favor.

This is the kind of shady behavior that no consumer should support.
Click to expand...

You said sites need to review a heavily oc'ed watercooled GTX980Ti vs a Fury X....

Like threatening review sites if they don't follow the reviewer's guide?
Like sending reviewers cards that boost 100-200mhz faster than retail?
 
Well.. they could do the EVGA AIO 980ti vs. Fury X OC vs. OC when it hits. See who has the performance crown. I have a feeling nVidia will, but we don't know yet. Then if the 980ti is 10% faster and $100 more it's justified by the perf. increase.

[H] has done OC vs OC in the past. Hopefully they will again. Even if it's something like a Strix 980ti vs. Fury X it'd be interesting. I've never seen [H] BIOs mod or hardware mod though for an official review. I understand why... that's an infinite rabbit hole.

Like this beast right here: http://www.evga.com/Products/Product.aspx?pn=06G-P4-1996-KR

Right now I'm tearing down my chilled system to possibly redo it or repurpose it... if I was not doing a chilled system I'm all about the AIOs for either team. It just makes sense.
 
PRIME1 said:
I fully admit that I am not fan of AMD and the way the conduct business.

Now feel free to look at his posts and decide if he has an agenda or not.

Saying that you should not compare retail cards, especially after AMD got caught sending review sites "golden sample" cards is a prime example of trying to show a product in a "better light' Just like AMD not wanting to give cards to sites that won't give them a review in their favor.

This is the kind of shady behavior that no consumer should support.
Click to expand...

Tell you what, how about tell us of this business practice/s that you speak of and how either Intel or Nvidia have never done it or worse.

And where is the proof of golden samples? Last we saw it was fan profile differences between cards. Most fixed after driver updates. But still would like to see proof of this, conclusively/finalized, if you have it.
 
LordEC911 said:
So you are disappointed that AMD is matching a DP-light GM200 with a smaller GPU that offers just as much performance, if not more, in the same price range?

From an engineering standpoint, they increased performance 50-60% while only increasing die size, over Hawaii, ~36% and keeping the average TDP roughly the same.
Click to expand...


Can't compare die size vs performance, when one has a memory controller and one doesn't.
 
LordEC911 said:
They both have memory controllers....
Click to expand...

the complexity of the memory controller and die size consumption is very different between the two, the memory controller of HBM uses very little die size and is no where near as complex, for the most part lets consider around 25% of the die for GDDR5 for a memory controller vs, something like 5% of the entirety of the all the silicon (with transposer) or possibly even less for an HBM memory controller, I'm saying very rough %'s of course, actually the memory controller is in the HBM stack, I thought it was part of the interposer, ok, so its not actually part of the GPU at all so no the GPU itself has no memory controller..
 
Last edited:
PRIME1 said:
Gibbo is one of AMD's biggest fans and he's recommending people buy 980ti's after testing a Fury card.

http://forums.overclockers.co.uk/showpost.php?p=28200197&postcount=40

Oh well. I was hoping for more of a showdown.
Click to expand...

he's recommending buying a temporarily discounted 980Ti @ 500 pound when the next cheapest 980Ti is @ 540 pound and thinks vendors are going to sell Fury @ 600 pound when it first gets released even though MSRP is supposed to be 549 pound

GTX 980Ti obviously becomes a better buy if there's a 100 pound difference between Fury & GTX 980Ti pricing
 
razor1 said:
the complexity of the memory controller and die size consumption is very different between the two, the memory controller of HBM uses very little die size and is no where near as complex, for the most part lets consider around 25% of the die for GDDR5 for a memory controller vs, something like 5% of the entirety of the all the silicon (with transposer) or possibly even less for an HBM memory controller, I'm saying very rough %'s of course, actually the memory controller is in the HBM stack, I thought it was part of the interposer, ok, so its not actually part of the GPU at all so no the GPU itself has no memory controller..
Click to expand...

So you found a new stick to play with and won't let go?
Welcome to 12 months ago.

The PHY still requires MC like logic to accept the signaling. Yes, it is simplified but not completely removed.
 
LordEC911 said:
So you found a new stick to play with and won't let go?
Welcome to 12 months ago.

The PHY still requires MC like logic to accept the signaling. Yes, it is simplified but not completely removed.
Click to expand...


In the neighborhood of what like 2% of die size in the GPU? Hmm yeah I can see where you are trying to go with that....

No its not a stick to play with, you try to compare die sizes with two different GPU's and totally ignore why the differences are there, its called selective reasoning. and that never works.


If nV went HBM this round their chip would be much smaller then Fury, and still have the same performance, if I made that argument, isn't that incorrect? Yes it is.
 
[jF];1041676236 said:
I guess I'm just a little disappointed that it just seems to be 'more' of what we already have with HBM bolted on!! I guess we'll have to wait until the 14/16nm shrinks to see some truly innovative stuff from both sides... :(
Click to expand...

when you're using something new like HBM it's better to stick with what you already know and is proven to work to test something like HBM on, fully expect we'll see a new GCN update next year with HBM 2.0 on the 16nm process.
 
razor1 said:
In the neighborhood of what like 2% of die size in the GPU? Hmm yeah I can see where you are trying to go with that....

No its not a stick to play with, you try to compare die sizes with two different GPU's and totally ignore why the differences are there, its called selective reasoning. and that never works.

If nV went HBM this round their chip would be much smaller then Fury, and still have the same performance, if I made that argument, isn't that incorrect? Yes it is.
Click to expand...

I'm not going to speculate on actual sizes since we have nothing to base it on.

It is a stick. You have never mentioned it before, even though it has been known about and discussed before. Now that you found out about it, it is this big deal... I'm not ignoring anything.
 
You must log in or register to reply here.
Back
Top