AMD "Tonga" Silicon Features 384-bit Wide Memory Interface

[btarunr]

In what could explain the rather large die-size and transistor-count of AMD's "Tonga" silicon, compared to "Tahiti," it turns out that the silicon features a 384-bit wide GDDR5 memory interface, and not the previously thought of 256-bit wide one. The die is placed on a package with pins for just 256-bit, on the Radeon R9 285, but it can be placed on a bigger package, with more pins, to wire out the full width of the memory bus, in future SKUs. This isn't the first time AMD has done something like this. Its "Tahiti LE" chip was essentially a "Tahiti" die placed on a smaller package with pins for just a 256-bit wide memory bus, on the oddball Radeon HD 7870 XT.

What this means is that AMD's next performance-segment graphics card based on the "Tonga" silicon, could feature 50% more memory bandwidth than the R9 285. The stream processor count is still 2,048, but these are more advanced Graphics CoreNext 1.2 stream processors, compared to first-generation ones on "Tahiti," offering more performance per Watt. The TMU count remains 128, although there's no clarity on the ROP count. Estimates are split between 32 and 48. The R9 285 has 32, and so does "Tahiti."



View at TechPowerUp Main Site

 

[dj-electric]

Let's just take a second to remember how non-efficient the R9 285 is.
It's sad to say it, but my expectations from an AMD GPU have never been lower.

 

[Naito]

Let's just take a second to remember how non-efficient the R9 285 is.
It's sad to say it, but my expectations from an AMD GPU have never been lower.

Well, it's not much worse than Hawaii and still better than original Tahiti (XT2 atleast), but I guess you're right... it's sad that the latest generation is not much better off than the two prior ones.

 

[arbiter]

At risk of flakk, wonder what kinda bump would be needed on gpu side to get it to match a gtx970 for example, and what kinda power it would draw.

 

[Serpent of Darkness]

Let's just take a second to remember how non-efficient the R9 285 is.
It's sad to say it, but my expectations from an AMD GPU have never been lower.

Well you're argument about AMD Graphic Cards being non-efficient is sound, AMD's new R9-390x is going to be the first Graphic Card on 20nm. In addition, it will come with stacked Ram. So while efficiency and lower power draws are being thrown out the window, AMD intends to offer better gains in other areas to it's consumers: Features found on Volta, or features that should have been implemented on Maxwell GTX 980 instead of GTX Titan-Maxwell in Feb 2015. I could care less about the power draw, but the AMD nuclear reactor will keep me warm at night during the Christmas Season while playing MMOs and rendering stuff on Maya/3DMax. That's pending 3rd party benchmarks of course...

 

[esrever]

Let's just take a second to remember how non-efficient the R9 285 is.
It's sad to say it, but my expectations from an AMD GPU have never been lower.

It is more efficient than maxwell at double precision compute most likely, AMD has been really pushing the Firepro lines.

 

[ErthWen]

It is more efficient than maxwell at double precision compute most likely, AMD has been really pushing the Firepro lines.

Ahah, good joke!!!

But you need to try very much hard to defend some piece of AMD's honor:

This is the "weak" dp performance of Maxwell (1/32 of SP performance, a weak one by definition):

Annnnd, THIS is the very much "stronger" performance of the heavenly provide and perfect R9 285:

WTF????

OHHH yeah!

Very much solid performance in DP, Tongo is a weaker chip than Tahiti for GPGPU (no better than R9 280, 3 years later), and a much weaker one with DP than the Maxwell chips. Its a pile of bullshit.

You are speaking nonsense searching some goodness with Tongo.

 

[esrever]

Ahah, good joke!!!

But you need to try very much hard to defend some piece of AMD's honor:

This is the "weak" dp performance of Maxwell (1/32 of SP performance, a weak one by definition):


Annnnd, THIS is the very much "stronger" performance of the heavenly provide and perfect R9 285:



OHHH yeah!

Very much solid performance in DP, Tongo is a weaker chip than Tahiti for GPGPU (no better than R9 280, 3 years later), and a much weaker one with DP than the Maxwell chips. Its a pile of bullshit.

You are speaking nonsense searching some goodness with Tongo.

Folding@home has always been better optimized on nvidia. Anyways, those numbers are cut down DP numbers from the gaming lines. If you look at the firepro lines, it will look a lot different.

 

[dj-electric]

it's not much worse than a 3 year-old GPU

That's exactly what i'm worried about. Worse than a 3 year-old GPU.

About next lithography - TSMC does not want 20nm, as they are already preparing 16nm nodes. We will see maxwell on these nodes, as well as whatever AMD is cooking. They better cook good, as good as in 50% performance increase per watt over CGN 1.0 at the same lithography.

Or, they cannot, and they will lose GPUs just like they lost the R9 290 series last week, and will probably lose R9 280X and R9 285 in a few weeks to the medium-sized maxwells

 

[HumanSmoke]

It is more efficient than maxwell at double precision compute most likely, AMD has been really pushing the Firepro lines.

A couple of points...
As ErthWen in the post above me pointed out, Tahiti's 1:4 FP64 rate has been cut to 1:16 with Radeon Tonga. IF you assume that Tonga natively (as opposed to AMD's limitation) supports 1:2 rate, then the Tonga based W7100 pretty much just competes with the existing W8100, and the full die will compete with the W9100.
Double precision isn't obligatory for professional cards. FP64 workloads tend to be more the province of HPC than workstation, and AMD's presence in the HPC market isn't great where performance-per-watt becomes an important metric. Even in the WS/HPC single precision arena, AMD face an uphill struggle when the GTX 980 has already demonstrated itself to be a stable 6+ TFlop performer.
Thirdly, while professional boards usually make up the cream of GPU revenue, AMD is basically offering them at very much reduced cost to their largest high profile customer- Apple, so I doubt there is much relief as there could be in that sector.

 

[esrever]

A couple of points...
As ErthWen in the post above me pointed out, Tahiti's 1:4 FP64 rate has been cut to 1:16 with Radeon Tonga. IF you assume that Tonga natively (as opposed to AMD's limitation) supports 1:2 rate, then the Tonga based W7100 pretty much just competes with the existing W8100, and the full die will compete with the W9100.
Double precision isn't obligatory for professional cards. FP64 workloads tend to be more the province of HPC than workstation, and AMD's presence in the HPC market isn't great where performance-per-watt becomes an important metric. Even in the WS/HPC single precision arena, AMD face an uphill struggle when the GTX 980 has already demonstrated itself to be a stable 6+ TFlop performer.
Thirdly, while professional boards usually make up the cream of GPU revenue, AMD is basically offering them at very much reduced cost to their largest high profile customer- Apple, so I doubt there is much relief as there could be in that sector.

Hawaii was also cut down in the 290x and 290 for DP, same is probably true for tonga. And how would a W7100 compete with W8100 when the 285 doesn't come close to the 290 in performance? They are the same gap when both are 1/2 DP.

If you look at the HPC arena, the W9100 has almost a 60% lead in performance/watt vs kelper in DP due to the 1/2 rate DP. AMD is aggressively pursuing that space just from their design perspective. Going from 1/4 DP in Tahiti to 1/2 DP in Hawaii shows that is what they are trying to target. As you see the trend, the more DP performance they try to squeeze out, the less efficient the cards become, same from nvidia's side as well.

 

[HumanSmoke]

Hawaii was also cut down in the 290x and 290 for DP, same is probably true for tonga. And how would a W7100 compete with W8100 when the 285 doesn't come close to the 290 in performance? They are the same gap when both are 1/2 DP.

The W8100 has conservative clocks because of its lack of efficiency. You'd expect Tonga to rectify at least a degree of that failing....if they don't what's the point?

If you look at the HPC arena, the W9100 has almost a 60% lead in performance/watt vs kelper in DP due to the 1/2 rate DP.

I guess you're talking about specification numbers? I think I'd put more stock in actual performance- as would any prospective customer. The W9100 has a theoretical FP64 figure of 2618.9 GFlops, the Nvidia Tesla K40 is 1430 GFlops - that's an 83% supposed advantage, yet...

Need more proof? Check out the Green500 and the Top500 list. Not much in the way of FirePro is there? There is a vast difference between theoretical and actual performance. Meanwhile Maxwell (which you seem to marginalize for esoteric workloads the GM 204 won't be used for) will likely also feature 1:2 FP64 as the GM200, and if it approaches the 96+% efficiency of GM 204 then AMD has its work cut out for it.
If AMD are hinging their future success on double precision workloads, then their driver team really need to start paying attention to their work. Strange that you are now pinning your hopes on FP64 when single precision has been traditionally the stronger aspect of compute for AMD, and that traditional strength is now being taken apart by Maxwell (see the last link).

 

[Assimilator]

I forsee a very short life for the R9 285. Unless AMD likes to keep getting curbstomped.

 

[ZoneDymo]

yay, AMD, bring out new good shit thanks

 

[Frick]

I forsee a very short life for the R9 285. Unless AMD likes to keep getting curbstomped.

Dunno, phase out the 280 and lower the price... My problem with it is the market. The midrange is so confusing right now, with a million OEM's selling cards at a million prices. In the €200-250 bracket we've got 270X, 280, 285, 280X, 760, 770... In a million different flavours.

 

[dj-electric]

The high end (350-500$) market had GTX 780, GTX 780 Ti, R9 290 and R9 290X. Now it has GTX 970 and all the rest can go to the trash (unless by some kind of magic the R9 290X will drop from 480$ to 300$).

I expect the 250-300$ market to act the same way. GTX 960 (or Ti) will come in baring a 250-270$ price point. It will outperform the R9 280X and kill the whole market, making AMD think of a proper solution instead of remixing the same architecture again.

 

[Hilux SSRG]

Dunno, phase out the 280 and lower the price... My problem with it is the market. The midrange is so confusing right now, with a million OEM's selling cards at a million prices. In the €200-250 bracket we've got 270X, 280, 285, 280X, 760, 770... In a million different flavours.

I agree. It's a sea of midrange mediocre cards and the new cards, with shiny new stickers, are barely better performance wise.

My friend got really excited wanting to buy a 285 recently until I showed him better options for far less.

 

[RejZoR]

Why can't they just stick framebuffer compression and GCN 1.2 cores into a R9-290X, give it a lower price and call it a day? It would easily harass GTX 900 series cards seeing how they aren't realyl any faster than R9-290X. 290X already has ultra wide memory bus, giving it additional edge with compression, more efficient shaders and higher tessellation rate, it'd be up there and most likely use slightly less power thn existing R9-290X. AMD could make a rather quick response to NVIDIA's release, but they are apparently playing it very safe...

 

[techy1]

I am still interested only in Price/performance not watt/performance... why, you might ask - well cuz 4K wont run it self with low-wats/med-performance, but 4K will need raw power to unseen levels (still even 2x gtx 980 demanding titles can not push beyond lousy 40 frames - :O ) ... so here might AMD come in to play and smoke GTX 980 with price/performance.. no mater that this card (so far imaginary) would be like 300w TDP vs nvidias 165w

 

[Casecutter]

Well considering the R9 285 is just the "Tahiti LE" of Tonga, I think we know pretty much exactly how this 348-Bit Tonga will bump while, figure the power stays close/better to what the 285 offered. It's not much to get excited about, they'll have to soldier-on; pip-pip...chin up and all that.

I can't see them offering this it as next-gen like 380X on GDDR, but there's that whole tweet from Robert Hallock the R9 285X graphics product officially doesn’t exist? This is a "Big Reach" perhaps with stacked HBM memory as a "pipe cleaner" . It’s hard to imagine how/what AMD can do within 28nm.

Not sure what can be taken from this yesterday... but it's a/something to read.
http://www.thinkcomputers.org/amd-to-launch-radeon-r9-380x-in-february/

 

[the54thvoid]

Not sure what can be taken from this yesterday... but it's a/something to read.
http://www.thinkcomputers.org/amd-to-launch-radeon-r9-380x-in-february/

Sounds completely reasonable tbh. But that would be 380X competing against a 5-6 month old 970/980. I'll stake a claim now (and may it be quoted against me later) that the 390X won't outperform the proper daddy Maxwell part (be it 990, 980ti or Titan II or other silly name).

Then again.... stranger things have happened. There is the remote possibility Nvidia will release a crippled 980ti and have an even bigger part for the 20nm process just to screw with AMD later in 2015.

 

[arbiter]

Sounds completely reasonable tbh. But that would be 380X competing against a 5-6 month old 970/980. I'll stake a claim now (and may it be quoted against me later) that the 390X won't outperform the proper daddy Maxwell part (be it 990, 980ti or Titan II or other silly name).

Then again.... stranger things have happened. There is the remote possibility Nvidia will release a crippled 980ti and have an even bigger part for the 20nm process just to screw with AMD later in 2015.

What will probably happen if its 5-6 months before it comes out, AMD will announce it then within a few days Nvidia will drop a new card that will beat it. It's happened a few times already.

 

[Steevo]

Well considering the R9 285 is just the "Tahiti LE" of Tonga, I think we know pretty much exactly how this 348-Bit Tonga will bump while, figure the power stays close/better to what the 285 offered. It's not much to get excited about, they'll have to soldier-on; pip-pip...chin up and all that.

I can't see them offering this it as next-gen like 380X on GDDR, but there's that whole tweet from Robert Hallock the R9 285X graphics product officially doesn’t exist? This is a "Big Reach" perhaps with stacked HBM memory as a "pipe cleaner" . It’s hard to imagine how/what AMD can do within 28nm.

Not sure what can be taken from this yesterday... but it's a/something to read.
http://www.thinkcomputers.org/amd-to-launch-radeon-r9-380x-in-february/

Its the memory bandwidth starvation that keeps the 285 from performing well at all, their delta compression wasn't enough to save it and possible overuse of on chip caches that consume too much power were a bad mix, but without a doubt this chip is a turd, and why they released it so crippled is beyond stupid, I hope whoever made that decision gets the ax. If they had used cheaper slower memory and binned it better, added a second clock generator for the UVD it would have been a much bigger win in power savings.

A 20-25% performance increase is still only slightly faster than the 280X.

 

[arbiter]

Not sure what can be taken from this yesterday... but it's a/something to read.
http://www.thinkcomputers.org/amd-to-launch-radeon-r9-380x-in-february/

AMD saying 380x will compete against the gtx980/970, yea We'll see what History repeats itself on that front for AMD.

 

[Hilux SSRG]

AMD saying 380x will compete against the gtx980/970, yea We'll see what History repeats itself on that front for AMD.

Rumors that the 390/390X will tackle the 980Ti/Titan 2 as well.

AMD in the second half of 2015 on 20nm and HBM will be very interesting to say the least.