FUD: AMD Tonga GPU comes as Radeon R9 285

[f1sherman]

In laptops I'd agree and that's why I bought a Maxwell. Power efficiency is not the top priority of enthusiasts who will generally buy desktops. We care about performance.

If you guys care about 10% more power in the same thermal envelope that's your opinion. I'm cheap but I'll splurge on good hardware if it gives me noticeable performance.

yeah... I've actually never met on forums ANYONE who makes his purchase based on 10% power difference, 30 W whatever...

and yet time and again I have to hear about "you guys who care about power so much".
I mean WHO ARE THESE PEOPLE you are arguing with?

For the umptieth time - it's not about 30W of power bill.
It's about who has more power efficient arch, it's about engineering excellence and it's about those 30W potentially deciding high-end perf. and mobile winner.

 

[3DVagabond]

In the big picture, 30W matters. Just like the size of the chip matters. There's a reason it's always perf/W and perf/mm². Both determine perf/$, which in the end is what truly matters. Now, which one if any is the winning spec? For instance we have Hawaii winning the perf/mm² metric and we have GK110 winning the perf/W comparison. In the end which one saves the most money for you and me?

 

[blackened23]

Actually, 10-30W can make a difference to OEMs. But it probably doesn't matter to a typical desktop user. 10% power doesn't mean jack for the desktop. Yeah. This is obvious. However, one has to consider the big picture.

People saying that 20-30W doesn't mean anything for desktops don't really get what's going on.

GPUs are created for mobile first and scaled upwards for mobile, data center, corporate use, all the way up to servers. You bet your A-- that performance per watt matters for all of those segments, but maybe it doesn't matter to Joe C. Desktop user.. Kepler was mobile first and scaled all the way to the 780ti. Hawaii starts out as a mobile part and scales up to the 290X. So on and so forth. Mobile graphics in the long term is far, far more profitable for everyone that can make a dent in the market; i'm sorry it has to be said that AMD has suffered significant losses in the mobile dGPU market because their efficiency has been quite frankly, horrible, compared to Kepler in prior years. Also, nvidia did not do well at all with Fermi, that was (in my opinion) why NV was really motivated to focus on efficiency with Kepler. Remember, NV is using the Kepler technology in everything from mobile dGPUs to tegra, and while tegra hasn't sold billions of units...their mobile dGPUs with Kepler have done very well, all because of the efficiency.

This applies to INTEL as well. Why do you think all of their parts are performance per watt focused? Duh, mobile first. For high performance oriented mobile ultrabooks and portables such as macbooks, intel is bar none the best. This is why efficiency matters. It scales up from the mobile parts, data center parts (YES, performance per watt MATTERS here), corporate sales, all the way up to the desktop. Intel gets it. Nvidia gets it. AMD gets it as well, but they just haven't delivered as well. Some of the fans who talk about 20W not being a big deal DON'T get it.

This is where I think Tonga comes in. Efficiency first and it will certainly be used in mobile designs. If the efficiency claims are true, then that will be AMD's start for making a dent in the mobile dGPU market again. But it all comes down to software. AMD's dual graphics mode is assinine compared to optimus, hopefully AMD can get this up to par. It hasn't been good for several years now.

Why do you guys even THINK that everything is based on efficiency now? Use your common sense. Everything is scaled from the bottom up , and the mobile parts DEPEND ON EFFICIENCY and performance per watt. Performance per watt may not mean JACK SQUAT for a 250W TDP dGPU, but I assure, if AMD's efficiency is garbage for the low end that means they will not be selling any mobile dGPUs. And they really haven't sold any in recent years - I do hope Tonga is successful in this respect. It'll take a combination of good performance per watt and correction of AMDs dual graphics mode for mobile GPUs. I really feel like while Tonga will be an interesting desktop part, the bigger picture will be AMD applying that same technology for mobile dGPUs and applying some of the power optimizations to their next gen APUs.

 

[3DVagabond]

Pitcairn didn't suffer any efficiency issues. What mobile device uses Hawaii or Tahiti? I don't think AMD's mobile issues are because GK110 uses 10% less power than Hawaii.

 

[blackened23]

GCN 1.0 consisted (and GCN 1.1 does as well) of many parts (including pitcairn) all of which were based on the same tech, just like Kepler consists of many parts ranging from the GK107, 106, 104, 110, etc etc etc all based on the same tech. I think you're well aware of this.

All of these GPUs use the same design architecture as the baseline and are scaled in terms of clockspeeds, memory, shaders, etc etc etc. GCN 1.0 scaled all the way (for the desktop) to the paltry 7850 (pitcairn) all the way up to the 280X and 7970. This same design was used for AMD's mobile dGPUs. All GCN 1.0 parts are based on similar tech just as Kepler GM104, 106, 107, 110 are. Kepler is based on mobile first, and just did it better than AMD's attempt with GCN 1.0.

 

[rgallant]

idd it is
(Hawaii vs GK-110)

100W from smaller cheap will lead to higher temps for two reasons:

• heat capacity of bigger chip is bigger than heat capacity of smaller chip, leading to smaller temp delta
• bigger die area means better conductivity, means easier to cool

Try OCing a 290 on air and see how it feels.

-yea but ref. gpu's don't draw much more power . op most likely moved to a bigger card and it draws more watts and gives off more heat.

yea but it does not create more heat per watt if thats what you mean .

under full load the nv = might be 20-30 watts less so if op can sense what go wc and put the rads below the floor like I do.

I have one core [3770k] running 4c under my water temps [at idle only] ,there are only a few crap sensors and free software giving numbers so don't loose sleep over it .

my gpu's per msi-ab run 3-5c [21-20] cooler than my watertemps [23c], if said hardware run's fine don't worry about it.imo

 

[3DVagabond]

GPUs are created for mobile first and scaled upwards for mobile, data center, corporate use, all the way up to servers.

Tahiti was designed first. It wasn't scaled up from a mobile design. What you are saying here is simply wrong.

 

[FatherMurphy]

The only GPU architecture from Nvidia or AMD that I recall having the stated design purpose of being designed "mobile first" is Maxwell. Even Kepler wasn't designed that way, I don't believe. Rather, lower power parts were fit into laptops because they were simply lower power.

From Anandtech's review of the 750ti:

"As we found out back at NVIDIA’s CES 2014 press conference, Maxwell is the first NVIDIA GPU that started out as a “mobile first” design, marking a significant change in NVIDIA’s product design philosophy. The days of designing a flagship GPU and scaling down already came to an end with Kepler, when NVIDIA designed GK104 before GK110. But NVIDIA still designed a desktop GPU first, with mobile and SoC-class designs following. However beginning with Maxwell that entire philosophy has come to an end, and as NVIDIA has chosen to embrace power efficiency and mobile-friendly designs as the foundation of their GPU architectures, this has led to them going mobile first on Maxwell. With Maxwell NVIDIA has made the complete transition from top to bottom, and are now designing GPUs bottom-up instead of top-down."

http://www.anandtech.com/show/7764/the-nvidia-geforce-gtx-750-ti-and-gtx-750-review-maxwell

 

[blackened23]

You're the one in the wrong here, Tahiti was a part of GCN 1.0 and it certainly scaled from the bottom up, as all of those parts use the same design technology. It wasn't apparent at retail but all of these parts use the same tech, from the mobile dGPUs to the high end dGPUs for desktops.

The difference is, AMD applies different codenames to each and every single part in their respective performance category - the low/mid range was pitcairn, the high performance category was Tahiti and Tahiti pro. NV does not share this name convention, everything is Kepler. Each and every Kepler part is a Kepler part, from the GK106 to the GK110.. But AMD's GCN 1.0 does have a different name depending on what performance category it targets (for the desktop at least). But AMD has the same GCN 1.0 tech in every single one of those parts, and the differences are related to clockspeeds, memory, bus width, shader counts, etc. They are all GCN 1.0 designs from AMD. They are all using the same tech. Just like while GK107 is a far cry from the performance of GK110, it IS THE SAME tech. Even a cheesy GT640, while being pretty bad performance wise, shares the same tech as does the 780ti. But NV doesn't apply different names to different performance segments - they just code name the chip with GKxxx. AMD uses actual code names depending on the performance category.

 

[3DVagabond]

You are simply ignoring the facts. Tahiti was not designed up from the smaller Pitcairn chip.

 

[tviceman]

yeah... I've actually never met on forums ANYONE who makes his purchase based on 10% power difference, 30 W whatever...

and yet time and again I have to hear about "you guys who care about power so much".
I mean WHO ARE THESE PEOPLE you are arguing with?

For the umptieth time - it's not about 30W of power bill.
It's about who has more power efficient arch, it's about engineering excellence and it's about those 30W potentially deciding high-end perf. and mobile winner.

Actually, 10-30W can make a difference to OEMs. But it probably doesn't matter to a typical desktop user. 10% power doesn't mean jack for the desktop. Yeah. This is obvious. However, one has to consider the big picture.

People saying that 20-30W doesn't mean anything for desktops don't really get what's going on.

When Silverforce first showed up on these forums, it was his #1 argument against Fermi. Of course that tune changed over time... but there you go, you've now met someone who at least used to use that as an excuse.


But seriously, I do care about power consumption. I'm running an SFF build with a 450 watt GPU. I don't overvolt my GPU, but I will overclock. And that said, GK110 has so much more headroom without breaking the power bank than Tahiti has. Which is why I got a gtx780.

 

[blackened23]

You are simply ignoring the facts. Tahiti was not designed up from the smaller Pitcairn chip.

I just explained this. Cape verde, and pitcairn are all GCN 1.0 based designs and share the same tech. Differences? Very minor, with the most obvious differences being the number of compute engines, shader engines, rasterizers, steam processors or CUDA cores, all scaling up and down. You can add or remove, you can go up or down in TDP based on the same GCN 1.0 tech or the Kepler tech. GK107 has a bunch of stuff removed just like cape verde has a bunch of stuff removed. I already explained that AMD applies codenames to each performance segment, while NV does not do this. Kepler is Kepler. GCN 1.0 is pitcairn and tahiti. GCN 1.0 scales from the bottom up. You go ahead and think what you will though and ignore it. Keep thinking that efficiency doesn't matter to AMD. IMO, the fact that they haven't sold many mobile dGPUs in a while probably was their main motivation behind updating their architecture with Tonga.

Performance per watt is the new metric for all silicon. Anyone who thinks otherwise doesn't get the bigger picture, period.... because nearly everything in the PC world is designed as scalable architectures. Haswell is designed like this, GCN 1.0 is designed like this, and Kepler/Maxwell are designed like this. These designs can in the small ultrabooks all the way up to enterprise professional cards with intel, AMD and NV adding and removing from the architecture (in terms of engines, clockspeeds, cores, etc) as they wish to fit their TDP or form factor goal.

 

[Techhog]

So the boxes are not printed yet?



....then I don't believe these cards will be called R9 285X.

Same number of stream processors as R9 280X, but significantly narrower memory bus?

I'll bet it gets called R9 275X or something else instead.

But then what would they name the Tonga XT?

 

[3DVagabond]

You can repeat yourself all you want to. It doesn't make it fact. I wasn't addressing naming conventions or shader arch. I never said efficiency doesn't matter (I actually said the opposite). You are assigning arguments to me which I never stated and then trying to say I'm wrong.

Fact: Tahiti was not scaled up from Pitcairn. (Tahiti was designed before Pitcairn.)
Fact: Tahiti is not a mobile GPU.
Fact: Pitcairn is AMD's top mobile GPU.
Fact: Pitcairn is very energy efficient. (Keeping in mind it's >2 years old now)

Also, however nVidia designed Kepler, Maxwell, etc... is irrelevant to anything I said. I'll repeat it one more time. Tahiti was not designed by scaling up Pitcairn. If anything it's the other way around. Tahiti was the full blown performance chip and Pitcairn was designed with compute functionality it didn't need removed and an improved memory bus to make it more efficient. More efficient in both metrics, perf/mm² and perf/W. Because both matter.

 

[Spanners]

You're the one in the wrong here, Tahiti was a part of GCN 1.0 and it certainly scaled from the bottom up, as all of those parts use the same design technology. It wasn't apparent at retail but all of these parts use the same tech, from the mobile dGPUs to the high end dGPUs for desktops.

The difference is, AMD applies different codenames to each and every single part in their respective performance category - the low/mid range was pitcairn, the high performance category was Tahiti and Tahiti pro. NV does not share this name convention, everything is Kepler. Each and every Kepler part is a Kepler part, from the GK106 to the GK110.. But AMD's GCN 1.0 does have a different name depending on what performance category it targets (for the desktop at least). But AMD has the same GCN 1.0 tech in every single one of those parts, and the differences are related to clockspeeds, memory, bus width, shader counts, etc. They are all GCN 1.0 designs from AMD. They are all using the same tech. Just like while GK107 is a far cry from the performance of GK110, it IS THE SAME tech. Even a cheesy GT640, while being pretty bad performance wise, shares the same tech as does the 780ti. But NV doesn't apply different names to different performance segments - they just code name the chip with GKxxx. AMD uses actual code names depending on the performance category.

None of this is any proof that things were designed from mobile up."IS THE SAME tech" is just as strong an argument for top-down as it is for bottom-up.

Edit: Nothing in your response to Techhog offers anything convincing either.

 

[Silverforce11]

When Silverforce first showed up on these forums, it was his #1 argument against Fermi. Of course that tune changed over time... but there you go, you've now met someone who at least used to use that as an excuse.


But seriously, I do care about power consumption. I'm running an SFF build with a 450 watt GPU. I don't overvolt my GPU, but I will overclock. And that said, GK110 has so much more headroom without breaking the power bank than Tahiti has. Which is why I got a gtx780.

No, it never changed. I still hate cards that are terribly inefficient compared to the competition, and the power bill does factor in the final price.

Fermi (GTX480) was horrible compared to the 5870, it literally doubled the power (I still cringe seeing Guru3d or AT's Crysis power load for it!) for 10-15% the performance gained.

Now with the R290/X situation, a custom cooled card not running at 94C, as have been shown by many sites, uses very close to the GTX780/ti in power. They only become inefficient when OC with vcore. That's the drawback of a high density smaller die approach, more vcore with those transistors crammed in, leads to more leakage.

Price-wise, at the time and even now, where I was (Europe & later moved to Australia), custom R290s are ~$150 cheaper than 780, and R290X was ~$250 cheaper than 780Ti. It wasn't even a fair comparison, as someone who values perf/$ as well as perf/w, R290s was the clear winner.

Power always matters, but it needs to be taken into context. What is the market? High end, power matters less, and often doesn't matter at all IF the performance is great enough. But sure, mobiles, perf/w is #1, there's no denying that.

ps. I pay ~30 USD cents per kWh. Power matters. This is also why I undervolt my CF R290, and its running at <60C, each of them draw ~200W.. under MINING load.

 

[Techhog]

I just explained this. Cape verde, and pitcairn are all GCN 1.0 based designs and share the same tech. Differences? Very minor, with the most obvious differences being the number of compute engines, shader engines, rasterizers, steam processors or CUDA cores, all scaling up and down. You can add or remove, you can go up or down in TDP based on the same GCN 1.0 tech or the Kepler tech. GK107 has a bunch of stuff removed just like cape verde has a bunch of stuff removed. I already explained that AMD applies codenames to each performance segment, while NV does not do this. Kepler is Kepler. GCN 1.0 is pitcairn and tahiti. GCN 1.0 scales from the bottom up. You go ahead and think what you will though and ignore it. Keep thinking that efficiency doesn't matter to AMD. IMO, the fact that they haven't sold many mobile dGPUs in a while probably was their main motivation behind updating their architecture with Tonga.

Performance per watt is the new metric for all silicon. Anyone who thinks otherwise doesn't get the bigger picture, period.... because nearly everything in the PC world is designed as scalable architectures. Haswell is designed like this, GCN 1.0 is designed like this, and Kepler/Maxwell are designed like this. These designs can in the small ultrabooks all the way up to enterprise professional cards with intel, AMD and NV adding and removing from the architecture (in terms of engines, clockspeeds, cores, etc) as they wish to fit their TDP or form factor goal.

I'm simply astonished with how bad some people here are at supporting arguments. Using the same tech isn't even remotely proof of it being designed from the bottom up. They've always used the same tech in the same generation (not counting some outliers). It could have been designed in either direction. I don't see what you're getting at here at all.

 

[blackened23]

Okay. Whatever you say buddy. Just keep thinking that AMD doesn't care about selling mobile dGPUs, and that they intentionally had a bad architecture for that purpose. I wonder Tonga is more efficient. HMM. I guess they REALLY think you guys want 20W less power consumption on the desktop. He-he. Not for ANY OTHER reason.

You just go ahead and think what you think. I'm done spelling out obvious things to you. The first product being "x" doesn't mean anything, the first maxwell product was a desktop GPU. But what was maxwell designed for. Mobile first. Why was Kepler designed to be so scalable with excellent performance per watt and scaling? It definitely isn't because they're using Kepler in everything from Tegra to Quadros. Right? I mean even though the first Kepler product wasn't a 2W SOC, but it was designed specifically to be scaled down to such. And it has. But why was it designed as it was. HMM. QUESTIONS. obvious answers.

Scalable architectures go from portables to high end, only, AMD's poor performance per watt in the past in their GCN 1.0 architecture meant that they weren't selling squat for mobile dGPUs. And their poor software of course (dual graphics). Meanwhile while NV and intel are using scalable architectures, i'm sure AMD just doesn't care. They'll just intentionally create mobile dGPUs that are poor in comparison to the competition in PPW. As I said though. Keep thinking that AMD doesn't care about that extra 20W in their architecture buddy. When scaled down it does make a difference. And that difference is either selling or not selling product to OEMs. Since the latter happened for the past several years for mobile dGPUs, i'm sure AMD just doesn't care.

 

[Rvenger]

Techhog and Blackened23, this will be your final warning. The arguments stop right after my post. Agree to disagree and move on.


-Rvenger

 

[Techhog]

Techhog and Blackened23, this will be your final warning. The arguments stop right after my post. Agree to disagree and move on.


-Rvenger

Aren't debates one of the main points of forums? But fine, whatever you say. Feel free to get on me after only one post...


Moderator callouts are against the forum rules. One post? You have more than that in this thread. Nice Try.

-Rvenger

 

[Blue_Max]

No, it never changed. I still hate cards that are terribly inefficient compared to the competition, and the power bill does factor in the final price.

Fermi (GTX480) was horrible compared to the 5870, it literally doubled the power (I still cringe seeing Guru3d or AT's Crysis power load for it!) for 10-15% the performance gained.

Now with the R290/X situation, a custom cooled card not running at 94C, as have been shown by many sites, uses very close to the GTX780/ti in power. They only become inefficient when OC with vcore. That's the drawback of a high density smaller die approach, more vcore with those transistors crammed in, leads to more leakage.

Price-wise, at the time and even now, where I was (Europe & later moved to Australia), custom R290s are ~$150 cheaper than 780, and R290X was ~$250 cheaper than 780Ti. It wasn't even a fair comparison, as someone who values perf/$ as well as perf/w, R290s was the clear winner.

Power always matters, but it needs to be taken into context. What is the market? High end, power matters less, and often doesn't matter at all IF the performance is great enough. But sure, mobiles, perf/w is #1, there's no denying that.

ps. I pay ~30 USD cents per kWh. Power matters. This is also why I undervolt my CF R290, and its running at <60C, each of them draw ~200W.. under MINING load.

Being in the market for something new, this is good to know!! I've had my eye on the R9-290 for its great value, but worried about high heat and noise... a slight under-volt and NO overclocking might make it run cool & quiet even after a day of gaming, then? Nice!

 

[tollingalong]

yeah... I've actually never met on forums ANYONE who makes his purchase based on 10% power difference, 30 W whatever...

and yet time and again I have to hear about "you guys who care about power so much".
I mean WHO ARE THESE PEOPLE you are arguing with?

For the umptieth time - it's not about 30W of power bill.
It's about who has more power efficient arch, it's about engineering excellence and it's about those 30W potentially deciding high-end perf. and mobile winner.

I'm not arguing with you on mobile. I'm agreeing with you on mobile.

On desktop I don't agree. I'll gladly buy a Devil's Canyon if it made a notable difference. I'll gladly buy a R9 295X2 if it made a notable difference. Neither will make a difference for me but if they did I'd gladly splurge.

 

[ocre]

I remember something a few yrs ago about nvidia moving towards fully scalable module designs. Even though for years they had product stacks that shared similar arrangements from the top to the bottom, the module design was a much cleaner approach that offered its own set of benefits. A single smx could be used in the k1 or many together to create a mobile or desktop chip.

Nvidia always had a top to bottom product stack that shared a similar layout, but the way they achieved this changed with the module. Nvidia seen promise in this new approach as they scale up or down performance targets. Its not the only way to go at it and surely there are a few draw backs but perhaps the pros far out weigh the cons for them.

See, by module design the mobile first means what? The same modules get used across the stack. Perhaps you could bring the mobile chip to market first but its more the module first than a mobile first design.

I am not sure how amd does it these days. Are the doing full fledged module or is each chip designed individual and cut down as needed. And could be somewhere between. Surely there is benefit to scaling modules but there could be much more flexibility going another route. There are other ways to do things.

If your using module designs such as smx or smm, mobile first is more of a marketing term. The goal is to make the very best module to scale up or down.

As for efficiency, and used to kill nvidia in that aspect. The move to more complex compute cores made an impact on them just as it did nvidia and Fermi. But it only seems logical that Amd will improve this is time. I expect them to make great progress for sure. The only issue is nvidia had a head start on solving their issues. And should be able to make great progress as well. For years they have been close, there might be a breakaway here and there but they always end up back close to one another. I think this trend should continue on in the future.

 

[3DVagabond]

Being in the market for something new, this is good to know!! I've had my eye on the R9-290 for its great value, but worried about high heat and noise... a slight under-volt and NO overclocking might make it run cool & quiet even after a day of gaming, then? Nice!

Just buy one with an effective cooler. :\

 

[Blue_Max]

Just buy one with an effective cooler. :\

Easier said than done... the Newegg user reviews usually prove pretty reliable... the 290 runs pretty hot so it seems you can usually have cool or quiet - not both.