Possible Radeon 390X / 390 and 380X Spec / Benchmark (do not hotlink images!!!!!!)

[Davedree]

I can still sit on the fence and state something is disappointing and worrying.
As a 7950 user I can look at the 285 tonga and like I did in the 285 thread, stated my disappointment, but also provided a viewpoint that scaling Tonga to 3584 on 28nm could be possible.
Nutella you're not worried that it takes a 4096 rumoured amd card to barely outperform a 2048 maxwell ? UMM I would be

I disagree because 970/980 were just cheaper replacements of variations of gk110's with power savings and cheaper dies/pcbs/memory congigs.
This maybe 390 speculation maybe a 30% increase over a Hawaii, but it needs to be 40%-50% to be worthwhile.

It's how well maxwell scales with more cores, that it becomes scary for Amd.

Yes I said it was based on speculation so you don't need to throw it back at me like we both don't understand it's speculation.

Yes discussion is fine but at the end of the day neither of us are right or wrong until the hardware is released.

edited 2nd line

 

[Gloomy]

"This is an unstable sample, the actual final product may have changes such as: spec changes and improved drivers"



Basically, this leak is so early basic things like clockspeed aren't hammered down. And it doesn't even have drivers.

 

[Yaayuh!]

Just because WCCF say so?

They themselves even admit they have no reason to believe that other than "it's the fastest card in the test".

Also, funny how 980 was a top-end card one day and mid-range the next (when an AMD leak happens to show 980 being pwned soundly)... despite the fact GM200 has yet to show it's face.

Also also, "made-up" LOL. If it showed GM200 I'm sure your enthusiasm would be positively palpable.

Calm down Orangey, no need to fly off the handle.

I was questioning whether it was the 390X or 380X, not stating that it was either.

I never said it wasn't top-end, read my post properly. I said that the GTX980 was a.) nVidia's currently released and highest performing card (which makes it their top-end card) and b.) The midrange Maxwell card.
This was a response to placing links to a rumour and showing it off as if it should be taken as fact/evidence. I was merely giving a realist version of your post.

And last but not least, "made-up", well it's graph posted on the internet before release, i'm going to believe it's made up until proven otherwise. You're naive to believe rumours and leaks this early on.

I hope it is true and is in fact just the 380X, that'd be some leap in performance! I'd also hope it'd come early next year so the war between red and green will allow me to pick for my birthday as i am due a replacement.

Prime example of why this section of the forum is the most infuriating and to be honest, one of the worst out there. Too many fanboys and people jumping to conclusions left right and centre. I would just like to see some decent evidence before people start showing it off as if it was true.

 

[bru]

17% Faster while using 7% more power, thats 10% better efficiency than Maxwell.

To be honest that is the thing that makes me doubt the whole thing. the performance in my opinion is quite feasible but the power consumption I'm not too sure about.
Looking at it this way, power wise Tonga is AMD's 750ti. ie the first run out the gate for the new power efficiency of their new architecture, then there is this leak, which is either inline with tonga's power efficiency if this is the 380x. But then the performance is in LA LA land or the performance is good and its the 390x in which case the power efficiency is in LA LA land.

AMD are good at making GPU's but they are not miracle workers.

One other thing, if this is the 390x and it does arrive first. It might force Nvidia hand to bring us the consumer big Maxwell before any Titan esque repeat.

Oh yes another thing, when did we hear about the less than 28nm taping out of these chips. Seeing as we always get to know when they have taped out months in advance. Of course if this is 28nm then one of those graphs really is living in LA LA land.

 

[Nutella33]

Nutella you're not worried that it takes a 4096 rumoured amd card to barely outperform a 2048 maxwell ?

nope. factor in that its a ES sample, has immature drivers and completely unknown clockspeed. I dont think anyone was too worried when the 9xx was released and showed 10% improvement over last gen and this is showing more than that already.

I disagree because 970/980 were just cheaper replacements of variations of gk110's with power savings

and yet multitudes bought them for the power saving alone and they have been a massive success. did you post on the 980's release saying its not good enough because AMD would beat it later? No company releases a product to beat a future mystery product, they release to beat the one in the market thats getting sales.

This maybe 390 speculation maybe a 30% increase over a Hawaii, but it needs to be 40%-50% to be worthwhile.

easily attainable if the figures are true with a matured release ready product and a driver.

It's how well maxwell scales with more cores, that it becomes scary for Amd.

not really. they will just stick two slightly lower clocked cores on a pcb to combat any titan type release. like they have done for every other release in recent memory.

Yes I said it was based on speculation so you don't need to throw it back at me like we both don't understand it's speculation.

then why bother bullet pointing it as if its something no-one understands?
equally stating we wont know until the cards released is a mute point and doesn't cancel any speculation out.

 

[humbug]

If these rumours are true than a 2880 Shader Maxwell would need some very impressive scaling.

2048 Shaders + 40% = 2867 Saders.

Just adding Shaders alone never scales 100%, GK104 to GK110 scaled about 50% (Clock for clock) in performance with a 33% wider bus, 50% more ROP's and 80% more Shaders.

Tahiti to Hawaii scaled about 45% (Clock for clock) with GCN 1.0 to GCN 1.2, 33% wider bus, 100% more ROP's and 40% more Shaders.

With 40% more Shaders alone Big Maxwell would be about 20% faster than the 980 and use about 15% more power, that would put it on par with the 380X? and use a little more power.

To be honest that is the thing that makes me doubt the whole thing. the performance in my opinion is quite feasible but the power consumption I'm not too sure about.
Looking at it this way, power wise Tonga is AMD's 750ti. ie the first run out the gate for the new power efficiency of their new architecture, then there is this leak, which is either inline with tonga's power efficiency if this is the 380x. But then the performance is in LA LA land or the performance is good and its the 390x in which case the power efficiency is in LA LA land.

AMD are good at making GPU's but they are not miracle workers.

One other thing, if this is the 390x and it does arrive first. It might force Nvidia hand to bring us the consumer big Maxwell before any Titan esque repeat.

Oh yes another thing, when did we hear about the less than 28nm taping out of these chips. Seeing as we always get to know when they have taped out months in advance. Of course if this is 28nm then one of those graphs really is living in LA LA land.

Your assuming that this GPU is the same as Tonga and on 28nm with no HBM, its a very narrow assumption to make and probably way off.

 

[Ayahuasca]

It's also a very narrow assumption to make that nvidias full fat maxwell will only be on par with the 380x lol.

 

[humbug]

It's also a very narrow assumption to make that nvidias full fat maxwell will only be on par with the 380x lol.

Nvidia have 40% Shaders to add to Maxwell to bring it up to 550mm^2, Adding 80% more Shaders + 50% More ROP's + a 33% Wider bus got 50% extra performance from GK104 to GK110

If Maxwell has 2880 Shaders it would be 20% faster than the 980 (par with 380X) + 96 ROP's + a 384Bit Bus it would be 30% faster or 10% faster than 380X.

 

[flopper]

It's also a very narrow assumption to make that nvidias full fat maxwell will only be on par with the 380x lol.

if samsungs suit goes trough there might not be any Nvidia cards to buy

 

[Ayahuasca]

Nvidia have 40% Shaders to add to Maxwell to bring it up to 550mm^2, Adding 80% more Shaders + 50% More ROP's + a 33% Wider bus got 50% extra performance from GK104 to GK110

If Maxwell has 2880 Shaders it would be 20% faster than the 980 (par with 380X) + 96 ROP's + a 384Bit Bus it would be 30% faster or 10% faster than 380X.

There's not a chance nvidia would release their high end card after AMDs offerings that could only match the "mid tier" AMD card.

if samsungs suit goes trough there might not be any Nvidia cards to buy

Yeah because that would be brilliant for PC gaming and graphics card prices.

 

[humbug]

One other thing.

AMD have just released an APU with an iGPU with a voltage and power regulation technology similar to Maxwell, on 28nm its <20% more power efficient than its predecessor (the CPU is also <40% more mower efficient)

A 290X with that ^^^ technology would be reduced from 270 Watts to 220 Watts, 290P from 240 Watts to 200 Watts on 28nm without HBM; with it it might take another 20 Watts off that power consumption.

It also likely that these are on one of the new processes, which could reduce the power a further 30%, bringing the end power consumption down on a 290X equivalent + New power tech + HBM to about 140 Watts.

 

[Kaapstad]

Nvidia have 40% Shaders to add to Maxwell to bring it up to 550mm^2, Adding 80% more Shaders + 50% More ROP's + a 33% Wider bus got 50% extra performance from GK104 to GK110

If Maxwell has 2880 Shaders it would be 20% faster than the 980 (par with 380X) + 96 ROP's + a 384Bit Bus it would be 30% faster or 10% faster than 380X.

When not bottlenecked by the CPU/Settings going from GK104 to GK110 scales almost perfectly in line with the number of SMX modules used. In the example below with the GTX 690s and Titans it is 8 to 14 SMX modules.

4 GPUs

1. Score 3525, GPU nvTitan @994/1777, CPU 3930k @5.0, Kaapstad Link
2. Score 3457, GPU 290X @1230/1500, CPU 4930k @4.8, Kaapstad Link
3. Score 2042, GPU 690 @1160/1800, CPU 3960X @4.9, Kaapstad Link
4. Score 1064, GPU 590 @613/855, CPU i7 980X @4.29, Kaapstad Link

The above table is taken from the Heaven 4 1600p bench.

Using this setting is enough to remove any CPU bottleneck once you reach about 4.4ghz.

 

[humbug]

The 690 you have listed there ^^^ is only running one GPU, IE its a 680. Haven is also very 'Shader and' Nvidia friendly. in a mixture of games it doesn't look like that, Unigine is pretty unique in that.

 

[Davedree]

The 690 you have listed there ^^^ is only running one GPU, IE its a 680. Haven is also very Nvidia friendly.

Why's that's irrelevant, it's comparing nvidia vs nvidia and scaling?

 

[TheRealDeal]

The 690 you have listed there ^^^ is only running one GPU, IE its a 680. Haven is also very Nvidia friendly.

No as it's a 4 gpu benchmark basically 4 Titans v 4 gtx680's. It's not completely accurate as clock speeds are different. I would say the Titans at boost clocks would be slightly higher. Maybe wrong about that as i am not sure if Kaap is listing the Titans at normal or boost and same for gtx690.

 

[humbug]

Why's that's irrelevant, it's comparing nvidia vs nvidia and scaling?

I'm not saying Its irrelevant, it is relevant for Unigine, but nothing else. Unigine is entirely Shader bound and somewhere unique in that.

 

[Davedree]

I'm not saying Its irrelevant, it is relevant for Unigine, but nothing else. Unigine is entirely Shader bound and somewhere unique in that.

edited your post after i made mine.

 

[bru]

It also likely that these are on one of the new processes, which could reduce the power a further 30%, bringing the end power consumption down on a 290X equivalent + New power tech + HBM to about 140 Watts.

I would have to agree with you that if this leak is remotely accurate then the new card will have to be on a new process. Which I am very surprised that we have no indication of which and even whose process it will be and yet this leaker has a fully working card that he was able to bench in 23 games against 17 other cards on surely un-optimised drivers with un-optimised clocks speeds and it still came out nearly 20% faster. Not forgetting that this is supposedly the 380x.

Also if it sips power in such a frugal way, why has it been suggested that it will use a hydro cooling solution as a lot of previous leaks have stated.

But hell if it is accurate then AMD will surely have a winner on their hands and seeing as this leaker has been able to run all that on it, it must be nearly ready to ship. Which will be good competition and hopefully move things along in the GPU world.

 

[humbug]

edited your post after i made mine.

lol...

Gota watch my Ninja edits.

 

[Davedree]

lol...

Gota watch my Ninja edits.

lol but i have proof !
peace xx