videocardzAMD Radeon R9 290X confirmed to feature 64 ROPs

[raghu78]

IF these specs are completely accurate, it sure looks as you say. Except it'll be the 290 (non X) that's ~10% faster than Titan, since it is effectively a 7870*2 on a single die.

Maybe?

normally pure 100% scaling is rare. 90% is more like it. and since R9 290 is running at 50 mhz lower clock than HD 7870 it should end up at around 85% R9 290X though will easily hit 2x HD 7870 perf.

 

[Saylick]

I don't think it'll use 250W average. Not even close. 250W max in games, is very likely though.

Ehhh, it will depends how hard games push it. With less demanding games, then yeah, I can see it being lower.

THW reports a power consumption of 92W, 124W, and 207W for their 260X, 270X, and 280X, respectively. Guru3D reports a "subjective" power consumption of 113W, 177W, and 241W for the 260X, 270X, and 280X, respectively.

Going by these numbers, you can scale each one up to get a ballpark figure for 290's and 290X's power consumption.

THW/Guru3D
260X to 290 is ~2.85x: ~262W/~322W
260X to 290X is ~3.1x: ~285W/~350W

270X to 290 is ~1.9x: ~236W/~336W
270X to 290X is ~2x: ~248W/~354W

280X to 290 is ~1.23x: ~259W/~296W
280X to 290X is ~1.3x: ~269W/~313W

Ignoring the silly results after scaling Guru3D's numbers, we're looking at a ballpark figure around 250W, but I'm thinking the most probably outcome is ~240W after architectural enhancements.

EDIT: Added links.

 

[3DVagabond]

Ehhh, it will depends how hard games push it. With less demanding games, then yeah, I can see it being lower.

THW reports a power consumption of 92W, 124W, and 207W for their 260X, 270X, and 280X, respectively. Guru3D reports a "subjective" power consumption of 113W, 177W, and 241W for the 260X, 270X, and 280X, respectively.

Going by these numbers, you can scale each one up to get a ballpark figure for 290's and 290X's power consumption.

THW/Guru3D
260X to 290 is ~2.85x: ~262W/~322W
260X to 290X is ~3.1x: ~285W/~350W

270X to 290 is ~1.9x: ~236W/~336W
270X to 290X is ~2x: ~248W/~354W

280X to 290 is ~1.23x: ~259W/~296W
280X to 290X is ~1.3x: ~269W/~313W

Ignoring the silly results after scaling Guru3D's numbers, we're looking at a ballpark figure around 250W, but I'm thinking the most probably outcome is ~240W after architectural enhancements.

Links would help.

Of course there is no absolutely correct figure to go by, but going by TPU...

The 270X uses 122 peak. That's highest figure recorded while playing Crysis 2. Assuming double, which I think is being very extreme, makes it 244W peak. I'm not saying that hawaii isn't ever going to exceed that figure, but saying it's going to use 250W average? I really don't think so.

 

[Saylick]

Links would help.

/snip

The 270X uses 122 peak. That's highest figure recorded while playing Crysis 2. Assuming double, which I think is being very extreme, makes it 244W peak. I'm not saying that hawaii isn't ever going to exceed that figure, but saying it's going to use 250W average? I really don't think so.

Yeah, sorry about that. I just added them in.

Hm, that's a fair assessment then. Interesting to note the jump in power from 280X to 7970 and 7970GE though. Same front end for all three cards, but Tahiti has 60% more shader power and a wider memory bus.

On another note, here's the peak power consumption for the 260X, i.e. Bonaire, via TPU:

93W at peak for a card which uses GCN 1.1, which is what Hawaii will most likely be based on. 290X has >3x the resources, so that translates into >279W. The same article posts an average power consumption of 84W for the 260X, so that translates into 252W. Again, you bring up a good point in that the scaling is never perfect, so you could probably shave off one or two dozen watts off those figures.

 

[Revolution 11]

If you can get more yield at x density, then that means that you'll also get acceptable (likely not as good) yields at the higher density.

Let me see if I get this. So if density goes down, yields improve, giving less failed die in the wafer but less die overall. If density goes up, yields decrease, giving more failed die in the wafer, but more die overall.

What AMD/Nvidia try to do is to find that sweet spot in density/yields which gives the most functional die per wafer. A sweet spot which changes over time for each node. Correct?

 

[insertcarehere]

Yeah, sorry about that. I just added them in.

Hm, that's a fair assessment then. Interesting to note the jump in power from 280X to 7970 and 7970GE though. Same front end for all three cards, but Tahiti has 60% more shader power and a wider memory bus.

On another note, here's the peak power consumption for the 260X, i.e. Bonaire, via TPU:

93W at peak for a card which uses GCN 1.1, which is what Hawaii will most likely be based on. 290X has >3x the resources, so that translates into >279W. The same article posts an average power consumption of 84W for the 260X, so that translates into 252W. Again, you bring up a good point in that the scaling is never perfect, so you could probably shave off one or two dozen watts off those figures.

Don't forget that the 260x runs at core 1.1ghz while the 290x runs at far lower clocks.

 

[Saylick]

Don't forget that the 260x runs at core 1.1ghz while the 290x runs at far lower clocks.

Heh, I didn't know that. I always thought that 260X sat at something closer to 1 GHz, like most of the "new" lineup. If that's the case, then it sounds like Vagabond's estimate is spot on. :thumbsup:

 

[csbin]

VS Titan

 

[24601]

VS Titan

That memory overclock doesn't look promising.

I wonder if the current bios editing tools for 7xxx will work for 290/290x.

Perhaps the Powertune Limit is too restrictive.
Perhaps the IMC just can't take much over 1250 mhz.

The chances of the GDDR5 itself not being able to clock at all higher than 1250 mhz is unlikely, even if it's rated at that.

People's experiences with the 7950 do not bear that out.

 

[csbin]

FSE

 

[Skurge]

That memory overclock doesn't look promising.

I wonder if the current bios editing tools for 7xxx will work for 290/290x.

Perhaps the Powertune Limit is too restrictive.
Perhaps the IMC just can't take much over 1250 mhz.

Looks to me like it's capped at those speeds. Probably just CCC not allowing you to go higher. Like when 5870/5850 came out.

 

[Haserath]

93W at peak for a card which uses GCN 1.1, which is what Hawaii will most likely be based on. 290X has >3x the resources, so that translates into >279W. The same article posts an average power consumption of 84W for the 260X, so that translates into 252W. Again, you bring up a good point in that the scaling is never perfect, so you could probably shave off one or two dozen watts off those figures.

Bonaire was based on Southern Island's GCN with some features added in(true audio and more clock states).

On the last page, Hawaii looks to have changed the arch. 3x Bonaire's die would be 480mm^2 vs low 400's for 290X. Might not be completely comparable.

 

[24601]

Looks to me like it's capped at those speeds. Probably just CCC not allowing you to go higher. Like when 5870/5850 came out.

I would assume for them to cap their VRAM speeds to only 20 mhz over stock, they would have a good reason. I think my statements still stand.

 

[Abwx]

If accurate thoses numbers give a few clues :

On the stock rating chart the 780 and 290 on a hand
and the Titan/290X on the other hand are above the rest
while in the oc chart the 290X seems to be in a class of itself ,
too bad for AMD that memory frequency is low otherwise
even the 290 would beat the titan in ocked conditions.

 

[Skurge]

I would assume for them to cap their VRAM speeds to only 20 mhz over stock, they would have a good reason. I think my statements still stand.

These cards aren't even out yet. So I don't know how much you can read into that. We don't even know what they used.

Also, its strange that the 290X and the 290 both oc to the exact same speeds.

 

[tolis626]

Regarding the memory modules used,I think we could see aftermarket 290x's with better RAM modules than the reference card (Better as in rated for higher clocks) that would clock at least up to 1500MHz.It may even be a deficiency of the reference cooler so that the memory isn't properly cooled,which could lead to thermal problems when memory is clocked high.These are all in the "maybe" sphere and just my speculations.I don't believe we won't see high memory overclocks in the future.

 

[csbin]

800mhz base(1000mhz boost)

 

[GaiaHunter]

If accurate thoses numbers give a few clues :

On the stock rating chart the 780 and 290 on a hand
and the Titan/290X on the other hand are above the rest
while in the oc chart the 290X seems to be in a class of itself ,
too bad for AMD that memory frequency is low otherwise
even the 290 would beat the titan in ocked conditions.

The OC difference in those charts is 8.5% performance. The core clock difference is 4% and the memory clock is the same.

I doubt the difference between the 290X OC and the 290 OC is the memory.

If the numbers are real.

 

[3DVagabond]

These cards aren't even out yet. So I don't know how much you can read into that. We don't even know what they used.

Also, its strange that the 290X and the 290 both oc to the exact same speeds.

True. It could just be what the catalyst limits are set at.