R9 380x rumor and speculation thread

[nurturedhate]

Did you see that article on NV's finances for this last quarter? Face it, we're all boned. NV saw the money, AMD wants the money, we're going to have to open our wallets up wide.

I guess I can alternate my upgrade pattern If NV/AMD resort to the current bisected release schedule, I can skip a card line up.

Yea.. I saw it, was a little depressed. Nvidia wants to be Apple, AMD wants to be Intel/Nvidia. There goes my wallet.

That is basically what I have resorted to, skipping a generation, then again 780ti to 980ti is definitely not the same as 580 to 780ti.

 

[railven]

Yea.. I saw it, was a little depressed. Nvidia wants to be Apple, AMD wants to be Intel/Nvidia. There goes my wallet.

That is basically what I have resorted to, skipping a generation, then again 780ti to 980ti is definitely not the same as 580 to 780ti.

I was forced to a two-year upgrade pattern because of the GF. But I ended up with a 780 out of desperation at my disappointment with CFX 7970 and SLI 660 Ti. When I sold my 7970s they had piss poor drivers, so I figured two 660's I had would tied me over until 290X became affordable. Nope. Snapped, bought cheapest fastest single card I could find.

This upgrade was my normal period, and I think the 980 Ti is going to last me until big pascal. At least I hope. Unless AMD totally brings it with Arctic Islands then I'll upgrade just to chill with AMD again (dat new CCC!).

[Yes, I know, I'm part of the problem, but no kids (yet), well paying job, no country club membership, second mortgage isn't a go (yet), sorry guys I need my fix!]

 

[nurturedhate]

I was forced to a two-year upgrade pattern because of the GF. But I ended up with a 780 out of desperation at my disappointment with CFX 7970 and SLI 660 Ti. When I sold my 7970s they had piss poor drivers, so I figured two 660's I had would tied me over until 290X became affordable. Nope. Snapped, bought cheapest fastest single card I could find.

This upgrade was my normal period, and I think the 980 Ti is going to last me until big pascal. At least I hope. Unless AMD totally brings it with Arctic Islands then I'll upgrade just to chill with AMD again (dat new CCC!).

[Yes, I know, I'm part of the problem, but no kids (yet), well paying job, no country club membership, second mortgage isn't a go (yet), sorry guys I need my fix!]

I'm basically in the same boat but with kids. I get a new card, old card goes to the wife, her old card goes to the kids. Been looking into replacing her 580 since I had to bake in in Oct 2013 to revive it but even $200 290s haven't enticed me to buy since it runs everything she plays fine; wow, ffxiv, and l4d2. It's just loud being a blower.

 

[Elixer]

Hmmm... launches Nov. 15th?

Let's hope this is wrong.

 

[stahlhart]

Thread cleaned. Stay on topic.
-- stahlhart

 

[master_shake_]

Hmmm... launches Nov. 15th?



Let's hope this is wrong.

why didn't they just use tahiti again?

that gpu is still doing quite well for itself.

a 6gb's card witha nice clock bump would have been sweet

 

[Mondozei]

Remember that the delta color compression technology in GCN 1.2 is what allowed the R8-285 to have a bandwidth throughput which matched and sometimes even exceeded the 280X's 384 bit interface. This is especially helpful for ROPs, which are always memory bandwidth starved. Thus, just comparing like for like without taking architecture into account is a fool's errand.

We're talking about a 40% improvement over GCN 1.1

 

[Red Hawk]

why didn't they just use tahiti again?

that gpu is still doing quite well for itself.

a 6gb's card witha nice clock bump would have been sweet

Why not use Tahiti? Because Tahiti is old and out of date. Weaker tessellation, fewer ACEs, limited to DirectX Feature Level 11.3, older video engine, less efficient ROPs. Tahiti is not competitive with Nvidia's product stack when it comes to feature support and power efficiency. Tonga is at least somewhat better, it's what AMD needs on the market.

6 GB is unnecessary at Tahiti/Tonga's performance tier and would just add expense over Tonga's 4 GB. I'm not even sure a "nice clock bump" is possible with Tahiti, they probably squeezed as much as they could out of its clock with the 280 series.

 

[3DVagabond]

why didn't they just use tahiti again?

that gpu is still doing quite well for itself.

a 6gb's card witha nice clock bump would have been sweet

Tonga is more efficient and supports more features.

 

[Mondozei]

Flagships were at 500mm2+ since GT200 onward.

GTX280 was 576mm2 at 65nm
GTX480 was 529mm2 at 40nm
GTX580 was 520mm2 at 40nm

GTX680 was 294mm2 at 28nm oups
GTX980 is 398mm2 at 28nm oups again

That's only for NV. AMD didn't have a large die strategy until very recently.

This is why I laugh at the "die size = flagship" meme. And why I laugh at the whole panic that somehow tries to equate the 980 with the x60 cards of the past. It's nonsense.

Fiji is a large die, but if you look back to the cards AMD put out to compete with those NV cards, they had much smaller die area, yet were flagships. But of course this complicates the meme so many are fond of here.

NV in particular has been pushing the GK104/GK110 or the GM204 and GM200 strategy for a long time. I find it perfectly reasonable, because it takes time to get to a new arch and a new node. What matters is the generational performance over the previous comparable arch. This is why the 980 should be compared to the 780 and not the 780 Ti and why the 980 Ti should be compared to the 780 Ti and not the 780.

Thus, the generational performance leap is 35-40%, which is in line what was normal historically. There is no conspiracy here.

 

[master_shake_]

pretty sure at the 960 level tessellation isn't that big of a deal.

that's where this card fits.

and if TPU performance summaries mean anything the 280x obliterates the 285.

http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/23.html

and as far as a clock bump. my reference 7970's could max out the bar in MSI afterburner at 1125 and still had more headroom.

and yes 6gbs is probably too much but it's better too much than too little.

my 2 cents.

 

[Red Hawk]

pretty sure at the 960 level tessellation isn't that big of a deal.

that's where this card fits.

and if TPU performance summaries mean anything the 280x obliterates the 285.

http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/23.html

and as far as a clock bump. my reference 7970's could max out the bar in MSI afterburner at 1125 and still had more headroom.

and yes 6gbs is probably too much but it's better too much than too little.

my 2 cents.

It depends on the game. In a tessellation heavy game like Metro: Last Light, the 285 pulls ahead of the 280 and even beats the regular 7970. http://www.anandtech.com/bench/GPU14/852. At the 960 level, Tahiti's power consumption is unacceptable.

But of course the 280X obliterates the 285 on whole. It has more shaders and memory. It's not a fair comparison. Comparing the 285 to the 280/7950 is more appropriate because they have the same shader count, but still somewhat uneven since the 285 only ever had 2 GB.

Anecdotal evidence is pointless...maybe you had a golden chip. Maybe you were overvolting to a point that was unacceptable for manufacturers. Whatever the case, if AMD could reliably clock Tahiti that high, they would have already done so.

 

[IllogicalGlory]

If you want information on 7950/7970 overclocking headroom, look no further: http://www.overclock.net/t/1196856/official-amd-radeon-hd-7950-7970-7990-owners-thread

 

[AtenRa]

That's only for NV. AMD didn't have a large die strategy until very recently.

This is why I laugh at the "die size = flagship" meme. And why I laugh at the whole panic that somehow tries to equate the 980 with the x60 cards of the past. It's nonsense.

Fiji is a large die, but if you look back to the cards AMD put out to compete with those NV cards, they had much smaller die area, yet were flagships. But of course this complicates the meme so many are fond of here.

NV in particular has been pushing the GK104/GK110 or the GM204 and GM200 strategy for a long time. I find it perfectly reasonable, because it takes time to get to a new arch and a new node. What matters is the generational performance over the previous comparable arch. This is why the 980 should be compared to the 780 and not the 780 Ti and why the 980 Ti should be compared to the 780 Ti and not the 780.

Thus, the generational performance leap is 35-40%, which is in line what was normal historically. There is no conspiracy here.

AMD was only competing up to the upper middle segment the last years, they didnt directly compete against NVIDIAs 500mm2+ flagships.
NVIDIA was alone in the top performance segment with $500-600 price flagships.
AMD also increasing die size with each generation to finally reach almost 600mm2 with Fiji.

HD3870 = 192mm2 at 55nm
HD4870 = 256mm2 at 55nm
HD5870 = 334mm2 at 40nm
HD6970 = 389mm2 at 40nm
HD7970 = 352mm2 at 28nm
R9 290X = 439mm2 at 28nm
Fury X = 596mm2 at 28nm

NVIDIA changed its release strategy after the Fermi (GF200) disaster. That was the last time they released the Flagship chip first, from 28nm onward they released the middle class (GK204) chips first at flagship prices.
But when they finally released the real flagship (GK200), it was at a higher price/point instead of replacing the old middle class chip. So now we have a $500 price point middle class and $750 price point Flagships. It used to be $250-$300 for the middle class and $500-550 for the Flagships.

 

[master_shake_]

It depends on the game. In a tessellation heavy game like Metro: Last Light, the 285 pulls ahead of the 280 and even beats the regular 7970. http://www.anandtech.com/bench/GPU14/852. At the 960 level, Tahiti's power consumption is unacceptable.

But of course the 280X obliterates the 285 on whole. It has more shaders and memory. It's not a fair comparison. Comparing the 285 to the 280/7950 is more appropriate because they have the same shader count, but still somewhat uneven since the 285 only ever had 2 GB.

Anecdotal evidence is pointless...maybe you had a golden chip. Maybe you were overvolting to a point that was unacceptable for manufacturers. Whatever the case, if AMD could reliably clock Tahiti that high, they would have already done so.

power consumption

are you really going to miss that 27 watts at a typical gaming scenario?

http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/21.html

metro last light? is that the only game it's better in?

why would i compare a full tonga gpu to a cut down 7970 gpu?

that's just silly.

i did not have a golden chip.

the release clocks on a 7970 were very very conservative.

like the poster above me pointed out an 1125 core clock was modest compared to some peoples on OCN.

 

[tential]

power consumption

are you really going to miss that 27 watts at a typical gaming scenario?

http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/21.html

metro last light? is that the only game it's better in?

why would i compare a full tonga gpu to a cut down 7970 gpu?

that's just silly.

i did not have a golden chip.

the release clocks on a 7970 were very very conservative.

like the poster above me pointed out an 1125 core clock was modest compared to some peoples on OCN.

I hit 1100 easily on my 7950. I'd say ~1100 is a good range to use.

 

[AtenRa]

why would i compare a full tonga gpu to a cut down 7970 gpu?

that's just silly.

R9 285 and R9 380 is not a full Tonga, R9 380X is.

So,

You compare 285/380 to HD7950 GHz/280 and 380X to 7970GHz/280X

In latest games R9 380 is faster or equal to 280X

 

[master_shake_]

R9 285 and R9 380 is not a full Tonga, R9 380X is.

So,

You compare 285/380 to HD7950 GHz/280 and 380X to 7970GHz/280X

In latest games R9 380 is faster or equal to 280X

no it is not.

http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/23.html

the 280x is consistently better than the 380.

 

[AtenRa]

I hit 1100 easily on my 7950. I'd say ~1100 is a good range to use.

Yes, the vast majority of HD7950 will reach 1100MHz easily if they can raise the voltage.
My ASUS CU II 7950 does 1050MHz on default voltage.

 

[master_shake_]

Yes, the vast majority of HD7950 will reach 1100MHz easily if they can raise the voltage.
My ASUS CU II 7950 does 1050MHz on default voltage.

my 7970's did 1125 on stock voltage.

and i'm not alone.

look at ocn.

 

[AtenRa]

no it is not.

http://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Lightning/23.html

the 280x is consistently better than the 380.

I said in latest games, not in general.

Examples,

Also when Tessellation is being used like in Metro 285 is equal to 280X.

And R9 380 is a little bit faster than 285, that will make the R9 380X to be a very nice card but it should had come 6-12 months earlier.

Edit: Now that i see those games again, R9 380X should be equal to GTX780 on those games and faster in DX-12 games.

 

[AtenRa]

my 7970's did 1125 on stock voltage.

and i'm not alone.

look at ocn.

I guess I forgot how much it could OC

 

[tential]

Yes, the vast majority of HD7950 will reach 1100MHz easily if they can raise the voltage.
My ASUS CU II 7950 does 1050MHz on default voltage.

Why would you need to change voltage to hit 1050?

I really thought 1000-1100 range was stock voltage easily.

 

[crisium]

And why I laugh at the whole panic that somehow tries to equate the 980 with the x60 cards of the past. It's nonsense.

It's nonsense that you ignore that, mathematically, the 980 is the 560 of its generation. Almost exactly.

http://forums.anandtech.com/showpost.php?p=37817106&postcount=161

NV in particular has been pushing the GK104/GK110 or the GM204 and GM200 strategy for a long time. I find it perfectly reasonable, because it takes time to get to a new arch and a new node. What matters is the generational performance over the previous comparable arch. This is why the 980 should be compared to the 780 and not the 780 Ti and why the 980 Ti should be compared to the 780 Ti and not the 780.

Thus, the generational performance leap is 35-40%, which is in line what was normal historically.

Is this your retort of the factual knowledge that the 980 has virtually the same relative configuration as the 560? To compare it to the 700 series which began this name shifting?

460, 560, and 660 (compared to the temporary flagship 680) are are equal or overall closer in configuration to the flagship then the 980 is. But yes, comparing the 980 to the 700 series is more accurate because that is precisely where the X60 card really lost its value, and the 960 went even further.

If you want to shift your goal posts to the 700 series then even you have to admit that the 980 is a lower class GPU then in that generation. I already established that the 980 is 2/3 of the flagship, and the 780 is much, much closer to the 780 Ti with 80% of the Shaders, 100% of the ROPs, and the same buswidth albeit clocked slower. So by even your GTX 700 series standards the 980 is a joke; a user who bought a 780 2 years ago got closer to the flagship performance than a user who bought a 980 today.

There is no conspiracy here.

Of course not. It's simple marketing. Nvidia are a business and realize they can milk customers nearly twice that of 5 years ago.

 

[master_shake_]

I said in latest games, not in general.

Examples,

Also when Tessellation is being used like in Metro 285 is equal to 280X.
And R9 380 is a little bit faster than 285, that will make the R9 380X to be a very nice card but it should had come 6-12 months earlier.

Edit: Now that i see those games again, R9 380X should be equal to GTX780 on those games and faster in DX-12 games.

out of 22 games the 280x is bested by the 285 about 5 times at 1080p and about 2 times at >1080p

http://www.techpowerup.com/reviews/MSI/GTX_950_Gaming/

and the margins where the 280x is beat it is slim.

when the 280x beats the 285 it's a thorough beating.

edit 21 games

and if you want newer

the 280x is above the 380\960 for reference

also fallout 4 is brand new like less than 24 hours old

and we can easily place it above the 285 and below a 290x when comparing with the benchmarks i have supplied.

http://www.gamersnexus.net/game-bench/2177-fallout-4-pc-video-card-fps-benchmark-all-resolutions