Nvidia GTX 960 and AMD R9 285X – The Specs, the Performance, the Price and the Story

Boomstick777 · 2 Oct 2014 at 00:21

Today we’re bringing you an update to AMD’s elusive and yet unreleased R9 285X graphics card and Nvidia’s much anticipated GTX 960. It was revealed that the Tonga GPU powering the R9 285 non-X actually has more than just cut down compute units. In addition to the disabled 256 GCN cores, the GPU also has one third of it’s memory interface locked. The fully unlocked Tonga XT GPU which is rumored to power the R9 285X has been confirmed to have a 384bit interface. Just like its older sibling Tahiti.

Read more: http://wccftech.com/nvidia-gtx-960-amd-r9-285x-specs-performance-price/#ixzz3EwFaA1z4

CAT-THE-FIFTH · 2 Oct 2014 at 00:30

So basically we should have in-between GTX770/R9 280X and GTX780/R9 290 performance at under £200??

humbug · 2 Oct 2014 at 00:32

A 2048 Shader 384Bit Tonga may render the R9 290 obsolete.

What's next?

wazza300 · 2 Oct 2014 at 00:34

if only there was a way to unlock the extra goodness

Boomstick777 · 2 Oct 2014 at 00:36

humbug said:
A 2048 Shader 384Bit Tonga may render the R9 290 obsolete.

Isn't the 285 weaker than the 280X?

Boomstick777 · 2 Oct 2014 at 00:39

Boomstick777 said:
Isn't the 285 weaker than the 280X?

Yeah it is weaker, wouldn't expect to much from the 285X if it does get a release..

Quite slow as performance has not improved
Not very power efficient
Could be quieter in idle and gaming
Elpida memory not optimal for overclocking
High Blu-ray power consumption

CAT-THE-FIFTH · 2 Oct 2014 at 00:46

Boomstick777 said:
Yeah it is weaker, wouldn't expect to much from the 285X if it does get a release..

Yes,but with a lower max clockspeed than an R9 280 and R9 280X and lower bandwidth the R9 285 ends up still being faster than a R9 280 and within 10% of the R9 280X which has more shaders and more bandwidth.

So what do you think a 15% increase in shaders,a 50% increase in bandwidth and a 50% increase in ROPs,and better tessellation than a R9 290 is going to do especially if it runs at a higher clockspeed??

Its going to be easily faster than a R9 280X,especially when effective bandwidth is more due to compression.

Personally I think the GTX960 and the R9 285X would trade blows.

Now,the only thing is to see whether the R9 285X dies would be better binned than the R9 285,ie, improved performance/watt.

I think the GTX960 would be ahead in this regard.

Boomstick777 · 2 Oct 2014 at 00:53

CAT-THE-FIFTH said:
Yes,but with a lower max clockspeed than an R9 280 and R9 280X and lower bandwidth the R9 285 ends up still being faster than a R9 280 and within 10% of the R9 280X which has more shaders and more bandwidth.

So what do you think a 15% increase in shaders,a 50% increase in bandwidth and a 50% increase in ROPs,and better tessellation than a R9 290 is going to do especially if it runs at a higher clockspeed??

Its going to be easily faster than a R9 280X,especially when effective bandwidth is more due to compression.

Personally I think the GTX960 and the R9 285X would trade blows.

Now,the only thing is to see whether the R9 285X dies would be better binned than the R9 285,ie, improved performance/watt.

In that review they say the power use is still high and performance isn't improved, just saying. I wouldn't expect to much from these. HBM + 20nm will be when AMD get interesting again. Enjoy Nvidia until then, the GTX 960 looks like it could be a cracking lil card, with really low power use. Nvidia are killing it atm. Next year 20nm + HBM from the red team.

Looking forward to seeing APU's with HBM as well, they have the core performance once you remove that bandwidth bottlneck they could be awesome, espcially if AMD work it so that it can be used as cache for the CPU as well.

CAT-THE-FIFTH · 2 Oct 2014 at 01:00

Boomstick777 said:
In that review they say the power use is still high and performance isn't improved, just saying. I wouldn't expect to much from these. HBM + 20nm will be when AMD get interesting again. Enjoy Nvidia until then, the GTX 960 looks like it could be a cracking lil card, with really low power use. Nvidia are killing it atm. Next year 20nm + HBM from the red team.

Looking forward to seeing APU's with HBM as well, they have the core performance once you remove that bandiwthc bottlenck they could be awesome, espcially if AMD work it so that it can be used as cache for the CPU as well.

Yes,the R9 285 has worse specs than a R9 280. Lower max clockspeed,33% less bandwidth,same number of ROPs but overall tends to be a bit faster than a R9 280.

The R9 285 will have a 50% increase in ROPS over the R9 280,R9 280X and R9 285X,and more effective bandwith over all three. Tesellation is better overall than the R9 290.

One of the reasons the R9 290 and R9 290X are faster than the R9 280X,is since Tahiti was bottlenecked by having only 32 ROPs.

Edit!!

Look at the AT review:

http://www.anandtech.com/show/8460/amd-radeon-r9-285-review/7

Rroff · 2 Oct 2014 at 01:03

Boomstick777 said:
the GTX 960 looks like it could be a cracking lil card, with really low power use. Nvidia are killing it atm. Next year 20nm + HBM from the red team.

In addition to refinements to the 28nm process the Maxwell architecture has a lot of tweaks, some of them quite clever, to reduce how much of the core is redundantly active for any given operation - if AMD haven't mastered something similar (and I have no idea on that front) they don't have a chance of competing with Maxwell on 28nm.

CAT-THE-FIFTH · 2 Oct 2014 at 01:06

Rroff said:
In addition to refinements to the 28nm process the Maxwell architecture has a lot of tweaks, some of them quite clever, to reduce how much of the core is redundantly active for any given operation - if AMD haven't mastered something similar (and I have no idea on that front) they don't have a chance of competing with Maxwell on 28nm.

I get the impression,AMD is trying to get onto 20NM as quickly as possible,so personally I don't see it happening anytime soon. Nvidia is going to lead in power consumption on 28NM.

humbug · 2 Oct 2014 at 01:08

Depends on where you go and what they test.

TPU also fail to test the improvements Tonga brings.

Its not as fast as the 280X all the time, but it is faster than the 7950 Boost which is higher clocked and has a wider bus.

The 280X has 15% more Shaders, is 10% higher clocked and has a 50% wider bus (all tested on 14.7) TPU tested all but the 285 on 14.7. the 285 was tested on 14.30, which was a pre-release tester based on 14.6

CAT-THE-FIFTH · 2 Oct 2014 at 01:15

Regarding power saving tech for AMD CPUs and GPUs,this was posted on AT during the Mullins launch:

http://images.anandtech.com/doci/7974/Screen-Shot-2014-04-29-at-1.08.08-AM_575px.jpg

That was in April this year.

The question is whether this will be implemented on any of the 28NM AMD GPUs or will it be for their 20NM ones??

Looking at some of the tech,it does look similar to what is in Maxwell.

humbug · 2 Oct 2014 at 01:29

CAT-THE-FIFTH said:
Regarding power saving tech for AMD CPUs and GPUs,this was posted on AT during the Mullins launch:

http://images.anandtech.com/doci/7974/Screen-Shot-2014-04-29-at-1.08.08-AM_575px.jpg

That was in April this year.

The question is whether this will be implemented on any of the 28NM AMD GPUs or will it be for their 20NM ones??

Looking at some of the tech,it does look similar to what is in Maxwell.

I doubt those will be on 28nm, but then i don't think they expected Nvidia to implement those technologies on 28nm, they may now be forced to act.

The R9 290 is 27% faster than the 280X according to TPU.

If the 285X gets to a few % of that i think it may spell the end of the Hawaii GPU's to be replaced by Tonga.

R9 295 Tonga: 2560 SP's @ 1000Mhz, 48 ROP's, 384Bit

R9 295X Tonga 2816 SP's @ 1050Mhz, 64 ROP's, 512Bit

Feburay/March 2015:
R9 380X Fiji 20nm: 3238 SP's @ ????Mhz, 72 ROP's, HBM

April/May 2015:
R9 390X Bermuda 20nm: 4224 SP's ?????Mhz, 96 ROP's, HBM

bru · 2 Oct 2014 at 01:37

I see this stems from the same leak as the last 384bit rumour, so i would say it is far from confirmed.

I may be wrong and I will be only too pleased if I am,but I'm not holding out much hope.

humbug · 2 Oct 2014 at 01:40

bru said:
I see this stems from the same leak as the last 384bit rumour, so i would say it is far from confirmed.

I may be wrong and I will be only too pleased if I am,but I'm not holding out much hope.

Its all rumour, its fun to speculate. what's a GPU enthusiast to do with his time other than speculate on future GPU's?

Use the one he has to play games on?

CAT-THE-FIFTH · 2 Oct 2014 at 01:46

Regarding Tonga it was hinted strongly by sites like Anandtech and TR that it was a 2048 shader chip,and the TR reviewer hints that there is a 384 bit memory controller.

Plus Anandtech says this:

but it’s notable since at an architectural level Tahiti had to use a memory crossbar between the ROPs and memory bus due to their mismatched size (each block of 4 ROPs wants to be paired with a 32bit memory channel)

Tahiti had a less than optimal setup which was changed in Hawaii(512 bit memory controller and 64 ROPs). So a 384 bit memory controller in Tonga will be paired with 48 ROPs.

Edit!!

This is the quote by the TR reviewer:

Surely we'll see a Radeon R9 285X card eventually with a fully-enabled Tonga GPU clocked at 1GHz or better. If I were betting, I'd put my money on that card having a 384-bit path to memory.

So this is why I think a fully enabled Tonga will have:
1.)2048 shaders
2.)a 384 bit memory controller
3.)48 ROPs
4.)Have a higher clockspeed than a R9 280 and R9 285 and comparable to that of an R9 280X or higher

Disco_P · 2 Oct 2014 at 01:52

Wasn't 285X confirmed as not being a thing by AMD? I know there was a twitter tweet by one of the AMD guys + Matt may have said something in the forums iirc.

humbug · 2 Oct 2014 at 01:54

CAT-THE-FIFTH said:
Regarding Tonga it was hinted strongly by sites like Anandtech and TR that it was a 2048 shader chip,and the TR reviewer hints that there is a 384 bit memory controller.

Plus Anandtech says this:

Tahiti had a less than optimal setup which was changed in Hawaii(512 bit memory controller and 64 ROPs). So a 384 bit memory controller in Tonga will be paired with 48 ROPs.

Right, its because Hawaii has 100% ROP's scaling to go with its +27% SP's is what makes it 35% (130% scaling) faster the 7970 GE, or at least the 100% part, the 30% over full scaling is probably down to a 25% wider bus.

If your just increasing the SP's by 1 you get 0.5 scaling, the other 0.5 comes from ROP's or Memory bandwidth.

CAT-THE-FIFTH · 2 Oct 2014 at 01:55

Disco_P said:
Wasn't 285X confirmed as not being a thing by AMD? I know there was a twitter tweet by one of the AMD guys + Matt may have said something in the forums iirc.

Yeah,I remember that!

Maybe it was canned and AMD is trying to pull the rumoured 20NM R9 380X launch forward as much as they can??