[ PCGH ] Maxwell GTX 880 specifications leaked

[KaRLiToS]

Translation: https://translate.google.ca/transla...880-Leak-Spezifikationen-Maxwell-GPU-1116952/

Original: http://www.pcgameshardware.de/Nvidi...880-Leak-Spezifikationen-Maxwell-GPU-1116952/

The Geforce GTX 880 is clearly indicated in 20 nm and is said to have 7.9 billion transistors, whereas the GK110 still comes currently to 7.1 billion. As with the current Maxwell representatives of the front-end to have been greatly expanded by reducing the back-end. So 3200 shader and 200 texture units are called, which would correspond to 25 SMM (Maxwell shader processors). The raster output stages, however, would be reduced to 40 in number. In a GPU clock of 900/950 MHz (base / Boost) a single-precision peak performance of 5.7 TFLOPS would be reached, an increase of almost seven per cent compared to the GTX 780 Ti (the values ​​given the GK110 offshoot vote not consistent with our values).

 

[Silverforce11]

That's the mid-range on 20nm that we are expecting. Not surprised at all but it will be a long time getting there with the current TSMC situation.

256bit bus mid-range selling for the same as a 780ti ... hilarious times ahead.

 

[Grooveriding]

If true, meh. It's gtx 680 redux. Mid-range sold as high end and still not even getting that till 2015, so maybe the real high end geforce Christmas of 2015?

Sounds great :| I hope this is bogus.

 

[SolMiester]

That's the mid-range on 20nm that we are expecting. Not surprised at all but it will be a long time getting there with the current TSMC situation.

256bit bus mid-range selling for the same as a 780ti ... hilarious times ahead.

Going backwards on the 780's 384bus?....doesnt make sense!

 

[Insomniator]

Looks bogus... why would the bus be so low and the price and shaders so high.

 

[Rvenger]

Can't be real. That would have to be the GTX 860ti or similar.

 

[Silverforce11]

Going backwards on the 780's 384bus?....doesnt make sense!

It's a repeat of the GK104 aka GM104 mid-range die. The node jump allows a ton of shaders at less die space so the 256 bit bus is spot on.

It just means NV is doing the same thing with Maxwell, shove out a mid-range part at top-range prices. Then later follow up with a full maxwell die, aka GM110 etc. I'm pretty sure they need all their full Maxwell dies for Teslas first and foremost.

 

[Silverforce11]

Can't be real. That would have to be the GTX 860ti or similar.

You guys have a short memory.

GTX680, mid-range die at top-range price at the time, aka the GTX580.

If it worked then, why wouldn't they do it again, mid-range maxwell selling at 780ti prices. Yay, profits!

 

[Fire&Blood]

Looks bogus... why would the bus be so low and the price and shaders so high.

Unfortunately, that doesn't lead me to think it's bogus nowadays. If anything, it's the same thing they did with this generation. They'll launch this 880 wait a while and then launch the real 880 priced even higher.

 

[toyota]

specs make very little sense...

its listing effective memory at 7400 when the fastest gddr5 is rated for 7000.

going back to 256 bit bus while having more cores than the 780ti would choke it.

Nvidia ties its ROPs to memory controllers so how can the 256 bit bus have 40 ROPs?

its hardly any bigger of a gpu than the 780ti and the clocks are pretty low yet the power consumption is only 20 watts less with 20nm and efficient Maxwell architecture?

 

[tviceman]

Going backwards on the 780's 384bus?....doesnt make sense!

The 680 "went backwards" as well, but was still faster in the end.

Looks bogus... why would the bus be so low and the price and shaders so high.

I didn't see where price was mentioned; but I think I'm sadly expecting prices to start higher than the $500-550 mark we saw with 28nm. Anyways, if this is GK104's successor, then 256-bit bus sounds right.

The core count doesn't sound unrealistic, either. GK104 had 4x has many cores as GK107. So if this chip is GM204, and it ends up with 5x as many cores as GM107, that is within the ballpark difference that the Kepler chips had.

Also, as we see, GM107 performs quite well even with a castrated 128-bit bus and paltry 86.4 gb/s bandwidth.

Can't be real. That would have to be the GTX 860ti or similar.

GTX 680, anyone?

 

[tviceman]

specs make very little sense. its listing effective memory at 7400 when the fastest gddr5 is rated for 7000. going back to 256 bit bus while having more cores than the 780ti would choke it. Nvidia ties its ROPs to memory controllers so how can the 256 bit bus have 40 ROPs? and its hardly any bigger of a gpu than the 780ti and the clocks are pretty low yet the power consumption is only 20 watts less with 20nm and efficient Haswell architecture?

Agreed, some of the specs look off/suspect. Including the power draw and clock speed. Keep in mind, though, that GK107's bus is 128-bit and bandwidth is a low 86.4, but it had no problem running benchmark circles around the gtx650 TI, which had the same bus and bandwidth. So it's not out of the realm of possibility that GM204 will have less bandwidth than GK110, yet outperform it.

Also, the core count is within the realm of possibility. GK104 had 4x as many cores as GK107, so GM204 should have at least 4x as many cores as GM107.

 

[Phynaz]

40 ROPS? Disappointing if true.

 

[SlowSpyder]

specs make very little sense...

its listing effective memory at 7400 when the fastest gddr5 is rated for 7000.

going back to 256 bit bus while having more cores than the 780ti would choke it.

Nvidia ties its ROPs to memory controllers so how can the 256 bit bus have 40 ROPs?

its hardly any bigger of a gpu than the 780ti and the clocks are pretty low yet the power consumption is only 20 watts less with 20nm and efficient Haswell architecture?

A lot of people said the same thing about the ROP's and 384bit memory bus on the 7970 before it launched. Just saying. :whiste: With that said, I don't trust these specs.

 

[toyota]

Agreed, some of the specs look off/suspect. Including the power draw and clock speed. Keep in mind, though, that GK107's bus is 128-bit and bandwidth is a low 86.4, but it had no problem running benchmark circles around the gtx650 TI, which had the same bus and bandwidth. So it's not out of the realm of possibility that GM204 will have less bandwidth than GK110, yet outperform it.

Also, the core count is within the realm of possibility. GK104 had 4x as many cores as GK107, so GM204 should have at least 4x as many cores as GM107.

it makes no sense for a 256bit card with low clocks and efficient architecture on 20nm to have a 230 watt tdp.

 

[TreVader]

I would expect at least a 320bit bus for 40 ROPs, but it's possible nvidia didn't think they needed the bandwidth.



If this is accurate the 880 is going to be a weird card that is faster than the 780ti in benchmarks and shader intensive games but gets killed at high resolutions and w/ AA.

 

[OCGuy]

LOL...clock speeds are set and a year + naming convention set + year away?

Sure....

 

[toyota]

A lot of people said the same thing about the ROP's and 384bit memory bus on the 7970 before it launched. Just saying. :whiste: With that said, I don't trust these specs.

yes they may no longer be tied to memory controller on the upcoming gpus but on the only Maxwell gpu that we have now they are as for as I know.

 

[SolMiester]

The 680 "went backwards" as well, but was still faster in the end.

the 680 did, but then 780 went back up as a new series, so does that mean 980 will be full monty 384bus?....while 880 only gets 256?...perhaps it will if NV rolls out the gpus like they did with Kepler....but i dont see it myself....

 

[Silverforce11]

LOL...clock speeds are set and a year + naming convention set + year away?

Sure....

Targets more than set, especially the price. It will inevitably depend on how competitive AMD is with VI.

 

[TreVader]

My guess is that the clock speeds are going to be much higher than 950 boost. If the 750ti is any indication, they should be clocking it closer to 1100-1200

 

[Silverforce11]

My guess is that the clock speeds are going to be much higher than 950 boost. If the 750ti is any indication, they should be clocking it closer to 1100-1200

Don't forget its on a very mature 28nm process.

Given TSMC history, early 20nm will be... meh.

 

[rgallant]

not to seem likes it's thread crapping

but after the stepped release of GK's , I'll be waiting till 2016 ish when most of the sku's are hinted at before buying anything MAXWELL , which will be roughly when my sli cards run out of the 3gb of vram per nv's planned marketing.
or move to the red team.

 

[Techhog]

If both these specs and the Pirate Islands specs are correct, it would mean that AMD wins until the true Maxwell high-end release, wouldn't it?

 

[Silverforce11]

If bothese specs and the Pirate Islands specs are correct, it would mean that AMD wins until the true Maxwell high-end release, wouldn't it?

We don't know how well the new architectures perform so its a guessing game.