• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

NVIDIA Deploys GP104 GPU for GDDR5X version of GeForce 1060

Raevenlord

News Editor
Joined
Aug 12, 2016
Messages
3,755 (1.18/day)
Location
Portugal
System Name The Ryzening
Processor AMD Ryzen 9 5900X
Motherboard MSI X570 MAG TOMAHAWK
Cooling Lian Li Galahad 360mm AIO
Memory 32 GB G.Skill Trident Z F4-3733 (4x 8 GB)
Video Card(s) Gigabyte RTX 3070 Ti
Storage Boot: Transcend MTE220S 2TB, Kintson A2000 1TB, Seagate Firewolf Pro 14 TB
Display(s) Acer Nitro VG270UP (1440p 144 Hz IPS)
Case Lian Li O11DX Dynamic White
Audio Device(s) iFi Audio Zen DAC
Power Supply Seasonic Focus+ 750 W
Mouse Cooler Master Masterkeys Lite L
Keyboard Cooler Master Masterkeys Lite L
Software Windows 10 x64
NVIDIA has just shown us one of the most ingenious ways of creating new custom, competitive SKUs for the midrange market without spending any additional amounts of money on R&D, wiring, or memory controller work: just reuse the chips that already have that work done. This is the case for NVIDIA's new GTX 1060 GDDR5X graphics card, which the company has "designed" to further fill in the gaps on its midrange offerings against a revamped Radeon RX 590.





Instead of reworking a purpose-built memory controller solution compatible with the GP106 GPU, the company has gone and carved the SKU from its existing GP104 silicon - which already supports the GDDR5X memory subsystem due to its implementation on the GTX 1080 (GP104) and 1080 Ti (GP102) graphics cards. A smart usage of GP104 inventories - which have been superseded by NVIDIA's new RTX 20-series in the high end - or of very defective dies (remember the GTX 1060 has half the shaders, at 1280, compared to the GTX 1080's 2560). This decision by NVIDIA could also go some way in explaining dwindling inventories and increasing pricing of GTX 1080 graphics cards, as chips that could have been used for that SKU are (possibly) being used for the new GTX 1060.



The discovery came courtesy of a teardown on iGame's GTX 1060 U-TOP V2, which features a triple-fan cooling solution, 2x 8-pin power connectors, and an 8+2 phase power delivery design, via Taobao. Apparently, even the SLI fingers remain on the card, which if you'll remember, never where supported on NVIDIA's GTX 1060 - a result of iGame's decision to simply reuse their PCB design for the usually much more powerful, GP104-based GTX 1080.

View at TechPowerUp Main Site
 
so technically speaking the standard 1070 has been cannibalized in favour of this card now?? regular 1070s were GP104's too werent they?

this is like an entirely different card or revision. Its more like a GTX1060 2.0
 
so technically speaking the standard 1070 has been cannibalized in favour of this card now?? regular 1070s were GP104's too werent they?

this is like an entirely different card or revision. Its more like a GTX1060 2.0
not exactly, nvidia has defective 1070 that are sold as 1060
 
So when do we see reviews for this card the AMD 590?
 
the company has gone and carved the SKU from its existing GP104 silicon - which already supports the GDDR5X memory subsystem due to its implementation on the GTX 1080 and 1080 Ti graphics cards.
Wait....I thought 1080Ti was GP102? Did you perhaps mean 1070 and 1080?
 
That a blushing write-up, for saying they're serving the same old Lemonade at a higher price.
Exactly this, you would think it's Christmas come early.

Though being English i honestly read most as a sarcastic jab , who knows.

Maybe ,no ,no shader unlock possible since their bios is a hard nut t crack these day's.
 
Maybe ,no ,no shader unlock possible
Given Nvidia has been for some amount years excellent at granularity fusing off sections of the chip I'd say that's not in these cards. Best you might get is good OC'n but that might come at the expense of power, depending on what sections and how they slice-n-dice to kill half of that large chip.


Can't figure out how to quote Raevenlord "This decision by NVIDIa could also go some way in explaining dwindling inventories and increasing pricing of GTX 1080 graphics cards, as chips that could have been used for that SKU could be used for the new GTX 1060."

How does gelding perfectly good GP104's that could be sold as GTX 1080's (or 1070-Ti) be a logical use? Other than they want to stop selling them so to drive folks to by into RXT... And between accomplishing that goal, and unloading what probably Bandini Mountain of chips that could not be used for a lowly GTX 1070 they give use this?
 
Last edited:
So basically this is GTX 1069
 
So basically this is GTX 1069
No that naming convention would have been better used for that inbred 9Gbps model.

Now, had the made a card that had 9Gb of GDDR5X it would've offer that cool WOW factor, but it won't have been anymore of a benefit than this.
 
Last edited:
sounds like another design accomplished by rolling the dice, "hmm what will we do now the the 1060?"
 
How does gelding perfectly good GP104's that could be sold as GTX 1080's (or 1070-Ti) be a logical use? Other than they want to stop selling them so to drive folks to by into RXT... And between accomplishing that goal, and unloading what probably Bandini Mountain of chips that could not be used for a lowly GTX 1070 they give use this?

You're forgetting the fact that 104 wasn't even a high end part to begin with. It's just 310 mm^2, a couple of years ago that would have been the xx60 part from the very start.
 
You're forgetting the fact that 104 wasn't even a high end part to begin with. It's just 310 mm^2, a couple of years ago that would have been the xx60 part from the very start.
True forgot about that. Was definitely a big debate about that at the time.
 
And for every 100 cards1060 they prolly sell 5 of 1080.
 
So basically this is GTX 1069
IMO it should be called 1060 Ti or 1065. But we all know Nvidia, let's just make so damn many versions under the same model number, just to confuse those customers who don't know much about these.

I wouldn't be surprised if they do this same later with RTX 2070, just use defective TU104 chips and it's good to go.
 
You're forgetting the fact that 104 wasn't even a high end part to begin with. It's just 310 mm^2, a couple of years ago that would have been the xx60 part from the very start.

Your argument is simply laughable.
The level of performance relative to the full lineup makes a GPU low-end, mid-range or high-end, not the fucking die size.
based on your silly argument the RTX 2070 is a higher end card than GTX 1080 which isn't the case.
 
Your argument is simply laughable.

I'm sure it is when you ain't got a clue how this industry operates and how companies build their product stack based on the size and transistor count of the chips. Neither do I expect you to understand that based on your rather ... simplistic way of viewing things.

based on your silly argument the RTX 2070 is a higher end card than GTX 1080 which isn't the case.

It is indeed a higher end card, from all aspects. Including die size. Whether that translates well into pricing and performance matters little.

Hence why all of you are baffled as to why does Nvidia chooses to sell a product based on a "high end" part in a lower end segment when in fact this makes perfect sense from within their product stack and its hierarchy.
 
Last edited:
You're forgetting the fact that 104 wasn't even a high end part to begin with. It's just 310 mm^2, a couple of years ago that would have been the xx60 part from the very start.
Well, your statement is only partially right. The size of the die is not important. The GP104 is serving both high end (GTX 1080) and upper-mid range (1070 & 1070 Ti). The Gx104 has for the last several families/ generations served the x70 and x80, not the 60’s.
 
Last edited:
  • Like
Reactions: M2B
I personally think that they should have given it the P4000's core config (1792 vs 1280/1152) but that might have competed too much with the 1070
 
Reminds me of the GTX 465.
 
Yeah no, something ain't adding up.

Are we really supposed to believe that after 2 years of producing Pascal, TSMC's yields have gone down, not up, to the point where they only are able to produce half-working GP104 chips? Nope.

So the only other option is that NVIDIA has killed off GTX 1070/1070 Ti/1080 production in favour of seriously neutering GP104 and putting it into this "GTX 1060 GDDR5X". But that doesn't make sense either: why take a chip that in its lowest-end configuration (GTX 1070) was paired with cheaper GDDR5 memory, and cut it down by half and pair it with more expensive GDDR5X? Why not just chop GP104 down a little less - say, to 1440 shaders - and leave it coupled with GDDR5, to create a GTX 1070 "lite" that would be able to convincingly beat the RX 590?

The only possibility I can imagine is that there is some f**kery with GP104's eight memory controllers that means they can't work with non-multiples-of-4GB GDDR5, but aren't limited in that way with GDDR5X - although that seems unlikely, since those controllers couple into 4 sets of 2 each for GDDR5 mode (4x 64-bit) and 8 sets of 1 each (8x 32-bit) for GDDR5X, so cutting the memory amount down to 6GB would imply 3x2 for GDDR5 which should work just fine.

More information is needed here.
 
Back
Top