Friday, October 19th 2018

NVIDIA Rushes in GTX 1060 with GDDR5X to Counter AMD Radeon RX 590 Threat

AMD is giving final touches to its Radeon RX 590 graphics card, which is rumored to be based on an efficient new rendition of the "Polaris" silicon, which could disturb NVIDIA's product lineup between the GTX 1060 series and the GTX 1070, as its new RTX 2060 series is nowhere in sight. In a bid to thwart this threat, NVIDIA is preparing a variant of the GeForce GTX 1060 with faster GDDR5X memory.

The current GTX 1060 6 GB is endowed with 8 Gbps GDDR5 memory, which at its 192-bit bus width works out to a memory bandwidth of 192 GB/s. NVIDIA had attempted to improve its competitive position once, by creating a shortlived sub-variant of this SKU with 9 Gbps GDDR5 memory (211 GB/s). Switching to 10 Gbps GDDR5X memory would give the chip 240 GB/s memory bandwidth, and 11 Gbps (unlikely because expensive), would yield 264 GB/s. With the GP106 silicon maxed out, it's also possible the new GTX 1060 could be based on a heavily cut down GP104, possibly even with 192-bit memory, which explains GDDR5X memory.
Source: NVIDIA
Add your own comment

98 Comments on NVIDIA Rushes in GTX 1060 with GDDR5X to Counter AMD Radeon RX 590 Threat

#26
ShurikN
Vya DomusIt would require a new die with a different memory controller. Unlikely Nvidia would bother with that.

Or they could use a severly cut down GP104 . That's not as crazy as it sounds , that is after all nothing more than a mid-range GPU size wise.
But what's the point if 2060 is around the corner.
OR
Does this show that there wont be any mid/low tier Turings for the foreseeable future.

Another thing that caught my attention is the "590 counter"
Because if look at regular 1060 and 580, they mostly trade blows depending of the game.
The 590 should be 10% faster at stock if you take those recent leaks into account, so a regular 1060 with a bit of a mem bump will not really counter it.
So it's either not a GP106 or they have a lot of extra inventory from the crypto crash that was returned to them a couple of months ago, and they wanna get rid of it fast.

Food for thought I guess.
Posted on Reply
#27
T4C Fantasy
CPU & GPU DB Maintainer
ShurikNBut what's the point if 2060 is around the corner.
OR
Does this show that there wont be any mid/low tier Turings for the foreseeable future.

Another thing that caught my attention is the "590 counter"
Because if look at regular 1060 and 580, they mostly trade blows depending of the game.
The 590 should be 10% faster at stock if you take those recent leaks into account, so a regular 1060 with a bit of a mem bump will not really counter it.
So it's either not a GP106 or they have a lot of extra inventory from the crypto crash that was returned to them a couple of months ago, and they wanna get rid of it fast.

Food for thought I guess.
there is literally already many many in the thousands of GP104 1060s people dont know they have, it was released secretely

www.techpowerup.com/gpuz/details/3zs4w

www.techpowerup.com/gpuz/details/a7pvz
Posted on Reply
#28
ppn
Now only if this is 1060 Ti with 1920 Cuda 192 bit bus D5X.

Gimping more than 1/2 GP104 at 1152 CUDA. Why do that and not release a 660Ti same Cuda count as 670.
Posted on Reply
#29
WikiFM
They should increase core counts (if on GP104). If what they want is get rid of GP106 then increase core clocks, 1060 has a lot of OC potential. Let's say a lot of factory OC, plus maybe again 9 Gbps memory. And it should be called 1060 Ti to make a name difference finally, or 2050/Ti if a equivalent die is not coming in the next 8 months.

Anyway I agree with most of you, regarding increasing memory bandwidth, since 580 and 590 will be just core clocks increases, but memory bandwidth stays the same.
Posted on Reply
#30
Vayra86
AMD: here's a refresh of a refresh! Its faster and even on a smaller node, some optimizations, its great!

Nvidia: here's a two year old card with some new memory, now pls go sit in your corner TY
Posted on Reply
#31
ppn
RX590 It is not on a smaller node, same 14, less likely 12nm and glo fo doesn't use 7,5T libraries. 12nm zen is 9T, same as 14nm.
Posted on Reply
#32
windwhirl
T4C Fantasythe rumor is this is a 1060 6 GB with GP104 on a 1080 PCB

here is the entire stack of cards
Vayra86AMD: here's a refresh of a refresh! Its faster and even on a smaller node, some optimizations, its great!

Nvidia: here's a two year old card with some new memory, now pls go sit in your corner TY
The rebranding & re-spec'ing game, unlimited edition...

Honestly, Nvidia's move seems excessive. Why make a new card when they could just lower the prices? Does this have anything to do with the 300k cards they had in storage or something?
Posted on Reply
#33
Vayra86
ppnRX590 It is not on a smaller node, same 14, less likely 12nm and glo fo doesn't use 7,5T libraries. 12nm zen is 9T, same as 14nm.
Oh? I thought Polaris would go smaller?
Posted on Reply
#35
Vya Domus
Vayra86AMD: here's a refresh of a refresh! Its faster and even on a smaller node, some optimizations, its great!

Nvidia: here's a two year old card with some new memory, now pls go sit in your corner TY
Arguably, the same GPU on a newer node is somewhat more interesting than the latter.
Posted on Reply
#36
Valantar
dj-electricNVIDIA has a clear chance to rebrand such card. Probably not RTX\GTX 2060, but possibly a 2050 Ti moniker.
I also wouldn't be surprised if this product will end up with a 299$ price tag
Ugh. 2 years later, launching a barely refreshed product for noticeably more than MSRP of the original? I might be naive, but that seems too low even for 2018-era "Hey buy our new $1200 GPUs! They're a bit faster than our two-year-old $700 GPUs!" Nvidia.

Still, this whole thing sounds weird. Just the concept of Nvidia GP104 dice more than 50% disabled (rather than launch a rebrand and sell the same hardware for more money) is baffling to me. Wouldn't they ultimately make more money by launching a refresh/rebrand (or just cutting MSRPs), giving 1070 performance at 1060 prices, and so on? There can't possibly be that many GP104 dice with that many defects. Not at this point.
Posted on Reply
#37
Fluffmeister
And to think, AMD have made more effort (apparently) going from the 580 to the 590, than they did going from the 480 to the 580.
Posted on Reply
#38
newtekie1
Semi-Retired Folder
ShurikNUsually when you buy a card that has different amount of memory, you assume the GPU is the same.
Well that is your mistake for assuming. The fact is the cards are distinguishable.
ShurikNMX150 has 2 variants (or was it the mx130 hmmm). One is the default (being advertised as such) and its, in name, indistinguishable slower brother, pushed into the market in silence. You wont know which one you are getting until you buy the laptop and run GPUZ, at which point it's too late.
It was the MX150, and the only difference was the max TDP. The lower TDP variant was only used in ulta-thin laptops, and IIRC it still thermal throttled, so it isn't like putting the higher TDP variant in the laptop would make a difference. The issue here is more with the designs of the laptops, not really nVidia.
ShurikNAs for AMD side of story, we have the 550 with different specs in the same name and I believe a 460. The new 580 for the Chinese market has it's core count in its name on the Chinese website.
Don't forget the RX 560. AMD released a different version with less shaders with no change in the name at all. The original RX 560 had 1024 shaders, then AMD released another version of the RX 560 with no name change at all, no way at all to distinguish between the versions that had 896 Shaders. No core count in the name, no added SE or whatever, just RX 560.
cdawallThe gtx 460 was a bit funky v1 was 336 cores and 256 bit the v2 was 336 cores and 192 bit and the SE was 288 cores and 256 bit plus the 768mb model that was 336 cores and 192 bit lol.
Funky, yes, but still distinguishable. They added the V2 to the name, as well as an SE for the other version as well as the memory amount being part of the model name as well, distinguishing between the different versions.
Posted on Reply
#39
GoldenX
Do we know anything about the 590?
Posted on Reply
#40
John Naylor
Do people actually think that this design hasn't been sitting on the shelf for 3 years ? No one remembers the 780 Ti dropping a week later after the 290 / 290x dropped ?
Posted on Reply
#41
SOUTHPAWMIKE
I really feel like Nvidia needs to change up it's naming convention when it comes to all of these cards with different configurations that are referred to by the same model signifier. For example, thinking about the controversy around different 1060s with differing amounts of VRAM, the 3GB version could be called the GTX1063, and the 6 GB version the GTX1066. Maybe there could be different "elements" stapled on (like how Ti is titanium on the periodic table of elements) with something like Au (gold) for GDDR5 and W (tungsten) for GDDR5X. Just spitballing.
Posted on Reply
#42
sergionography
Lame, unless they sell it for 150-200usd
Meanwhile what they need is gp104 at 300usd. If rtx 2070 is about 500 for a 445mm die size, id expect they can afford to make a profit selling a 1080 equivalent for 300usd being that it's 314mm in size. Wishful thinking i know
Posted on Reply
#43
newtekie1
Semi-Retired Folder
SOUTHPAWMIKEI really feel like Nvidia needs to change up it's naming convention when it comes to all of these cards with different configurations that are referred to by the same model signifier. For example, thinking about the controversy around different 1060s with differing amounts of VRAM, the 3GB version could be called the GTX1063, and the 6 GB version the GTX1066. Maybe there could be different "elements" stapled on (like how Ti is titanium on the periodic table of elements) with something like Au (gold) for GDDR5 and W (tungsten) for GDDR5X. Just spitballing.
I really don't know why they wouldn't call the GDDR5X version the GTX1060 Ti. Other than the fact the GDDR5X isn't going to make much difference...
Posted on Reply
#44
GoldenX
Unless they add a substantial core OC.
Posted on Reply
#45
Nkd
Not sure bandwidth will help this card. Its too much cut down in other areas. Maybe its just a way for them to charge more for the same card. Afterall they are selling brand name anyways. Its easy to fool fanboys.
Posted on Reply
#46
hat
Enthusiast
cdawallI remember the old NV gtx 260 core 216 and they bragged that up all over the box. 1070 gained a "ti" model... Lol what a joke from the other side.
Take a step or two further back: remember the 9600GSO? One variant had 96 shaders, the other had 48 I believe...
Posted on Reply
#47
Nkd
GoldenXUnless they add a substantial core OC.
LOL it wont matter much. They already boost to the moon anyways. They will all run in to cap.
Posted on Reply
#48
cdawall
where the hell are my stars
T4C Fantasythere is literally already many many in the thousands of GP104 1060s people dont know they have, it was released secretely

www.techpowerup.com/gpuz/details/3zs4w

www.techpowerup.com/gpuz/details/a7pvz
It was announced that it was released and the PCB's I have seen are all double drilled for GP106/GP104. They do seem to be mostly chinese market that actually got the GP104 parts.

I would imagine when nvidia releases this card it will be a GP104 based model. Probably just to burn through the stock they have lol
newtekie1Funky, yes, but still distinguishable. They added the V2 to the name, as well as an SE for the other version as well as the memory amount being part of the model name as well, distinguishing between the different versions.
I agree I never disagreed that NV seemed to label products for the most part. I am sure something has slipped through the cracks over the decades of their existence, but we are talking a slip and not an open flood gate of IDGAF that AMD seems to have and this coming from someone who still has a stack as tall of him of their cards. :roll:
Posted on Reply
#49
GoldenX
hatTake a step or two further back: remember the 9600GSO? One variant had 96 shaders, the other had 48 I believe...
Better yet.
G92=8800GT - 8800GTS 512MB - 9600GSO - 9800GT - 9800GTX - 9800GTX+ - 9800GX2
Posted on Reply
#50
cdawall
where the hell are my stars
GoldenXBetter yet.
G92=8800GT - 8800GTS 512MB - 9600GSO - 9800GT - 9800GTX - 9800GTX+ - 9800GX2
I mean g92 was the best thing nvidia ever touched. It was in a lot more cards than that as well.
Posted on Reply
Add your own comment
Dec 23rd, 2024 16:25 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts