New AMD Polaris based GPU

[Headfoot]

I'm interested in a 1060 GDDR5x from an academic standpoint to see a true apples to apples comparison of what the effects are of GDDR5 vs GDDR5x at the same frequency on the exact same GPU (if its really still GP106). But I wouldn't buy one

 

[neblogai]

GTX 1060 6GB GDDR5x looks completely underwhelming and completely worthless. I mean the 9gbps version 1060 barely gained like 1-2fps in just a few games, so its not like the GTX 1060 is memory bound, in most cases it core bound and I can't see anything more than 1fps gain with the GDDR5x in few titles.

So overall from vanilla GTX 1060 6GB to GDDR5x we will see 2-4fps difference in some games. Hardly anything.

Current info is that GTX 1060 GDDR5x will run it's memory at 8.8Gbps, so about the same clocks as GDDR5 version. That is why I wrote, that if nVidia wants to compete vs RX590, they need either a new chip, or a GP104 that is not cut down as much.
As for the reason why we see GTX 1060 GDDR5x at all- it could be that nVidia has too many GP104 dies, or many GP104 dies with damaged memory controllers, or good deals on GDDR5X VRAM now that GDDR5X cards are being phased out.

 

[PeterScott]

Current info is that GTX 1060 GDDR5x will run it's memory at 8.8Gbps, so about the same clocks as GDDR5 version. That is why I wrote, that if nVidia wants to compete vs RX590, they need either a new chip, or a GP104 that is not cut down as much.
As for the reason why we see GTX 1060 GDDR5x at all- it could be that nVidia has too many GP104 dies, or many GP104 dies with damaged memory controllers, or good deals on GDDR5X VRAM now that GDDR5X cards are being phased out.

Why would you think it has GP104 dies? Are you assuming they can't connect GDDR5x to GP106?

GTX 1070 uses GP104 without GDDR5x.

 

[neblogai]

Why would you think it has GP104 dies? Are you assuming they can't connect GDDR5x to GP106?

GTX 1070 uses GP104 without GDDR5x.

https://videocardz.com/78786/palit-launches-geforce-gtx-1060-gamingpro-oc-with-gddr5x-memory

Says GP104. But the article still mentions they'd be happy to confirm it is GP104 themselves.

 

[LTC8K6]

Wiki lists 5 GP106 variants and 2 GP104 variants for the GTX1060.

https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units#GeForce_10_series

 

[PeterScott]

https://videocardz.com/78786/palit-launches-geforce-gtx-1060-gamingpro-oc-with-gddr5x-memory

Says GP104. But the article still mentions they'd be happy to confirm it is GP104 themselves.

It says:
"In terms of design, the card looks pretty much identical to GTX 1060 8Gbps, but Palit did not release a picture (render) of the back side. This would ultimately confirm if the GDDR5X model features GP104 or GP106 GPU."

Sounds like GP104 is a rumor and no one really knows if it is 104, or 106.

Either way, as reported this card makes little sense. It is reported to essentially have identical specs to current 6GB GTX 1060.

That seems like quite a waste of a GP104. Between 1080, 1070 and 1070 Ti there probably aren't significant numbers of 104 chips that are too damaged to be one of the 1080/1070 series, but still workable for 1060.

And even if it is a GP104, why go GDDRx on the lower end card when 1070/1070ti use regular GDDR?

I could see it if this were more like a GTX 1060 Ti with more execution units, but with all the same unit counts, I can't make any sense of this one.

 

[LTC8K6]

Preparatory to re-labeling it a 2060?

 

[lifeblood]

According to the two pics released by Andreas Schilling, 590 will be built on 12nm FinFET. That still doesn't absolutely confirm GloFo, or does it? According to Prof Google, TSMC also has a 12nm FinFET. Since Polaris is fabbed at GloFo, I really can't see AMD doing the work to move it to TSMC so I'm just going to assume GloFo.

And these pics tend to put the icing on the cake that RX 590 is real. So we have a new card with ~10% performance boost. The only question now is price (and heat, power consumption, etc).

 

[PeterScott]

According to the two pics released by Andreas Schilling, 590 will be built on 12nm FinFET. That still doesn't absolutely confirm GloFo, or does it? According to Prof Google, TSMC also has a 12nm FinFET. Since Polaris is fabbed at GloFo, I really can't see AMD doing the work to move it to TSMC so I'm just going to assume GloFo.

And these pics tend to put the icing on the cake that RX 590 is real. So we have a new card with ~10% performance boost. The only question now is price (and heat, power consumption, etc).

It's GloFo. The only reason this exists is because it's just a process tweak. IOW there is no AWD design work involved. They reuse all the design and probably most of the masks. It isn't even a shrink. The features are exactly the same size.

If they went to another fab, they would need to build expensive brand new custom masks and spend a lot more time testing and tweaking.

The is nothing but a mild process update at GloFo.

 

[kawi6rr]

I was under the impression that the 580 was 14nm and the 590 will be 12nm.

14 to 12 seems like a shrink to me but I'm not an expert.

 

[itsmydamnation]

I was under the impression that the 580 was 14nm and the 590 will be 12nm.

14 to 12 seems like a shrink to me but I'm not an expert.

You dont have to shink if you dont want to, exactly like Ryzen2 . Take existing masks and just get a incremental benefit at little development cost, or do far more layout and physical work for a bigger yet still small gain.

 

[lifeblood]

I woke up this morning and checked my news feeds. At the top of the list was a item from wccftech saying something to the effect of "Radeon RX 590 graphics card will have 40 CUs with 2560 stream processors". What?! So I followed the link to the article which said "Radeon RX 590 is going to have a total of 36 CUs or 2304 SPs". Huh? I checked the link, it said 40 CUs with 2560 SPs. I look at the article title and text, it says 36 CUs or 2304 SPs. I refresh my news feed and it then changed to 36 CUs and 2304 SPs.

Was wccftech playing a serious click bait game or was I just blind?

 

[psolord]

Why didn't you take a screenshot? Not that we don't believe you. Clickbait is here for years, to remind us of the quality of human beings.

 

[lifeblood]

Why didn't you take a screenshot? Not that we don't believe you. Clickbait is here for years, to remind us of the quality of human beings.

Didn't even think to take a screenshot. I was thinking it was linking to the wrong article or something, then it was too late.

 

[EXCellR8]

Sapphire releasing another Special Edition NITRO card for the 590

https://videocardz.com/newz/sapphire-radeon-rx-590-nitro-special-edition-smiles-for-camera

Be nicer if they offered it with the regular black shroud coz that aqua color matches literally nothing I've ever assembled lol

 

[LTC8K6]

I woke up this morning and checked my news feeds. At the top of the list was a item from wccftech saying something to the effect of "Radeon RX 590 graphics card will have 40 CUs with 2560 stream processors". What?! So I followed the link to the article which said "Radeon RX 590 is going to have a total of 36 CUs or 2304 SPs". Huh? I checked the link, it said 40 CUs with 2560 SPs. I look at the article title and text, it says 36 CUs or 2304 SPs. I refresh my news feed and it then changed to 36 CUs and 2304 SPs.

Was wccftech playing a serious click bait game or was I just blind?

https://i.imgur.com/ly156wc.png

 

[Justinbaileyman]

Sapphire releasing another Special Edition NITRO card for the 590

https://videocardz.com/newz/sapphire-radeon-rx-590-nitro-special-edition-smiles-for-camera

Be nicer if they offered it with the regular black shroud coz that aqua color matches literally nothing I've ever assembled lol

Or how about an all white shroud?? Wonder how close these will perform compared to rx vega 56??

 

[Justinbaileyman]

https://i.imgur.com/ly156wc.png

So are these rx 590's having less CU's then the 580's??

 

[LTC8K6]

So are these rx 590's having less CU's then the 580's??

Looks like the same, but probably higher clocks.

 

[EXCellR8]

So are these rx 590's having less CU's then the 580's??

should be the same iirc (2304)

 

[lifeblood]

https://i.imgur.com/ly156wc.png

So I'm not losing my mine (or maybe I am but I'm still right about this). That's not even fake news, that's just blatant lying for the sake of clicks. No wonder everyone here has such a low opinion of wccftech.

 

[Justinbaileyman]

If this is really a 2nm shrink going from 14nm to 12nm then why not just add those 4 extra CU's?
Why do a 2nm shrink just for a very little speed bump??
This release is totally pointless and a waist of resources for AMD!!

 

[PeterScott]

If this is really a 2nm shrink going from 14nm to 12nm then why not just add those 4 extra CU's?
Why do a 2nm shrink just for a very little speed bump??
This release is totally pointless and a waist of resources for AMD!!

Because it's not really a shrink, and not a new design, it takes nearly zero AMD resources. It's like Summit Ridge (Ryzen 1000) to Pinnacle Ridge (Ryzen 2000).

Its the same exact die, same exact size, same exact units. It's just a process tweak. It should more properly be called 14nm+ because the size doesn't actually change.

 

[zinfamous]

If this is really a 2nm shrink going from 14nm to 12nm then why not just add those 4 extra CU's?
Why do a 2nm shrink just for a very little speed bump??
This release is totally pointless and a waist of resources for AMD!!

It's not wasted resources. They have contractual obligations with GloFo that they still must meet, so it's actually a greater loss to not send out new orders from them. AMD is still very likely to see some profit off of this, though I guess margins are super tiny. They are currently focused on TSMC 7nm with Zen 2 and Vega 20 for high-margin enterprise products.

 

[Stuka87]

If this is really a 2nm shrink going from 14nm to 12nm then why not just add those 4 extra CU's?
Why do a 2nm shrink just for a very little speed bump??
This release is totally pointless and a waist of resources for AMD!!

Node "sizes" stopped meaning actual physical sizes ages ago (basically when finfet came around). In this case, going from 14 to 12 is just a slight tweak, no real changes from AMD.