New AMD Polaris based GPU

[LTC8K6]

Looks like the Hynix 10gbps chips need 1.55V whereas the Samsung 9Gbps chips just need 1.5V?

 

[Elfear]

I'm going to entirely disagree with you here. You either draw the line in the sand or you don't. The 1030's were clearly labeled with DDR4, which is every bit as explanatory as "2048SP" to the average consumer. There is no blurring. Lets not defend one company by saying another did the bad thing better.

I already addressed that above. I wasn't aware that the 1030's had "DDR4" plastered on the packaging to distinguish them from the much faster DDR5 models. That's much better than making no distinction.

 

[IEC]

We know for sure it supports 5% higher clock speeds. 8400 MHz in the Sapphire Special Edition RX 580 is the example here. Could it go higher? Possibly.

I've personally overclocked to 8800-9000MHz on early RX 480 reference cards. So I don't think it requires a revision of the silicon to handle the higher speeds.

 

[RTX2080]

https://videocardz.com/78757/amd-radeon-rx-590-final-fantasy-xv-benchmarks-results-emerge

AMD Radeon RX 590
The Radeon RX 590 is believed to offer a bigger jump in performance over RX 580 compared to RX 580/480 refresh. We are looking a 112 to 124% of RX 480 performance, while RX 580 was at most offering 102-109% of Polaris 10 performance.

The charts leaked through FFXV website indicate that NVIDIA’s GTX 1060 is in trouble. This is likely why the company confirmed that GDDR5X variants of this Pascal-based model will be available, as there is no trance of GTX/RTX 2060 cards coming to market yet.

 

[tviceman]

https://videocardz.com/78757/amd-radeon-rx-590-final-fantasy-xv-benchmarks-results-emerge

I'm questioning whether or not this is real, and if it is what kind of thermal performance it will come with. I hope it's true though; it would go to show that Nvidia jumped to gun with ray tracing too soon given it's prices, thermal envelopes, and die sizes.

 

[LTC8K6]

Well, the 56 seems to be beating the 64 in some of those charts...

 

[EXCellR8]

What's the MSRP on the RX 590? Can't be any more than $250...

 

[IntelUser2000]

https://videocardz.com/78757/amd-radeon-rx-590-final-fantasy-xv-benchmarks-results-emerge

The RX 580 at TPU is 5-7% faster than RX 480. 10% gain on top of that is quite decent, but nothing that'll change the landscape.

 

[darkswordsman17]

A 15-20% improvement over the RX480 seems possible, as I think Polaris on early GF14nm was especially hindered, and so should be able to gain a good amount from 12nm. I think GF said 12nm was good for 10% better performance at increased density. Which binning/better voltage tweaking alone could probably get AMD an extra 5-10% over the RX480 which seemed especially poor in that regard. And if they can get more memory bandwidth that should help as well. Which I think ~10-15% was the most likely and this seems to fall in line. I thought 25% might be possible if they moved to GDDR5X or 6 and upped bandwidth a good amount.

But still nowhere close to the 1070/Vega 56 region (which is what 50% higher than the 480/1060?). If Nvidia drops the 1070 to $300, then this RX 590 or 680 or whatever would have to be below $250 or come with significant value add (free games). And even then the 1070 is still pretty appealing.

I'm questioning whether or not this is real, and if it is what kind of thermal performance it will come with. I hope it's true though; it would go to show that Nvidia jumped to gun with ray tracing too soon given it's prices, thermal envelopes, and die sizes.

It seems like Nvidia was expecting Samsung's 10nm to be a decent step up from the tweaked 16/14nm that 12nm was. Supposedly the RTX chips were slated for that (Samsung 10nm), but they had to settle for TSMCs 12nm. Even then, I have a hunch that RTX for consumers wasn't likely to have been seriously considered beyond maybe a Titan and/or top end Ti, until 7nm, where it would have made the RTX components much less of an issue, but they needed the volume to offset the cost on 12nm. I think perhaps Nvidia banked on Samsung's 7nm which is taking longer than TSMCs. That or AMD managed to secure a lot of early 7nm at TSMC which squeezed Nvidia out (so Nvidia would have been making the 2070/2080 on TSMC, where they'd announce it probably at CES and then have it out early spring, followed by the Titan on 7nm, then the 2060 series and Ti for the holidays).

 

[SirDinadan]

I can't see huge gains if they don't up the memory clocks (currently 2,000 MHz) and only push the pedal once again on the gfx core clocks.
The difference between the RX 480 vs RX 580 with high-end models of Sapphire is usually not more than 5%.

Boost Clocks:
Sapphire RX 480 (1342 MHz)
Sapphire RX 580 (1450 MHz)
ASUS RX 590 ROG STRIX (1545 MHz)

Why not go for GDDR5 9Gbps, is it that much different and would mean additional complexity and not worth the hassle and cost?

 

[lifeblood]

A few sites are now reporting a 15 November release date for the RX 590. Given all the rumors and leaks, I’m now willing to say this thing is probably for real. That date also makes sense as it's in time for the Christmas buying season.

The bad news is that they are predicting a price of around $300 for the Chinese market. If the US price is near that then I think this thing is DOA. Assuming performance is only a little better than the current gtx 1060 6GB then they are asking too much for too little.

Could the US price be cheaper than the Chinese price? I would think the US price would be higher because of Trumps Tariffs, but I'm no expert. I would think $250 would be more the sweet spot for the RX 590.

 

[SirDinadan]

Slide deck:
POLARIS EVOLVED - A better version to upgrade your old GPU
- Latest Generation FinFET
- Aggressive Tuning for Higher Clocks
- Enhanced Multi-Monitor Efficiency
- Expanded Options & Robust Partner Designs

No mention of memory, so it is surely GDDR5 8 Gbps. Bummer!

GeForce GTX 1060 8Gbps - July 2016
GeForce GTX 1060 9Gbps - April 2017
GeForce GTX 1060 GDDR5X - October 2018
RX 480 - June 2016
RX 580 - April 2017
RX 590 - November 2018

I don't really see the point, I mean anyone who wanted to upgrade to this price / performance bracket had the option to do so with the RX480/580 and GTX1060 8/9 Gbps. AMD must have free capacities at the foundries and stockpiles of unused regular GDDR5.

 

[lifeblood]

I wonder if part of the reason RX 590 exists is as part of the renegotiation's between GloFo and AMD following cancellation of 7nm. AMD and GloFo would have had multiple contracts covering multiple products and as part of the back and forth negotiations AMD may have said “let us move products x, y, & z to TSMC and we’ll buy another Polaris respin on your 12nm process.” That happens in business all the time. While RX 590 may not make much sense per se, it may be necessary in the broader picture of moving to TSMC.

This is all guessing on my part and assumes RX 590 is being made at GloFo. If it ends up being a TSMC product then forget what I said.

 

[PeterScott]

I wonder if part of the reason RX 590 exists is as part of the renegotiation's between GloFo and AMD following cancellation of 7nm. AMD and GloFo would have had multiple contracts covering multiple products and as part of the back and forth negotiations AMD may have said “let us move products x, y, & z to TSMC and we’ll buy another Polaris respin on your 12nm process.” That happens in business all the time. While RX 590 may not make much sense per se, it may be necessary in the broader picture of moving to TSMC.

This is all guessing on my part and assumes RX 590 is being made at GloFo. If it ends up being a TSMC product then forget what I said.

590 exists because it's a cheap process upgrade, requiring no new masks. It just like 2000 series Ryzen.

 

[lifeblood]

590 exists because it's a cheap process upgrade, requiring no new masks. It just like 2000 series Ryzen.

Yes/No. Yes, your right, Polaris 30 is an easy upgrade (assuming it actually is made at GloFo), just like Ryzen 2XXX was. Also no, Ryzen 2XXX was the second half of a one two punch right in Intel's gut. Ryzen was a new architecture, a much hoped for return to competition. Polaris 30 will just be a slight speed bump that really won't effect the landscape. As easy as moving to 12nm is, it's still not free. Their has to be a realistic business case for AMD to go through the effort.

 

[neblogai]

Yes/No. Yes, your right, Polaris 30 is an easy upgrade (assuming it actually is made at GloFo), just like Ryzen 2XXX was. Also no, Ryzen 2XXX was the second half of a one two punch right in Intel's gut. Ryzen was a new architecture, a much hoped for return to competition. Polaris 30 will just be a slight speed bump that really won't effect the landscape. As easy as moving to 12nm is, it's still not free. Their has to be a realistic business case for AMD to go through the effort.

Ryzen 2000 series are the same design as Ryzen 1000, the difference is 12nm and firmware. Polaris30 is no different.
As for affecting the landscape- for competing GPUs, whovever is 1% faster is the winner. AMD was already becoming a winner with the RX580 in early 2017, only miners deleted the stock for gamers. Now- AMD can claim that win again- if nVidia does not release a competitive Turing, or cut a GP104 that is cut down to a lesser extent than GTX1060 GDDR5X.

 

[PeterScott]

Yes/No. Yes, your right, Polaris 30 is an easy upgrade (assuming it actually is made at GloFo), just like Ryzen 2XXX was. Also no, Ryzen 2XXX was the second half of a one two punch right in Intel's gut. Ryzen was a new architecture, a much hoped for return to competition. Polaris 30 will just be a slight speed bump that really won't effect the landscape. As easy as moving to 12nm is, it's still not free. Their has to be a realistic business case for AMD to go through the effort.

Effort is negligible. It's a process tweak.

 

[lifeblood]

...Now- AMD can claim that win again- if nVidia does not release a competitive Turing, or cut a GP104 that is cut down to a lesser extent than GTX1060 GDDR5X.

Which appears to be happening. I've read a few articles out there about a new gtx 1060 with GDDR5X memory that some suspect is a cut down GP104. Thrust and counter thrust. The war continues...

 

[PeterScott]

Which appears to be happening. I've read a few articles out there about a new gtx 1060 with GDDR5X memory that some suspect is a cut down GP104. Thrust and counter thrust. The war continues...

It looks more like GP106, with no real performance benefit. Maybe a way to clear some old stocks of GDDR5X.

 

[GodisanAtheist]

GTX 1070 performance for $250 or GTFO.

Polaris tweaked could absolutely be capable of this, and I can't imagine the expenditure would be great.

Would make Vega look like [a bigger] failure in the gaming space though...

 

[jpiniero]

GTX 1070 performance for $250 or GTFO.

Polaris tweaked could absolutely be capable of this, and I can't imagine the expenditure would be great.

1070 is too much to ask without bumping up the shader count.

 

[GodisanAtheist]

1070 is too much to ask without bumping up the shader count.

-True, I was referencing the arch more than the die/chip. A scaled up arch with ~2800SP and equivalent parts + some faster ram would definitely land in 1070 territory...

Not to rehash the same old arguments but a ~3500SP Polaris would be in 1080 territory.

Hell it makes me wonder why they didn't...

 

[SpaceBeer]

With 32 ROPs? It will be rebranded in a few months as RX 660/670 to match RTX/GTX 2050 with $149/159 MSRP

 

[LTC8K6]

With the dying 2080ti rumors swirling, it's too bad AMD can't take advantage.

 

[Guru]

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.

The RX 680 and RX 690 on the other hand seem like significant upgrades over their previous models, much higher clock speeds at lower power consumption is a significant step forward if they cost the same amount of money. I'm curious to see if maybe the RX 690 features a slightly different core design in terms of rop's, though I doubt it. But I'd take a RX 690 that is 10% faster than the RX 580 as the same price point, all the while consuming less power as well.