[News/Rumor] AMD Radeon RX 490 launching in December?

[StyM]

http://www.guru3d.com/news-story/amd...-december.html

It has been topic of discussion many times, but the chatter on the web is much stronger this time. From the looks of it AMD might actually release the Radeon RX 490, maybe even before Christmas.

AMD on their end spilled a couple of beans a couple of times now, IF this new rumor is true then the Radeon RX 490 series could be released in December.

If released, two things can happen here, it'll be a Dual-GPU Polaris based graphics card (2x RX 480), which is shatter that popped up the past week. If that is so it is likely a full Polaris 10 with 2304 shader processors, x2 = 4608 shader processors. Thing is that multi-GPU solutions like SLI and Crossfire are on the decline in popularity, so we hope that is not the case.

Now if we follow the old indication, it is based on a GPU called VEGA 10. Let's assume the latter one. After Polaris comes Vega, Vega is the brightest star in the constellation Lyra, the fifth brightest star in the night sky and the second brightest star in the northern celestial hemisphere, after Arcturus. On earlier roadmaps VEGA was tagged as a HBM2 product. The start of the "Vega" would happen with Vega 10 (code-name) GPU, a series to compete with Nvidias more high-end gear like the GeForce GTX 1080. It's rumored to have 64 Compute Units, multiply that with 64 shader processors per cluster and you'll get to 4096 shader processors. It can offer up to 24 TFLOP/s 16-bit (half-precision) floating point performance. You read it right, half-precision. The GPU will be paired with either 8 or 16 GB HBM2 memory at an up-to 512 GB/s memory bandwidth, logic would dictate 8GB though.

A few months ago AMD's Radeon Technologies Group (RTG) chief Raja Koduri tweeted that development of "Vega10" had just crossed a milestone, although it's a long way to go before you can see it". A little while ago an RX 490 entry was listed on AMDs own website as well.

AMD has hasn't announced anything official yet about the RX 490, but December is now very likely. It'll be interesting to see if it is VEGA 10 or a Dual-GPU Polaris 10 based product, we'll have to wait and see what rabbit AMD pulls out of its hat.

[AliG]

I do not trust a word from guru3d anymore. Perhaps if another, more reputable site backed them, I'd believe it. But from AMD's perspective wouldn't it just be smarter to focus all of their headcount on getting Vega 10 out faster?

[drmrlordx]

It needs to launch in time for Zen/Summit Ridge at the latest . . .

[zanzabar]

It needs to launch in time for Zen/Summit Ridge at the latest . . .

they are saying it is a 480x2, i dont see the point of a lower mid range xfire single card that would need to be around 350-400W.

[AliG]

they are saying it is a 480x2, i dont see the point of a lower mid range xfire single card that would need to be around 350-400W.

Exactly. It would probably compete performance wise with the 1070, but would have horrible power consumption. It just makes more sense to focus on getting Vega out as fast as possible.

[drmrlordx]

they are saying it is a 480x2, i dont see the point of a lower mid range xfire single card that would need to be around 350-400W.

I would rather they launched Vega in time for Summit Ridge, not a 480x2 card. Bleh.

[Chrono Detector]

If this turns out to be a dual 480 card, it would be disappointing. Some games do not even support multi GPU setups properly which will limit the card's full potential.

[SinOfLiberty]

Pascal Bios editor should be released around December-ish time frame too. Surely, Nvidia flagship cards like 1080Ti will still need to be sold at a high msrp BUT the time frame has been long enough already.

[SinOfLiberty]

AMD is playing "All in one" game. VEGA 10 with 24 TFlops of half precision performance and 12 TFlops of single precision performance will come out this year. Along with Zen, just like Intel and Nvidia...

Obligatory, 1080Ti is slated to launch very soon...

Edit: quotes

[Darakian]

Here's hoping for vega as well. Taking half a year to get the dual gpu version of polaris out would be massively disappointing.

[zanzabar]

AMD is playing "All in one" game. VEGA 10 with 24 TFlops of half precision performance and 12 TFlops of single precision performance will come out this year. Along with Zen, just like Intel and Nvidia...

Obligatory, 1080Ti wont be launched otherwise, but slated to; very soon...

amd is all in on HMB2, i would bet they have been waiting for samsung since hynx had problems and micron delayed due to low orders since samsung and hynx were quoting low.

[BeepBeep2]

amd is all in on HMB2, i would bet they have been waiting for samsung since hynx had problems and micron delayed due to low orders since samsung and hynx were quoting low.

I wouldn't put money on that. NVIDIA's P100 has been around for a little while now.
A dual Polaris card would probably be 275-325w TDP as far as I can tell.

[zanzabar]

I wouldn't put money on that. NVIDIA's P100 has been around for a little while now.
A dual Polaris card would probably be 275-325w TDP as far as I can tell.

p100 has no confirmed shipments and nvidia bought all of the non sample test run from micron. it apparently did not meet clock expectations but should be more than enough for the p100.

[Darakian]

I wonder if the 490 might be single chip polaris with HBM2.

[Particle]

That would seem to be unlikely. To my knowledge, HBM2 availability is the primary limiting factor for a Vega launch.

[AliG]

That would seem to be unlikely. To my knowledge, HBM2 availability is the primary limiting factor for a Vega launch.

I always wondered why they didn't create a cutdown version with GDDR5X to launch this year. I feel that would be a better investment of their R&D money simply because it would still fit a clear market segment (i.e. RX 580 with Fury xxxx being the full version).

[StAndrew]

I always wondered why they didn't create a cutdown version with GDDR5X to launch this year. I feel that would be a better investment of their R&D money simply because it would still fit a clear market segment (i.e. RX 580 with Fury xxxx being the full version).

That's assuming the Vega memory controller is compatible with GDDR5 I don't know.

[AliG]

That's assuming the Vega memory controller is compatible with HBM2 I don't know.

I thought we were already confirmed that Vega worked with HBM2?

As far as going backwards, I'm pretty sure they could just reuse the RX 480 or 390X memory controller for a cutdown version.

[StAndrew]

I thought we were already confirmed that Vega worked with HBM2?

As far as going backwards, I'm pretty sure they could just reuse the RX 480 or 390X memory controller for a cutdown version.

Oops, meant GDDR5 - edited. Would probably have to invest in new a new die to run GDDR5. Just wouldn't be worth it. AMD gambled with HBM2 and is so far not getting any payout. I curious just how much performance the extra bandwidth will net. VEGA 10 will must have some serious throughput unless this is just a "new technology" gimmick.

[Darakian]

Slightly bigger polaris with a 384 bit memory interface then? I really want to believe that it didn't take 6 months to stick two chips on a board.

[BeepBeep2]

Slightly bigger polaris with a 384 bit memory interface then? I really want to believe that it didn't take 6 months to stick two chips on a board.

It isn't as easy as "slap some more cores on it and widen the memory controller", assuming they started on that two months ago @ AMD's pace it would probably make it to retail in Q3/Q4. Interesting info was given in 2010/2011 too with R&D costs in the tens of millions to bring a new SKU to market, at least on the CPU side.

I think AMD made Polaris too small. Had they added 512 more cores to the current Polaris chip (2304+512=2816, Hawaii size), I think they would have been much more competitive in the market with RX480 @ 2816 and RX470 @ 2560/2304, even with a $250 MSRP. Polaris is currently running too fast to be as efficient as they claimed, probably because core voltage / clock had to be pushed for performance reasons. Users have noted significant power savings with undervolting and underclocking...

Upcoming Vega @ 4096 is also going to get outpaced by NVIDIA's large designs methinks.

zanzabar: You are right, the P100 does not have black/white confirmed shipments but it is already available in machines for order such as the IBM S822LC

[AliG]

Polaris isn't too small, it's already bigger than the GTX 1060 chip despite being slower.

It's an unbalanced architecture. It makes tons of TFLOPS but is greatly held back by the lack of ROPs in the backend. While DX12 clearly does benefit from this, I'm surprised they never balanced out the GCN architecture over their 4 revisions to better adapt to current games.

[Particle]

It's an unbalanced architecture. It makes tons of TFLOPS but is greatly held back by the lack of ROPs in the backend. While DX12 clearly does benefit from this, I'm surprised they never balanced out the GCN architecture over their 4 revisions to better adapt to current games.

If it was worth the power and silicon to add more ROPs to their designs in relation to the performance it gains, their design team would have done so by now. They do have real engineers.

[AliG]

If it was worth the power and silicon to add more ROPs to their designs in relation to the performance it gains, their design team would have done so by now. They do have real engineers.

I'm not sure that's necessarily true. Most architectural overhauls are done 5-6 years in advance, with only minor tweaks being financially feasible. Just look at how long Bulldozer has been around, and how vastly different Zen is.

As a "real engineer" of the electrical variety, I can confirm first hand that a lot of the times you know what the problem is, but have to wait for the next major design change due to development costs.

[Particle]

I'm not sure how to further the discussion at this point. You've expressed both that it's often necessary to wait for major revisions to make certain changes (and presumably you mean to suggest that this is one of them or there wouldn't have been a point to your last post) and also that you are surprised that they have not made such changes. You do not seem to have a clear conception of which point you'd like to make.