AMD rumored to release Radeon RX 500X series in Q3 2018

[AtenRa]

Will we see any review from you? off topic (APU war RR vs rest)

Perhaps in the summer, im not home right now

 

[lifeblood]

Honestly, in order of most to least likely my prediction is:

1. Shrink to 12nm, make a few tweaks to improve memory compression, and call it a day.
2. Shrink to 12nm, switch to GDDR5X, make a few small tweaks
3. Shrink to 12nm, switch to GDDR6, make a few small tweaks

I think option 2 is the best bang for the buck (assuming memory bandwidth is truly the problem).

Ok, do absolutely nothing and just slap a new name on it was not one of the options I considered. Well, the good news is for those of you with RX 500 series GPU's, you can download the newest drivers from AMD then change the specs in your signatures to RX 500X. Congratulations, you got a free video card upgrade! Don't you feel more awesome now?

 

[Qwertilot]

Shrink was probably always going to be too expensive to bother with - they're either selling all of them because of mining or basically none for gaming vs Volta/whatever it is called - but having done nothing at all really is phoning it in!

 

[TimCh]

So why don't they just release a new Vega card with cheaper 8GB GDDR5 memory (instead of HBM2) as a mid range card? Can't understand why they'd still want to keep pushing Polaris at this point.

Because they don't think spending a truck load of money and resources on chip that is marginally better than what they already have is a good idea.

 

[William Gaatjes]

Polaris is really not that bad as some claim it to be.
It runs all current game titles very well at HD resolution, with high settings and easily over 70 fps which is needed for a 60Hz monitor and when vsynced.
With a little bit of undervolting, power consumption is reduced and it runs even better because of less throttling.
For most gamers @ HD resolution it is a very good card. The only thing to watch out for is to get the 8GB versions. And never look at the 4GB versions.
Because most games of these days will require more than 4GB.

 

[Kenmitch]

AMD GPU department is playing lame duck and will turn into high flying eagle when the time is right....One can only hope.

 

[kawi6rr]

What's next for AMD GPU's? I've not heard much of what's coming next.

 

[jpiniero]

What's next for AMD GPU's? I've not heard much of what's coming next.

Well, besides Vega Mobile...

It's possible that AMD could release a gaming GPU based upon Vega 20 but it's unclear if they will do so. But if they do, it will be something like Vega Frontier Edition, aka super expensive.

After that, Navi sometime in the second half of next year.

 

[Ankur Saxena]

Thanks for the update

 

[NostaSeronx]

It will be interesting if it is actually Navi. Polaris and Vega are not the same GPU generation, nor part of the same family. Vega is part of the exclusive HBM family; Fiji(28nm), Alto(Intel), Vega(FinFET), etc.

Polaris to Navi.
Then, Vega to something else.

https://imagescdn.tweaktown.com/news/5/2/52983_09_amd-vega-high-end-architecture-high-gamers.jpg
https://www.tweaktown.com/news/52983/amd-vega-high-end-architecture-high-gamers/index.html

Navi is not explicit to be a high-end architecture for gamers like Vega was.

(not in actual relationship to the post 500X by the way)
Since, the 500X series launched in Q2. There might still be a series in Q3/Q4.

There is two scalability factors that can be taken from Zen.
- the IPC/Frequency/Power curve (Zen covers Jaguar to Excavator)
- the xPU side (Multiple dies)

Super-SIMD(VLIW2-GCN units; 1 Full VALU(Heavy computational; Tensor(ML),etc)+1 Core VALU(SFU-like(TAN/NGG/FireRay/Mostly Graphics+Sound))), 128-bit GDDR6, Infinity Fabric
I have investigated briefly, 7nm GDDR6 and HBM by Rambus for AMD is finished. So, product launch is soon-ish, OEM launches of the thread usually prelude with a retail launch of next-gen.

Super-SIMD -> IPC side
128-bit GDDR6/Infinity Fab -> xPU
//N7
N7 is, of course, the first generation of TSMC's 7nm process. YJ said that it has passed all qualification, is in volume production now, with over 50 tapeouts planned by the end of the year, roughly a third in mobile, 2/3 in HPC, and a couple of automotive. (HPC is the traditional GPU area)

Extended:
Super-SIMD based compute unit can provide the same ALU to texture ratio found in a typical compute unit while allowing for better L1 performance. Super-SIMD based compute unit can provide a similar level of performance with potentially less area savings as compared to SIMDs (shown as in FIG. 1A) two compute units. Super-SIMD based compute unit can also include 128k LDS with relative small area overhead for improved VGPR spilling and filling to enable more waves.

Big SuperSIMD CU array => 4 CUs => 8 Standard CUs (Replaces Vega -> Next-gen 7nm)
Small SuperSIMD CU array => 2 CUs => 4 Standard CU (Replaces Polaris -> Navi)

Full VALU (Core ALU) and Core VALU (Side ALU); just to confuse things.
http://www.freepatentsonline.com/y2018/0121386.html
(a design patent application is not published and is kept secret until granted.)

 

[Eug]

Any updates for rumours on the Vega Mobile release date? If 500X in Q3, what about Vega Mobile?

I'm just thinking that if new MacBook Pros come out in June, they will just have Polaris.

 

[beginner99]

I think he was frustrated and left because the situation was/is what it is there.

Well he was the boss of RTG so any situation he was in was of his own making...I'm more thinking he is much less capable than assumed and him gone will actually help AMD. I mean what has been released since he was there? Nothing. The last good card was Hawaii in 2013...before rajas time. After that? pretty much meh.

Videocardz is now claiming that this is just OEM rebrands.Which does make sense, the 570 and 580 were released in April last year.

https://videocardz.com/75817/amd-reading-radeon-rx-500x-series

Believable. Didn't the pull the same bs with RX 4xxX series?


They won't shrink Polaris to 12nm. Makes no sense. better to use that money saved on pushing EPYC. And and small VEGA besides vega mobile was never planned or canceled because HBM is still too expensive and vega doesn't have a gddr5 controller. Adding that would mean creating a new chip and new driver. Way too costly...Especially if you can sell your 2 year old cards for higher prices than at release due to mining craze.

 

[G2M7 zc]

So this would be a 2nd Polaris refresh, if true? It might make sense as a more mid range alternative to Vega or Vega+.

are you the real shmee?

 

[Puffnstuff]

Why would they do this in the first place? I just purchased a XFX rx 580 8gb to replace my gtx 780 ti because it was affordable to me compared to current nvidia offerings that I need to drive my 3440x1440 monitor and it barely does the job.

 

[Shmee]

Why not? It would be a refresh for a mid-high tier product. Something put out in the mean time while they work on new chips. Keep in mind, it depends on games and settings, but for a monitor with higher res than 2560x1440, I would say that in general you should go with a vega/1070Ti minimum if buying now. RX580 and a possible refresh are still good cards, but they may have trouble at high resolutions unless you start to really tinker with settings.

 

[Puffnstuff]

I'm sending this rx 580 back because the fans started making this horrible buzzing sound that increases with temperatures. I'll stick to Nvidia cards.

 

[Despoiler]

I'm sending this rx 580 back because the fans started making this horrible buzzing sound that increases with temperatures. I'll stick to Nvidia cards.

Nvidia doesn't make the fans....

 

[PeterScott]

So who should be responsible for this epic f***-up? They have brand new architecture, which was miles better than Polaris, but still release a Polaris update? Maybe Raja wasn't so bright after all...

Really the IPC of the shader/computer units about the same between Polaris and Vega. A smaller Vega design with same compute units as polaris probably wouldn't be significantly faster.

Latest rumors are that AMD won't have new GPU for consumers until Q2 2019.

It seems pretty clear they decided to wait for 7nm for newer GPUs.

This makes sense for AMD. They are smaller and have less money than NVidia, so it makes sense for them to be more frugal on expensive tapeouts, especially at 12nm, which has negligible benefits.

Though it could get very painful if things don't go smoothly with 7nm.

 

[lifeblood]

I'm not hopeful for AMD's next gen GPU's. They got the drop on Intel because Intel got lazy and complacent. Intel thought they had no competition and quit innovating and pushing the envelope. This allowed AMD to catch up and respond with a near parity CPU at a superior price.

Nvidia does not appear to be doing the same. They appear to be continuing to push forward. AMD will have a hard time catching up. I expect AMD will continue to provide competitive products in the mid-range while having only a token presence in the high end.

 

[Despoiler]

Nvidia does not appear to be doing the same. They appear to be continuing to push forward. AMD will have a hard time catching up. I expect AMD will continue to provide competitive products in the mid-range while having only a token presence in the high end.

What? Nvidia is launching late this generation compared to what they have done for the last several generations. Nvidia also went with a custom node which means AMD will hit 7nm before them. Both of these are doing the opposite of what you claim. They create room for AMD to catch up.

 

[PeterScott]

What? Nvidia is launching late this generation compared to what they have done for the last several generations. Nvidia also went with a custom node which means AMD will hit 7nm before them. Both of these are doing the opposite of what you claim. They create room for AMD to catch up.

NVidia did seem to take their foot of the gas a bit, but not nearly to the extent that Intel did.

Given what we have seen so far, it still looks like NVidia will have new generation cards about 6 months sooner than AMD, so they aren't exactly giving AMD a break.

Going straight for 7nm is something of a gamble for AMD, and it is one that I hope pays off, but even if it does, it won't be anything but a short lived gain, unless AMD has some really good designs coming.

 

[lifeblood]

What? Nvidia is launching late this generation compared to what they have done for the last several generations. Nvidia also went with a custom node which means AMD will hit 7nm before them. Both of these are doing the opposite of what you claim. They create room for AMD to catch up.

As the above person observed, Nvidia did slow down, but so has AMD. Also, yes Nvidia is taking a risk by going with a half node, but AMD is also taking a risk with 7nm. I think AMD is taking the bigger risk. Please correct me if I'm wrong but isn't AMD using GloFo for their GPU's? GloFo has failed AMD so many times in the past. TSMC, which Nvidia uses, has a much better record of delivering on their promises. Having said that I must admit GloFo has seemed to have gotten better recently so maybe the tide has turned.

Believe me, I hope AMD releases a GPU that kicks Nvidia to the curb, but I'll not hold my breath. They've disappointed me a few too many times.

 

[SpaceBeer]

AMD confirmed they'll make some products in TSMC fabs. It is expected to be (high-end) GPUs, but we'll see

 

[coercitiv]

I think AMD is taking the bigger risk. Please correct me if I'm wrong but isn't AMD using GloFo for their GPU's?

GloFo would pour press release after another if the 7nm Vega AMD showed running came from them. At least some of the 7nm GPU on AMDs roadmap will be produced at TSMC.

 

[Headfoot]

Hope they have something to show soon. It has been a very slow 12 months for consumer GPUs