Vega/Navi Rumors (Updated)

[railven]

What are the odds next incarnation of Vega runs of GDDR5/X/6 memory types?

I'm starting to feel HBM isn't the future memory for GPUs most anticipated (at least not yet.)

 

[swilli89]

At this point I just am forcing myself to wait to draw any further conclusions. I will say this situation is going to suck if Vega doesn't shake things up.

 

[3DVagabond]

But they are advertising Rx as such which uses the same uarch, besides titan started -and is the market leader in prosumer GPUs, and the titan is judged on a mix of gaming and semi pro, it excels in gaming- Vega Fe should be judged accordingly.

An anology would be Vauxhall bringing out a new hatchback car, and saying to everyone it is targeted for carrying the shopping/kids to school and shouldn't be judged on its driving experience-when the market leader ford fiesta is judged on such things and excels- just like every hatchback is judged on certain parameters that's the standard in its class. (Jack of all trades)
In the GPU world the prosumer cards are the hatchbacks, except they are expensive premium products unlike small cars.

Consumer = gaming.
Prosumer= uber gaming/gpgpu/semi professional mix (uber price)
Professional= uber gpgpu compute, stable certified drivers (crazy price)
What you and other people are suggesting is that AMD purposely turned off great gaming performance for that card ie artificially segmenting against the market expectations, which is daft.
You better believe if Vega Fe had uber titan like gaming capabilities they would enable it and market it as such, they are not because frankly it cant-hence this rubbish about just for game development.

The alternative to that Is what I mentioned in an earlier post, that AMD is pulling some masterful sandbagging for an early Christmas present, not impossible but extremely unlikely imo.

Im dismissing entirely the notion that AMD is doing neither of those and is working on an unready driver that is going to increase performance by 35℅ in one month of work- they have had 8 months to work on drivers- most of the gains would be in code already.
Edt; Just to restate- I would love some crazy driver action as I would probably go buy one, also I still think the bandwidth limitation rumours are fishy as hell, I expect if true just fixing that would bring 20℅ or so I would think, but would also hamper compute, so they would be shooting themselves in the foot with Vega Fe either way.

Titan doesn't define "prosumer" because it was a fake. An elaborate ruse to raise the price of gaming cards. It was strong in DP. And people who would never in a lifetime take advantage of that used it as a reason to buy it.

In the end Vega might prove merely adequate at gaming and excellent at compute. That doesn't make it a failure as a design. Just not a good gaming design. And then that can be mitigated depending on price.

 

[Elixer]

Apparently, someone is going to win a Vega RX at PDXLAN, and they will tell people system specs at the end of the show. (source: https://www.twitch.tv/elmnator the guy live streaming from there)
No idea if they will get the card at the show, or if they will get it later.
I assume after the launch, but, you never know.

 

[Dribble]

Titan doesn't define "prosumer" because it was a fake. An elaborate ruse to raise the price of gaming cards. It was strong in DP. And people who would never in a lifetime take advantage of that used it as a reason to buy it.

As far as I can see there is no such thing as a prosumer market - it seems to have been invented during all this Vega hype.

 

[_UP_]

As far as I can see there is no such thing as a prosumer market - it seems to have been invented during all this Vega hype.

That's not entirely accurate.
People forget this market, but it's a pretty big one.
At my university, Titans are used as a machine learning and neural nets GPUs. They allow the university to buy more cards as they are cheaper than pro cards, and still have ample memory and have many CUs.
Nvidia also has a program for giving Titans to students doing research that we were encouraged to submit an application for.
I won't be surprised if the FE would work well in that market too soon.
Cheaper does play a part for many universities.

 

[Mopetar]

What are the odds next incarnation of Vega runs of GDDR5/X/6 memory types?

Unlikely as AMD wants to eventually use HBM for APUs which means continuing to invest in it and working to get costs down.

 

[DeathReborn]

What are the odds next incarnation of Vega runs of GDDR5/X/6 memory types?

I'm starting to feel HBM isn't the future memory for GPUs most anticipated (at least not yet.)

HBM is fine, AMD's implementation in VEGA is questionable. 3-4 stacks would have been a better choice, at a lower voltage and clock speed they could put out 640-768GB BW.

 

[geoxile]

Not sure how anyone can question HBM. It's just expensive for now because it's new but even HBM2 can slam dunk over GDDR6's projected bandwidth with a 4-stack at full speed (1000MHz) IIRC. Something like 1024GB/s with a 4-stack @1000MHz and at lower than GDDR5X power consumption. Price is the only issue for now, but I'm sure that'll change considering Samsung is starting production too.

 

[Elixer]

Not sure how anyone can question HBM. It's just expensive for now because it's new but even HBM2 can slam dunk over GDDR6's projected bandwidth with a 4-stack at full speed (1000MHz) IIRC. Something like 1024GB/s with a 4-stack @1000MHz and at lower than GDDR5X power consumption. Price is the only issue for now, but I'm sure that'll change considering Samsung is starting production too.

HBM is much more complex & expensive than GDDR5X.
So, sure, on paper, it seems like it is a winner, but, so far, the products that have been using it have been more expensive, and uses more voltage than expected.

 

[geoxile]

HBM is much more complex & expensive than GDDR5X.
So, sure, on paper, it seems like it is a winner, but, so far, the products that have been using it have been more expensive, and uses more voltage than expected.

Based on the content creation program benchmarks for Vega FE it seems to match big Pascal in those programs just fine at a similar level of efficiency. So HBM is just fine for those purposes apparently, the only problem is price. Until we know more about why games perform so much worse I wouldn't say it has anything to do with the memory when AMD's cards historically have known weaknesses (worse tri and pixel fillrates than Nvidia's cards) that have more direct relationships with performance in games.

Regarding price, it'll get better since Samsung is starting production.

 

[railven]

Unlikely as AMD wants to eventually use HBM for APUs which means continuing to invest in it and working to get costs down.

HBM is fine, AMD's implementation in VEGA is questionable. 3-4 stacks would have been a better choice, at a lower voltage and clock speed they could put out 640-768GB BW.

Not sure how anyone can question HBM. It's just expensive for now because it's new but even HBM2 can slam dunk over GDDR6's projected bandwidth with a 4-stack at full speed (1000MHz) IIRC. Something like 1024GB/s with a 4-stack @1000MHz and at lower than GDDR5X power consumption. Price is the only issue for now, but I'm sure that'll change considering Samsung is starting production too.

It's more so because at least to me it feels like HBM isn't ready for mass production products. If smaller Vega is aimed at a more mainstream segment, it just makes more sense for AMD to use a cheaper memory type.

Of course things can changing going forward, but so far it seems HBM has become an anchor at least for Fury and the verdict still isn't in for Vega. However, I don't see roses in their future.

Just feel it be more practical for AMD to move to a memory type that is readily available.

 

[OatisCampbell]

Ok we know they showed a VEGA RX system against NVIDIA - who says this was a 1080 heavy-weight fight? Polaris 580 is positioned against 1060, they could have demoed SMALL VEGA as 1070 competitor, or a 1080 competitor with 1070 price tag. We'll see.

Annnnnddd....they are holding back BIG VEGA for the big surprise on NVIDIA because NVIDIA doesn't have any chips more powerful than the GTX1080....like the GTX1080Ti...or the old Titan Xp...or the new Titan Xp...or that bigger pro chip that beats all three of those....and possibly some Volta engineering samples....

It's been over two years since AMD launched a high end chip, you think they're playing games like this?

It makes no sense.

I'll be pretty shocked if RX Vega turns out much different than FE Watercooled, but unlike many, I think that would be OK for the right price in relation to 1080. What I wonder about is how profitable Vega can be for AMD as a 1080 competitor. :-(

 

[SickBeast]

It's looking extremely bad for Vega right now, however AMD has shown from their recent launches that they are able to quickly refine their products and turn them into winners. Polaris and Ryzen both had terrible launches, but AMD was able to fix them quickly (Polaris had the PCI-E power issue that was sorted out with the aftermarket cards, and Ryzen had poor gaming performance that was fixed with new motherboard BIOSes and faster memory). My personal opinion on Vega is that the launch is going to be a complete flop. At best I expect it to beat the GTX 1080 in a few games, with much higher power consumption. The Vega FE card has been tested to consume 440 watts when overclocked (!). It's just not looking good right now. However I do believe that the aftermarket cards will save AMD once again. Well cooled aftermarket cards will eliminate the throttling we are seeing now, plus I also expect them to release a slightly crippled Vega core for much cheaper. The real problems for AMD are timing and availability. The market has become completely saturated due to the mining boom. I'm personally running GTX 1070 SLI when I had absolutely no plans for an upgrade. The graphics cards have been flying off the shelves. This has been eating into AMD's market, day by day. Then there is also the fact that nVidia is poised to release Volta at any time. If AMD gains any type of momentum with Vega I expect nVidia to drop Volta. The saving grace for AMD is that Vega will be used in the Xbox One X. However analysts are predicting that console to be complete flop because it's too expensive and only 10% of US households own a 4k TV. Hopefully Ryzen will provide AMD with enough revenue to keep going for at least a few more years. They have to make hay while the sun shines though. Coffee Lake looks like a beast. At that point AMD could wind up right back where they were a short time ago. Near bankruptcy.

 

[maddie]

What are the odds next incarnation of Vega runs of GDDR5/X/6 memory types?

I'm starting to feel HBM isn't the future memory for GPUs most anticipated (at least not yet.)

If true, then that whole excursion into a HBCC world is wasted and will have to be abandoned.

GDDR will not work for an effective HBCC implementation.

 

[Elixer]

Heh, there was actually a "reporter" (Shawn) from wccftech @PDXLAN asking people what they thought of the Vega setup, and which "side" was better.

 

[HurleyBird]

The saving grace for AMD is that Vega will be used in the Xbox One X.

It isn't. Indications are that the XOX GPU is modified Polaris. It may have some vega features, but lacks, for example, packed math ability.

The PS4 Pro's GPU does have packed math, but also isn't Vega.

Hopefully Vega has the ID buffer from the PS4 Pro. If so, we could see some pretty sweet (and cheap) upscaling, anti-aliasing, and motion based effects.

 

[3DVagabond]

As far as I can see there is no such thing as a prosumer market - it seems to have been invented during all this Vega hype.

You must not have been around for the original Titan then.

 

[3DVagabond]

HBM is much more complex & expensive than GDDR5X.
So, sure, on paper, it seems like it is a winner, but, so far, the products that have been using it have been more expensive, and uses more voltage than expected.

You can't use GDDR5X/6 on APU's. It's not all about DGPU. They are working on making them extinct.

 

[happyflow]

It's been over two years since AMD launched a high end chip, you think they're playing games like this?

It makes no sense.

I think they ran into serious issues with Vega and needed time to fix it (re-spin or more radical change). The now-useless-for-gaming cards are sold as 'Prosumer' and potentially lower end models.

Sounds like wishful thinking, but I find it hard to believe that a company like AMD is not able to deliver an at least competive model until they had almost one year more to develop it compared to 1080(ti)...

 

[Tup3x]

I think they ran into serious issues with Vega and needed time to fix it (re-spin or more radical change). The now-useless-for-gaming cards are sold as 'Prosumer' and potentially lower end models.

Sounds like wishful thinking, but I find it hard to believe that a company like AMD is not able to deliver an at least competive model until they had almost one year more to develop it compared to 1080(ti)...

If they are really selling basically broken chips at massive price premium, they are really risking it. I don't believe it.

 

[Head1985]

HBM is fine, AMD's implementation in VEGA is questionable. 3-4 stacks would have been a better choice, at a lower voltage and clock speed they could put out 640-768GB BW.

Yeah i still dont know whos idea was OC GPU +50% and reduce memory bandwidth by 6% is good thing.That person should be fired.

 

[Karnak]

Would've been way more expensive. That's the point.

 

[DeathReborn]

It isn't. Indications are that the XOX GPU is modified Polaris. It may have some vega features, but lacks, for example, packed math ability.

The PS4 Pro's GPU does have packed math, but also isn't Vega.

Hopefully Vega has the ID buffer from the PS4 Pro. If so, we could see some pretty sweet (and cheap) upscaling, anti-aliasing, and motion based effects.

The more I read about what was in the PS4 or Xbones GPU's (and their pro etc variants) the more I think AMD has has outsourced GPU design ideas to Sony & Microsoft.

 

[AtenRa]

Got to stop with rather vague and misleading statements. AMD has shown RX Vega's performance a couple of times. Is it quantifiable? Not, but it's still performance they've demonstrated multiple times. They've made the comparison themselves. When someone over at AMD said something along the lines of "it looks good against Titan Xp" people took that as an example of the performance it is capable (and actually ran with it).

We've seen RX Vega performance. We just need the hard numbers to line it up to AMD's own current live demonstrations.

The fact that what we've seen of RX Vega's performance is lackluster is on AMD themselves, since they're the ones that trotted it out and said "look."

So by your logic, RX VEGA could be 15-20% faster than GTX1080 and this goes in line with what AMD said about being close to GTX1080 and why they used the GTX1080 on the last event . Ok thanks for clearing the misleading RX performance shown by AMD.