Poll: ** The AMD VEGA Thread **

[Matkinson]

Vega is going to have to be damn good value for money at this rate.

 

[LoadsaMoney]

AMD's Vega GPU is late

July 30 is the first day

Vega is delayed and while AMD said it will arrive in Q2 2017, it won't.

AMD went on the record at a Computex press conference and said it will launch Vega at the Siggraph conference that starts on the July 30.

We have been hearing that Vega has been delayed for quite some time, but we cannot yet understand the reason for the delay. The card that we saw back in December and February appeared fine but apparently, AMD is struggling to bring this piece of technology to the market.

High Bandwidth Memory 2 (HBM2) is not in short supply, so this can be ruled out. However, it remains to be seen if there is a manufacturing problem. It is hard to imagine that this could be driver-related, as the driver team has mostly improved the quality of drivers for the Polaris series tremendously.

http://www.fudzilla.com/news/graphics/43781-vega-gpu-coming-at-siggraph

 

[N19h7m4r3]

http://www.fudzilla.com/news/graphics/43781-vega-gpu-coming-at-siggraph

Claims no supply issues for HBM2, while SK Hynix themselves haven't been able to deliver any 2.0Gbps modules, despite stating mass production for it in Q3 2017. No no estimate or 2017 at all.

1.6Gbps was also mass production for Q3 2016, then Q1 2017, and now Q2 2017.
Only HBM2 that seems to be in some supply is Samsung's slower 1.4Gbps which NVIDIA have been using on Tesla, and latest Quadro GP100.

As a result Vega won't even match Fury X memory bandwidth of 512GB/s; and needs custom 1.88Gbps HBM2 to reach 480GB/s bandwidth.

Certainly looks like HBM2 issues are there; although it's clear there are other issues as well.

 

[Rezident]

Hopefully Vega will be another HD7900/R290 beast and not another Fury/RX400 fail.

My 290X was an excellent card alright, but I have to say, the much-maligned Fury is still doing the business at 1440p in all the games I’ve played (except DXMD but performance was broken in that game for Nvidia too – 100% greedy and inept Square Enix’s fault, not AMD or Nvidia). The 8gb RX480 in my VR rig is good too, and for the price it’s pretty sweet.

The fail was the ‘overclockers dream’ marketing nonsense etc. but since my first qualification (before I left the Dark Side) was in marketing I know not to listen to it and I don’t watch ads anymore (they're bad for you). As HG Wells said: ‘Marketing is legalised lying’. Hardware > Marketing for me.

 

[nashathedog]

There are loads on a popular auction site and they also pop up in the MM. Great cards.

I put mine on the MM but no-one replied

 

[Razor Time]

I put mine on the MM but no-one replied

Bless ya.


I was talking about the 980Ti, no one wants a fury.

 

[ICDP]

I was spending £400-£600 on GPUs every 6 months, all for maybe 20% higher FPS. I decided to stop this cycle when AMD dropped out of the high end and the way Nvidia was going re price increases and drip feeding releases. This choice was made much easier with a Freesync monitor because 20% - 50% extra FPS means nothing if it means turning off Freesync.

I have a Fury X (bought cheap second hand for less then I got for my 980Ti) and a Freesync monitor. From my own testing with freesync on vs off, I get much smoother and better lag, stutter and tearing free game-play at 35-45 compared to 60-80 FPS Freesync off.

I am happy holding out for Vega to keep Freesync because a 1080Ti and no Freesync would be a downgrade for me.

 

[benjii]

AMD's Vega GPU is late



http://www.fudzilla.com/news/graphics/43781-vega-gpu-coming-at-siggraph

Technically they said they'd launch Vega Q2, they didn't say what aspects of Vega. Vega Frontier is still launching Q2. If that is what they intended when they made that statement months ago, they need a slap up the side of the head. If not, they need to just own the problem.

 

[nomadd]

Technically they said they'd launch Vega Q2, they didn't say what aspects of Vega.

Seems it's just the Crossfire Bridge Cable at the moment.

 

[N19h7m4r3]

Seems it's just the Crossfire Bridge Cable at the moment.

It's the Frontier Edition card.

 

[TNA]

My 290X was an excellent card alright, but I have to say, the much-maligned Fury is still doing the business at 1440p in all the games I’ve played (except DXMD but performance was broken in that game for Nvidia too – 100% greedy and inept Square Enix’s fault, not AMD or Nvidia). The 8gb RX480 in my VR rig is good too, and for the price it’s pretty sweet.

The fail was the ‘overclockers dream’ marketing nonsense etc. but since my first qualification (before I left the Dark Side) was in marketing I know not to listen to it and I don’t watch ads anymore (they're bad for you). As HG Wells said: ‘Marketing is legalised lying’. Hardware > Marketing for me.

Well said

 

[JediFragger]

not fussed on the delay, my fury does the job nicely for now and means I can save up more and get a nice factory oc'd aib card if vega is actually any good. Plus its summer and too hot to game most of the time anyway

Yup, more beer drinking time!!!

 

[old gamer]

I put mine on the MM but no-one replied

yes they did, just took a month that's all

 

[akarypid]

They Prey demo was another horrible PR move in a long list of AMD PR mistakes. It was supposed to be showing off Threadripper IO, instead it has caused people to panic and started the rumour that Vega is junk.

This is a market where PR is all about bragging rights.

Anything short of showing something at 30% faster than a 1080ti would be met with scorn and pitchforks by the press and a very vocal part of the gamer community.

But AMD don't have a card that is faster than the 1080ti. So there's not much they can do now, is there?

The sensible people that will take that performance and enjoy it with their Freesync monitors will silently do that anyway so I don't see any reason to bash AMD on that.

If HBM is short supply (one of the reasons that been touted around as to why Vega is being delayed) couldn't AMD get Global Foundries to produce this memory? After all AMD did help develop the technology and it could be a good way of using that wafer agreement that they have to produce something helpful.

I'm also wondering why that is...

 

[benjii]

But AMD don't have a card that is faster than the 1080ti. So there's not much they can do now, is there?

What's your source for that statement? As far as I'm aware, we have no confirmed benchmarks. All we can do is speculate at this point. I've been saying since day 1 I thought Vega performance would come in between the 1080 and 1080ti, however, until AMD release solid data we can't say one way or another.

 

[JediFragger]

I'm also wondering why that is...

Special tooling maybe? It's not your run of the mill kind of item...

 

[harney]

As HG Wells said: ‘Marketing is legalised lying’.

Great quote Oh my hatred for any form of ads is stratosphere level

 

[ubersonic]

I was talking about the 980Ti, no one wants a fury.

Depends on the use, the Fury Nano has the same compute performance as a 980ti but for 75w less.

 

[Razor Time]

Depends on the use, the Fury Nano has the same compute performance as a 980ti but for 75w less.

Sarcasm totally missed.

 

[D.P.]

Claims no supply issues for HBM2, while SK Hynix themselves haven't been able to deliver any 2.0Gbps modules, despite stating mass production for it in Q3 2017. No no estimate or 2017 at all.

1.6Gbps was also mass production for Q3 2016, then Q1 2017, and now Q2 2017.
Only HBM2 that seems to be in some supply is Samsung's slower 1.4Gbps which NVIDIA have been using on Tesla, and latest Quadro GP100.

As a result Vega won't even match Fury X memory bandwidth of 512GB/s; and needs custom 1.88Gbps HBM2 to reach 480GB/s bandwidth.

Certainly looks like HBM2 issues are there; although it's clear there are other issues as well.

I don't buy the HBM2 supply not being an issue, they haven't provided any proof and as you say Hynix' product catalogue quite clearly shows major delays and missing products. It would be one thing if Hynix never even announced these products in the catalogue but they clearly had Q32016 and then a whole series of delays and cutting the faster chips altogether.

And if if it not HBM2 supply then what is it? We know the 14nm process is not going to be giving yield issue in itself. It may not be quite as good as TSMC's but the node is very mature now and AMD have a lot of experience with it form Polaris so it can't be new fab node which has caused headaches in the passed.

I don't think it can be any interposer and mounting issues, AMD already went through that nightmare with Fiji so I hope they have that sorted.


The drivers definitely can't be a real reason. AMD and Nvidia start developing drivers long before before there is even a first working silicon using advanced simulations. Plus there have been working engineering samples for months now.


If it isn't HBM2 hen it looks more liekly some fundamental design flaw that requires a whole new respin which would delay things a few months. But then they wouldn't release the professional FE cards, unless some how the flaw was only with specific graphics related functionality which they can completely block for the HPC cards. But then AMD have ben marketing the FE cards for graphics so....



SO no, by far the simplest explanation is HBM2 is in tight supply from SK Hynix. There may be some deal in place where AMD get a big discoutn form sourcing form Hynix, while Samsung might eb charging a pretty premium for Nvidia but Nvidia doesn't care when they are selling Pascal and Volta at $13K a card.