Poll: ** The AMD VEGA Thread **

[DrBombcrater]

I wonder what today would be like if AMD never bought ATi...

ATI would have died, or been bought up by a large company for their IP and patents. When AMD came knocking ATI was preparing to launch the R600 GPU in the form of the Radeon 2900XT card. As history records, that was a catastrophic failure and there was a very strong chance it would have sunk ATI if they didn't have AMD money to keep them going - AMD was very successful in those days, and kept ATI going until they became competitive again with the Radeon 4000 series.

The real 'what if' scenario was that AMD intended to buy NVidia, but couldn't agree terms with NVidia's CEO.

 

[malachi]

ATI Rage 2 SLI'ed with Voodoo 2, after selling my Voodoo Rush......fun times!

 

[CAT-THE-FIFTH]

i


its people that create the hype, and again people expect this card to be a great gaming card, yet AMD have said this is NOT! A gaming card.

They're going up against intel and nvidia with half the R&D budget, yet still producing great product. Personally i can wait, and if it is around 1080 performance then ill be happy with that. Progress is progress imo

Its nothing to do with hype - its a £1000 card with a 500MM2 GPU. What if Nvidia launched the Titan X and Titan Xp and said this is not really a true gaming card,but it can game,but its more for other stuff even though its not certified for professional stuff,but we will be selling a professional version later and then gaming performance was barely better than a GTX980TI and AMD had a year old card which was faster??

Nvidia had just over 80% of the market at one point when AMD did have highish end cards,and even with the Fury X in tow,AMD managed to get to 30% which is historically still very low.

ATI marketshare never even fell that low when they had the HD2000 and HD3000 series and Nvidia were kicking their backsides with the G80 and G92 series. ATI had missteps but at least they seemed to have a better idea on average how to pitch their products and could actually have near perfect launches.

Look at the HD7970 which was massively underclocked at launch.

It was 40% faster than the HD6970 and was on a new node,and overclocked you could gain a good performance uplift,and at stock clocks consumed less power than an HD6970:

https://tpucdn.com/reviews/AMD/HD_7970/images/power_average.gif

Look at the performance of the Vega Frontier Edition by comparison.

If it were to follow the same performance uplift the HD7970 had over the HD6970 it would need to faster than a GTX1080 consistently.

Except all we are seeing is in-between GTX1070 and GTX1080 level performance and a big increase in power consumption over the Fury X.

PCPER said around 280W running The Witcher 3,now look at the Fury X:

https://www.pcper.com/image/view/79393?return=node/67284

Its around 200W to 240W.

Even if its a combination of poor drivers and a poor cooler which is causing the relatively low performance,why is AMD launching a card in this state??

It appears they have learnt nothing from the last few years.

FFS,they should have the honestly to actually tell people if the gaming drivers are not up to scratch,but they don't to save face,so it leads to this current debacle.

They better try and make sure the RX Vega is launched with much better drivers,otherwise it only takes Nvidia to drop the prices a bit on their current cards to compete,and by that time people won't bother waiting again for another AMD card.

 

[zornyan]

ATI would have died, or been bought up by a large company for their IP and patents. When AMD came knocking ATI was preparing to launch the R600 GPU in the form of the Radeon 2900XT card. As history records, that was a catastrophic failure and there was a very strong chance it would have sunk ATI if they didn't have AMD money to keep them going - AMD was very successful in those days, and kept ATI going until they became competitive again with the Radeon 4000 series.

The real 'what if' scenario was that AMD intended to buy NVidia, but couldn't agree terms with NVidia's CEO.

I do honestly think that amd could benefit by selling off the gpu section.

first off guys I'm not saying this to be all doomsday like.

trying to do both at once has clearly hurt amd broadly speaking over the years, and I do think if ati/radeon was sold to another company that would invest as a gpu only company it could benefit both.

amd would get some much needed money from the transaction to keep funding cpu development.

Radeon could be bought out and have a good deal of investment and be real competition for nvidia.

 

[five8five]

My memory is poor but did anyone see the card when Vega was demoed all that way back? I am just wondering if they didn't do something naughty, as things are not really adding up with what we are expecting.

I was thinking the same thing! There was very little information given and they disguised the card.

It was probably a Titan in Vega shroud

 

[nashathedog]

at this point amd should sell Radeon off, literially buying ATI was one of the worst decisions the company ever had, it basically nearly bankrupt amd, they havnt gotten out of it what they needed.

clearly they can't R&D for both divisions at once, even with much bigger budgets than they currently have.

if they release a good gpu, the cpus are ****, if they release a good cpu, the gpus are ****.

I'd say thank God they bought ATI, If not for the GCN 1.0 cards like the 7000 series they'd of likely gone under with just Bulldozer products.

We all knew that AMD's development teams were focused on Zen as we were all speculating on the topic leading up to the release of the 400 series where there was just low and mid range cards.
AMD told us as much so really we all knew the GPU teams were struggling to keep up with Nvidia, As always it's the hype train that's taken the lead.
Vega was developed during the lean years so expecting so much was pretty silly.
Now Ryzens out we should see more funds going to RTG but we won't feel the effect for another 2 or 3 years.
I'm only running uw1440 so 1080 performance will be fine, my concern is where AMD's going to price it at.

 

[five8five]

While I can see his point. But if he was taking requests then time isn't as tight as it sounds. Also how are they not curious about this rubbish performance and the regression. It's their job. When the written review comes out, we'll see if they are worth their salt, or if they are just going to spit out numbers and call it done.

Think he mentioned quick tests. He also said he was going to do more tests later.

 

[zornyan]

ATI Rage 2 SLI'ed with Voodoo 2, after selling my Voodoo Rush......fun times!

playing quake, cs source.....half life 1.....god I miss the 90s.

 

[CAT-THE-FIFTH]

I'd say thank God they bought ATI, If not for the GCN 1.0 cards like the 7000 series they'd of likely gone under with just Bulldozer products.

We all knew that AMD's development teams were focused on Zen as we were all speculating on the topic leading up to the release of the 400 series where there was just low and mid range cards.
AMD told us as much so really we all knew the GPU teams were struggling to keep up with Nvidia, As always it's the hype train that's took the lead.
Vega was developed during the lean years so expecting so much was pretty silly.
Now Ryzens out we should see more funds going to RTG but we won't feel the effect for another 2 or 3 years.
I'm only running uw1440 so 1080 performance will be fine my concern is how AMD's going to price it.

The problem is that their PR just screwed this up - they could have easily sold this card directly from the AMD website in limited quantities as a first edition card for people who were not gaming and for those who want to "preview the latest GCN features" for developing games on,etc and put a statement out on the website that it was not indicative as final gaming performance,as it was an early release preview card for feature validation,etc.

How can it be any hype train when a Fury X according to TPU is quite close to a GTX1070 already.

I mean not even beating a GTX1080 consistently would make this a lower jump than the R9 290X/390X to the Fury X on a new node and with two generations more newer tech.

They have a habit of panic rushing out stuff to meat deadlines,when TBH if they actually launched it when it was actually properly working at a later date,it would at least a better impression.

Now until the RX Vega is out AMD will be in damage limitation mode,and if RX Vega does not have a decent performance bump,99% of people will probably ever want to wait for an AMD card again and just look at what is currently available on the market,ie,mostly Nvidia cards.

 

[five8five]

Yes very strange indeed
to see some doom results on the FE so to compare to the tests done on the black gaffa taped VEGA 6 months ago would have been great

PCper seemed to have a lot of dialogue with AMD so I wonder if AMD asked them not to test Doom on the FE to save face from the potential fallout as it's one of the few performance indicators they released.

 

[alec]

I love hype train crashes. Loud and spectacular.

Drivers will save the day, believe it.

 

[CAT-THE-FIFTH]

PCper seemed to have a lot of dialogue with AMD so I wonder if AMD asked them not to test Doom on the FE to save face from the potential fallout as it's one of the few performance indicators they released.

This is the Fury X all over again,but potentially worse,as they panic rush stuff out to meet some internal deadline,and yet again its half baked and destroys any momentum they might have built up. AMD has so much goodwill from Ryzen,Intel having issues with X299,etc and like clutching defeat from the jaws of victory,AMD seems to then manage to give their other competition(Nvidia) even more room.

Seriously people are talking about hype trains - expecting Vega to match or slightly exceeding a 314MM2 GP104 based card,launched over a year ago with a 500MM2 GPU based card,with double the VRAM,50% more memory bandwidth,with a decent performance/watt improvement over Fiji is not a hype train,its a hype tram.

You are talking around 30% more performance over a Fury X FFS!! How can anybody think 30% maybe 35% is much after two years??

A hype train would be expecting it to match or beat a Titan X.

I have been using a GTX1080 since last year!!

 

[DragonQ]

I love hype train crashes. Loud and spectacular.

Drivers will save the day, believe it.

If the driver was good, the train wouldn't have crashed.

 

[CAT-THE-FIFTH]

If the driver was good, the train wouldn't have crashed.

 

[D.P.]

We really should put this to bed. It's not a PRO card nor a consumer card. It's that grey in-between.

Let me put it like this:

Suppose you do job X and at work you use (courtesy of your employer) a Quadro X or RadeonPro X that costs $4000 on a 6900K.

Want to do some stuff at home but can't afford that setup? You can buy a Threadripper + Vega FE.

It's a niche, but it exists...

But then you could also buy a Quadro P4000 for much less cash for example


And you also have the issues that the professional software often costs $4000-10,000, so saving a little on the GPU can be dwarfed by software license. Of course pirated and hacked software exists but the software pricing alone put a into perspective the hardware costs an explains why people would pay $2k for a Quadro, because it is peanuts compare to the software they are using.
You then have to factor if someone is using illegal pirated copies of SolidWorks or whatever for their hobby then they would also have no issues in softmoding their Geforce cards to allow the Quadro drivers to install.

Then there is the fact nothing stops you gaming on a Quadro or Pro Radeon card. The drivers almost the same. in benchmarks you see no performance difference between the Quadros and their geforce counterparts. So it is not like this is some new, unique disruptive selling point. Its never been the case of one or the other, merely no sane person would pay pro prices to get a card to game with.

And for this niche to work it would still need to have gaming performance up their with the 1080ti to make real sense. In that case I could imagine there is a niche of people with some productivity software who don't need they certification but also want to game.



I actually think the biggest niche could be deep learning hobbyist and small start-ups. the double rate FP16 performance could be a huge performance gain over any Nvidia card outside the insanely expensive P100 (none of the others support double rate FP16). The problem here is mNvidia are rally runnign away with the deep elarning industry support like CUDA. Software like tensorFlow and Caffe support CUDA accelerated neural network learning and inference, but at least the last time I checked they didn't support OpenCL and AMD GPUs. that stuff is supposedly changing and AMD have talked about new support so Vega wont be shut out completely, however, the industry develops around Nvidia CUDA and anything is a side thought if there happens to be resources available.

I do a little deep learning and related machine learning technologies with my employer. We have been using simply an Maxwel 980 for development, but moving code to AWS to get scaling. If Vega offered genuinely great deep learning performance and integration with existing sftware was properly supported and kept up to date it could be an attractive option.

 

[zornyan]

PCper seemed to have a lot of dialogue with AMD so I wonder if AMD asked them not to test Doom on the FE to save face from the potential fallout as it's one of the few performance indicators they released.

apparently the original guy doing the first reviews tested doom yesterday, got 60-70fps on his VEGA FE, basically identical to the showcase back in January.

saw this on reddit.

 

[harney]

I love hype train crashes. Loud and spectacular.

Drivers will save the day, believe it.

Well from what i have seen allready they have a long way to go to get the power down and performance up ...I am hoping its a flaw with the rev chip and the reason why we have the FE card ...i hope that the RX vega is a another rev stepping ...but this is just HOPE... I really do want AMD to succeed in the GFX department for all are sakes....

 

[harney]

If the driver was good, the train wouldn't have crashed.

 

[CAT-THE-FIFTH]

Well from what i have seen allready they have a long way to go to get the power down and performance up ...I am hoping its a flaw with the rev chip and the reason why we have the FE card ...i hope that the RX vega is a another rev stepping ...but this is just HOPE... I really do want AMD to succeed in the GFX department for all are sakes....

Yeah,not being able to match a GTX1080 on average whilst consuming more power than a Fury X on a new node,is a disappointment especially when the Fury X is quite close to a GTX1070FE according to TPU and probably hitting a VRAM bottleneck too. I hate to think how this would look if the Fury X was an 8GB card.

 

[EsaT]

I do honestly think that amd could benefit by selling off the gpu section.

first off guys I'm not saying this to be all doomsday like.

trying to do both at once has clearly hurt amd broadly speaking over the years, and I do think if ati/radeon was sold to another company that would invest as a gpu only company it could benefit both.

And kiss goodbyes to chances of staying in best position for future console CPU/GPU market?
Which has kept AMD going in past years.

What hurt AMD was Intel's dirty and illegal market tactics when AMD had superior CPU while Intel had its failure with NetBurst/Pentium 4.
http://jolt.law.harvard.edu/digest/intel-and-the-x86-architecture-a-legal-perspective

After that AMD had similar architectural failure with Bulldozer, killing their market share in high profit margin high end worstation/server markets.
Ryzen is that long needed answer for getting CPU business competitive.


As for graphics cards HD5000-serie had superior architecture to Nvidias Fermi (with fake cards) which consumed heck lot more power.
But instead of setting up lynching party lots of people completely disregarded heating Fermistor's power consumption.

While actually "reference always bad - non-reference good" fashionable factory overclocked GP102 cards now consumer more power than Fermi.
Neither was anyone calling for lynching party when Nvidia had all those manufactured to fail cards during "bumpgate".
(with people baking their cards in oven to reflow broken solder joints)