Sodapopjones
[H]ard|Gawd
- Joined
- Aug 22, 2012
- Messages
- 1,936
That's generally what happens when someone is trying to prove a point and won't admit they're wrong. I noticed he likes to plaster paragraphs of crap, rather than a concise statement.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Vega Rumors
- Thread starter grtitan
- Start date
That's generally what happens when someone is trying to prove a point and won't admit they're wrong. I noticed he likes to plaster paragraphs of crap, rather than a concise statement.
I don't care if he admits he is wrong, just the simple thought process will suffice. I can counter the latency issue, but it requires a fundamental change in architecture and how it communicates with its self and other chips, and I have a pretty good idea of how to do it too even with current architectures to maintain scalablity. But it will introduce a new type of control silicon specific for GPU's. Now I'm not an EE, but theoretically I can figure it out actual implementation lol, hell no.
The general theory is part of what came about by my posts with Anarachist, close to a year ago. It was a heated discussion but mutually I understood where he was coming from, and modified what I was thinking.
This guy wants to discuss possible rumors, but he isn't willing to understand what the current pitfalls are, which is kinda what we have to look at when the problem is there. Can't fix something without looking at the problem first :/
Dayaks
[H]F Junkie
- Joined
- Feb 22, 2012
- Messages
- 9,774
I would go ape shit for "Voltaripper"! How are they going to cool 750 watts of GPU though? That's around 600 amps...
Honestly I'd pay a couple grand for a seamless mGPU that acts like one.
_mockingbird
Gawd
- Joined
- Feb 20, 2017
- Messages
- 992
It's too bad the event tomorrow won't be livestream.
I was hoping to see Raja babbling like an idiot on stage.
Hopefully, Lisa Su fires him and bring new talents to RTG so we can have healthy competition again.
I was hoping to see Raja babbling like an idiot on stage.
Hopefully, Lisa Su fires him and bring new talents to RTG so we can have healthy competition again.
Lisa Su is actually a better CEO than I thought, she has turned the company around. AMD just posted a profit, what a shocker. Ryzen has been a huge success for them. Raja could be different though. If Vega does indeed turn out to be a turkey then he could be in hot water. It will not reflect well on him. We will see, just one more day to go.
_mockingbird
Gawd
- Joined
- Feb 20, 2017
- Messages
- 992
Exactly. Leadership can make all the difference in the world.
Can you imagine that if after the Bulldozer disaster, AMD kept the same leadership?
The company would be going down the tubes.
OutOfPhase
Supreme [H]ardness
- Joined
- May 11, 2005
- Messages
- 6,579
I have a ton of respect for Dr. Su. She's great, and perhaps I'm jaded being an engineer, but I think she's what they needed for a long time.
5150Joker
Supreme [H]ardness
- Joined
- Aug 1, 2005
- Messages
- 4,568
Console market has nothing to do with their success in the consumer GPU sector where their market has only now recovered somewhat. However, if Vega disappoints (as expected) and AMD cannot manage investor expectations because they aren't making money on it, then their stocks will dip again and so will their marketshare. Volta won't kill AMD off but it will hurt RTG really bad since Vega is all they've got and if it can't even beat Pascal then it has no chance in the world against Volta. What can they do at that point? Keep dropping prices and lose money? It wouldn't even be worth manufacturing the cards if they lose money on each sale.
She is extremely capable as a CEO, an excellent strategist.
_mockingbird
Gawd
- Joined
- Feb 20, 2017
- Messages
- 992
Su increased the headcount in RTG by 60% so it's not likely that the division was lacking resource.
Rather, what we are seeing is a failure in execution.
Rather, what we are seeing is a failure in execution.
I get the sense that there is some sort of bottleneck or severe flaw in the design. It has a similar die size and transistor count to a GTX 1080 Ti, it should be showing much better results. I'm wondering if they did not give it enough ROPs. AMD tends to give their GPUs a ton of raw processing power but not enough ROPs. I'm also wondering if GlobalFoundries could be playing a role here. I'm pretty sure Vega is being fabbed at GlobalFoundries IIRC. Perhaps had they fabbed them at TSMC they could have clocked Vega higher.
For Vega to have worse IPC compared to Fiji tells us that AMD botched the design. This is still speculation though. We won't know until tomorrow.
So nV has a holodeck, and AMD has a Holocube 
https://videocardz.com/71334/amd-showcases-radeon-holocube-and-rx-vega-64-le
https://videocardz.com/71334/amd-showcases-radeon-holocube-and-rx-vega-64-le
Vega is clocked pretty decent. I think they fucked on binning rasterizer. it simply doesn't work as they intended or they still can't get it working. Again As far as clocks, Vega seems like it's pretty decent.
I don't think they screwed up anything, I think it as too far into design for them to make any changes after seeing what Pascal can do. Pascal was not a typical increase from gen to gen, Pascal's increase when above and beyond Maxwell's overclocking performance, and that was something nV wasn't expecting, nV stated they got lucky with Pascal with clocks. So there was no way for AMD to predict that, they couldn't plan for it in Vega.
The thing is, like I said, Vega has worse IPC than Fiji. Why did they even bother? It reminds me of Tonga. Such a complete and utter waste of engineering resources.
As for AMD adding 60% more engineers to RTG...sometimes with engineering you should focus on quality over quantity. Jim Keller for example on the CPU side, he's a mad genius and I credit him more than anyone else for how well Ryzen turned out. The guy is a mad genius. He's left AMD now though. My point is that they need someone like that on the GPU side. They would probably have to poach someone from nVidia. Raja does not look up to the task.
Its going to be very hard for them to pouch for nV, they have to spend money on that.
As _mockingbird said they did increase RTG's headcount but not where it counts cause in the short term it won't matter (learning curve is too great for soon to be released products to bring on people in the middle of a project), and that is because engineers are much more expensive. AMD knows where the faults of Vega are, they knew it well before its release, they knew it in designing the thing. So hire more marketing and try to get any kind of sales out of it. Its the same approach nV did with FX, instead of focusing on the tech they pushed TWIMTBP program, it doesn't work. But in the long term it might set up a good marketing program for future products to capitalize on, as nV did. The problem with AMD in the past with their marketing programs though is they never followed through. And we still see those problems with their naming conventions and branding. Their is no stability in marketing. Lets see how it turns out. We see this in their software development programs too. When they see no traction in one aspect they drop the whole thing and move on to something else. Pretty much its money wasted, perfect example is with HPC programs. What was their first HSA program can't remember the name of it now, but it didn't do too well, so now they moved over to ROCm, if that falls short of their expectations in the short term, they will move to something else. nV with CUDA, took them what 5 years before it was monitizable, which is a typical business turn around. Then another 5 years so 10 years for DL? They spent the time and energy and money to see it through. AMD never has been able to do such a thing, even though AMD is an older company and had much tougher battles than nV ever had against Intel. They should be aware when going into competition against a larger company they need to take a step back and target specific areas of weakness against the competition. Otherwise they won't get anywhere, but we see it time and time again with AMD, they seem like they look good and then have a bad spell. This isn't the first time AMD and (ATi was too) was on the brink of bankruptcy. Granted Intel had a lot to do with the first time but still leaves a mark.
IPC the issues we see, might be just drivers. When we see 1070 level performance out of Vega, something is wrong, but I don't think its the silicon, drivers can push it up a bit, it should be getting 1080 level performance. Getting past 1080 performance now that is most likely not going to happen. But once Vega comes out we will see how many respins were needed, then we can guess on if there were any issues in the silicon
Raja is really good, if I remember correctly he was one of the principle engineers for the r300, plus he is from my home town in India, so kinda favor him for that
He left AMD due to the budget cuts as many of the other sr. engineers did too. Can't do anything without a proper budget. And I think he was promised a lot when he went back to AMD. So its not something they can probably easily torch him because of Vega. Shit any one in his position knows HBM was a waste of resources, but once its there, he can't do anything, his hands were tied. Everything so far points to one thing, not enough money. Ryzen might turn that around but its going to take time.
Last edited:
Dadebraafsie
n00b
- Joined
- Feb 27, 2017
- Messages
- 52
https://www.caseking.de/pc-komponenten/grafikkarten/amd/radeon-vega
some workstation cards available
some workstation cards available
SighTurtle
[H]ard|Gawd
- Joined
- Jul 29, 2016
- Messages
- 1,410
Preempting it. Don't believe it. It's probably hype. Hardware Canucks is pro-AMD anyway. Etc. Etc. I hope NDA ends Sunday. This is intolerable.
Magic Hate Ball
Limp Gawd
- Joined
- Jul 28, 2016
- Messages
- 366
Yeah, I'm probably going to buy Vega as a "thank you" for all the stock gains... but I'll be real surprised if it somehow gets 10% above the GTX1080 at this point....
They can't. This was a secret press event with a NDA. Price and stuff will be talked about tomorrow at AMD's presentation.
Magic Hate Ball
Limp Gawd
- Joined
- Jul 28, 2016
- Messages
- 366
Good news then.
These are the "Limited Edition" cards. The black/red one that looks like the RX480 reference should be the standard air cooled one.... if you really want a blower.
If I was to buy Vega, it would either be an AIB or Watercooled version. Not a fan of AMD reference design cooling.
I like the ref special edition look, the only thing I would change is the red V eye slits, I would put LED's there instead of paint. Cooling wise yeah they aren't the best but looks I like the industrial clean look.
Jimmy Whicher
n00b
- Joined
- Nov 10, 2013
- Messages
- 20
Console market DOES have something to do with consumer GPU sector because a. The APUs contain versions of their consumer GPUs b. Games are optimized based on consoles
As far as Volta goes, Vega might actually stall its launch if Pascal sales are still high margin and going well. They wont sell them at a loss, but my guess is they have some room to drop.
The new consoles are further steps away from current GCN chips. At least Xbox Scorpio is, and with Vega, Vega is much more different than either of the two consoles. So optimizations aren't going to be automatic. Some will port over easily some won't. This is why the only games we have seen with any kind of reasonable advantage on AMD hardware were part of AMD's game program to begin with. Goes both ways for nV's game program too.
Consoles don't drive PC sales either.
Volta is going on nV's time table from gen to gen, which is 1.5 years. Now if they delay to maintain sales of Pascal, they might be making increased margins (for more mature node) but the cost will be total volume sales, how many more people will buy Pascal after next year? There is no revolving door in GPU sales. nV can't wait for the market to stagnate, they need to push forward to maintain sales momentum. Also OEM's, board partners have to maintain sales too. So there will be some pressure from them as well, although limited, its more from keeping their sales targets in line with their bottom line.
Also have to look at the 12nm process from TSMC, it is cheaper than the current 16nm, less layers, so they might be able to get the same margins with Volta with a bigger chip. To what degree not sure, but its possible.
now with Vega launch tomorrow, if it turns out to be not so great, Q3 is the best selling quarter for hardware, you can expect nV's Q3 to be damn good, then Q4 is usually fairly good too, and of that Q is the best time to launch since Q1 is weak, but with Volta launching in that time frame they will also get a strong Q1 quarter. Makes shareholders very happy.
Last edited:
nVidia spent $3 billion developing Volta. They will bring it out when it's ready. Yields probably aren't good yet. They launched the professional cards last week. I expect we'll know something before the end of the year. It's also going to depend on GDDR6 availability, for the Volta cards that use that memory.
Well they could use the faster GDDR5x chips right now as well. GDDR5x with 12Gbs, gives 20% more bandwidth right now, so add that to a larger bus size, its can cover the need easily.
Yields are fine, at least for professional price range for an 800mm2 chip. Otherwise they won't go into mass production. Now for consumer products just taking out the tensor cores it looks to be nV's top end chips will be around 500-600mm2, which is extremely doable on 12nm (just a modified 16nm process). So if they decide to go with performance first like did for the last 2 gens, those chips will be in the 400-450mm2 range. We can see with the gtx 1080 its not bandwidth limited with the faster GDDR5x, unfortunately we don't know how much performance will it take to saturate that extra bandwidth.
Last edited:
From what I understand, the type of memory supported by the card has to be baked into the silicon. So this would have to have been a design decision nVidia made a long time ago. I hope they didn't put all their eggs in one basket with GDDR6. As you said, GDDR5X and even GDDR5 are fine for mainstream cards.
im sure they built in support for DDR5, DDR5X and DDR6 with DDR5/5X for lower end cards
That is true, but for a performance range with bigger bus size GDDR5x should be plenty, for the flag ship products most likely not though, and GDDR 6 is coming out next year and we know nV is planning on GDDR6 for their flag ship products. So it fits in with their current launch schedules.
Last edited:
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
Wait, you are gonna take stabs at the possibility of others paying too much for a level of performance, when you own a Titan. Poor argument. Granted anyone can buy whatever they want, but outside the original Titan the last 3 Titans were absolute wastes of money. Especially the last Titan, but given the fact of the TI releases shortly after cant say that buying one is ever a good idea nor is a purchaser of one in any position to throw stones at those that may mirror that decision albeit at differing price/performance tiers.
What the fuck? Rx 480 launched too high? It was 249.99 for after Market cards. If you are bitching about rx 480 price. Glad to tell you that was the only thing everyone said was good about it. Clearly your expectations were too high.