Vega Rumors

[Bahanime]

Nope, just stating how it is.

Seriously, you seem very pro NV, I get that. But why is your post history full of anti- AMD rhetoric and shill-like flame baits? Do you HATE AMD?

I figured if you love NV and enjoy buying their GPUs, you would want there to be competition so you can enjoy lower prices. The logical position for gamers who appreciate their money would be for both AMD & NV to have great GPUs, right?

 

[bezant]

No its specific slide 32 in the deck you linked to, Frontier is a pro product.

Yeah it was kind of disjointed. Why compare it to the Fury then?

 

[razor1]

cause there is nothing else they have to compare to, they can't compare to nV products in the pro segment it would make them look bad. And if they compare to the Fury Pro duo, that too won't look good for them.

 

[bezant]

New VR headsets in development are pushing for higher resolutions and maintain high frame rates. Here is the specs on the $300 Microsoft VR headset. Think of these early entries into VR as the 640 x 480 monitors of yesteryear.

Headset specs

2x LCD displays, 1440 x 1440
2.89” diagonal display size (x2)
Front hinged display
95 degrees horizontal field of view
Display refresh rate up to 90 Hz (native)
Built-in audio out and microphone support through 3.5mm jack
Single cable with HDMI 2.0 (display) and USB 3.0 (data) for connectivity
Inside-out tracking
4.0 meter cable
Sensors: Accelerometer, Gyroscope, Magnetometer, Proximity
Dimensions (L x W x H) 195.8 x 94.8 x 106.59 mm
Weight: 350 grams

Before you say, "VR who cares!" remember that we are getting new Nvidia libraries to push hardware, HDR, and more to push these GPUs. Also you can buy a Dell 8k monitor. Next year I expect that price to be sub 2K and VEGA to be within its refresh window.

Interesting that they recommend a 6+ Core Intel or 8+ Core AMD CPU, and a GTX 1060/AMD480 for developers.

 

[darkeen]

Note this render from the financial presentation

 

[Magic Hate Ball]

Note this render from the financial presentation

GDDR instead of HBM?

Maybe the art/marketing department didn't get the memo, or they repurposed an art asset for Polaris based Pro cards?

 

[darkeen]

GDDR instead of HBM?

Maybe the art/marketing department didn't get the memo, or they repurposed an art asset for Polaris based Pro cards?

Ya, I'm not sure. The art department clearly knows it's vega given the big V on the rendered silicon. I guess we'll find out at computex. The Polaris die art was angled. This looks more like a reused hawaii render if anything.

Edit:
Hawaii die for comparison

 

[Chimpee]

GDDR instead of HBM?

Maybe the art/marketing department didn't get the memo, or they repurposed an art asset for Polaris based Pro cards?

Maybe, I do recalled Robert Hallock saying Vega can use GDDR or HBM, though the most likely scenario is AMD marketing goof again.

 

[Deleted member 93354]

Aw, poor Shintai doesn't know how to gracefully admit being incorrect. It's almost as if being right or wrong isnt her/his ultimate goal...

In psychology it's called cognative disonance. The same thing happens to politics. It's actually a personality flaw where a person's brain actually induces feeling of being threatened if they are presented a contradictory information that is counter to their own beliefs.

I'm neutral territory. I slammed both the rx480 and fury. But I still believe amd makes decent stuff even if they aren't out in front. And I'm still neutral on the Vega until I see benchmarks.

I look forward to Kyle's review.

 

[Alkaladur]

In psychology it's called cognative disonance. The same thing happens to politics. It's actually a personality flaw where a person's brain actually induces feeling of being threatened if they are presented a contradictory information that is counter to their own beliefs.

I'm neutral territory. I slammed both the rx480 and fury. But I still believe amd makes decent stuff even if they aren't out in front. And I'm still neutral on the Vega until I see benchmarks.

I look forward to Kyle's review.

8Hi stacks of HBM running at 1.87Gbps will run extremely hot, it is quite natural to be skeptical of 16GB claims using only a 2048 interface. Supply will be ***extremely*** limited and it should even interfere with waterblock installation due to height. Its a surprise to everyone that 8Hi stacks exist because they do not appear on any catalogues at all.

 

[razor1]

8Hi stacks of HBM running at 1.87Gbps will run extremely hot, it is quite natural to be skeptical of 16GB claims using only a 2048 interface. Supply will be ***extremely*** limited and it should even interfere with waterblock installation due to height. Its a surprise to everyone that 8Hi stacks exist because they do not appear on any catalogues at all.

Always have risk samples and also die thinning for the height. Which would just add more cost though.

 

[Deleted member 93354]

8Hi stacks of HBM running at 1.87Gbps will run extremely hot, it is quite natural to be skeptical of 16GB claims using only a 2048 interface. Supply will be ***extremely*** limited and it should even interfere with waterblock installation due to height. Its a surprise to everyone that 8Hi stacks exist because they do not appear on any catalogues at all.

As someone in engineering I can tell you that not all products appear in catalogs for competitive reasons or they are hidden secrets.

I'm not saying that is the case here and they may be only capable of producing a couple thousand per month. But it is a plausible explination.

 

[Crosshairs]

address the topic, not the poster.

 

[Anarchist4000]

I think 8+6 is 300W in this instance with Frontier.
75W motherboard
75W 6pin
150W 8pin

Yeah sorry it may be a bit pedantic I know, because in reality 8+6 is spec as 225W/300W.
Cheers

Different applications makes sense. Presumably there is a Threadripper or Epyc with a rather large amount of GPUs attached where pulling 75W each through the motherboard isn't ideal.

Really want to see the 4/8x Vega gaming results as HBCC should make that viable with transparent memory management and less data to replicate. That's the design for the server market.

Ya, I'm not sure. The art department clearly knows it's vega given the big V on the rendered silicon. I guess we'll find out at computex. The Polaris die art was angled. This looks more like a reused hawaii render if anything.

Might be a socketed GPU/APU which was alluded to on one of the slides. Those stacks look more like MOSFETs than RAM.

 

[lolfail9001]

I figured if you love NV and enjoy buying their GPUs, you would want there to be competition so you can enjoy lower prices. The logical position for gamers who appreciate their money would be for both AMD & NV to have great GPUs, right?

The desires, whether they are mine, yours or anyone else's are irrelevant.

P. S. There is competition on high end between nV's own products, just saying.

 

[sparks]

There is competition on high end between nV's own products.
NV can not compete with itself, its just makes cards and charges as much as it can for each one. Like any other co.
They have basically had a monopoly for several years and the prices have gone insane.
I agree 100% with competition is good but AMD blowing smoke is not competition.
Last year all we heard was the 480 will be as fast as a 1080 for half the price.
F a 480 can't even beat a 1060 you call that competition, more smoke and mirrors.
This year its called a 580 and it still can't beat it, Next year will we have a 680?
The Vega fake post were talking about speed and price...and I say again, more talk and a year of BS
IF AND I SAY IF they had something to market they would be bouncing off the walls and screaming but they don't.

OH did I forget to say that AMD stock fell 10.79% this morning...the day after their super talk

 

[kalston]

There is competition on high end between nV's own products.
NV can not compete with itself, its just makes cards and charges as much as it can for each one. Like any other co.
They have basically had a monopoly for several years and the prices have gone insane.
I agree 100% with competition is good but AMD blowing smoke is not competition.
Last year all we heard was the 480 will be as fast as a 1080 for half the price.
F a 480 can't even beat a 1060 you call that competition, more smoke and mirrors.
This year its called a 580 and it still can't beat it, Next year will we have a 680?
The Vega fake post were talking about speed and price...and I say again, more talk and a year of BS
IF AND I SAY IF they had something to market they would be bouncing off the walls and screaming but they don't.

OH did I forget to say that AMD stock fell 10.79% this morning...the day after their super talk

You say NV competes with itself and then that they can't? Typo? And no, prices haven't gone insane.

 

[razor1]

well titan prices are crazy lol, that is the only place they are price gouging. Now if AMD did have performacne comparative parts in the enthusiest and high end I'm sure the prices will be a tad lower on nV's products, damn 60% margins is way above typical. Good for nV but those margins are coming from us.

 

[Digital Viper-X-]

well titan prices are crazy lol, that is the only place they are price gouging. Now if AMD did have performacne comparative parts in the enthusiest and high end I'm sure the prices will be a tad lower on nV's products, damn 60% margins is way above typical. Good for nV but those margins are coming from us.

That's because people keep buying it at that price too

 

[Archaea]

You say NV competes with itself and then that they can't? Typo? And no, prices haven't gone insane.

Those adjusted for inflation charts always seem off to me.

Around the year 2000 I was in the middle of my college years, living in Kansas City working evenings at FedEx for about $11.25 per hour IIRC. I'd imagine the pay is about the same now for a package handler. I remember a fast food worker in my area, would make about $8 an hour starting. I hear it's about $9 now.

Price of gas is higher now. Price of rent is higher now, insurance is higher now, food cost is higher now, heck my water bill used to be $35 every 2 months, now it's $80 every month. Same water service, same city.

I realize my perspective is antecdotal, but it sure seems like typical living costs rose faster than wages over those 17 years. What I'm saying is put yourself in the shoes of a 18 year old with a basic job -- it seems it would have been easier to buy a $400 card in the year 2000 than it is to buy a $700 card today.

 

[Magic Hate Ball]

Those adjusted for inflation charts always seem off to me.

Realistically. In the year 2000 I was in the middle of my college years working evenings at FedEx for about $11.25 per hour IIRC. I'd imagine the pay is about the same now for a package handler. I remember a fast food worker in my area, in the city would make about $8 an hour starting. I hear it's about $9 now.

Price of gas is higher now. Price of rent is higher now, insurance is higher now, food cost is higher now, heck my water bill used to be $35 every 2 months, now it's 80 every month. Same water service, same city.

I realize my perspective is antecdotal, but it sure seems like typical living costs rose faster than wages over those 17 years. What I'm saying is put yourself in the shoes of a 18 year old with a basic job. It was easier to buy a $400 card in the year 2000 than it is to buy a $700 card today.

You're not wrong about the cost of living vs wages.

 

[SolidSnake3035]

Pay hasn't increased along with inflation and production output, which is one of the reasons we need to have a UBI a decade or two ago... People aren't able to buy the things being produced.

 

[Rizen]

Those adjusted for inflation charts always seem off to me.

Around the year 2000 I was in the middle of my college years, living in Kansas City working evenings at FedEx for about $11.25 per hour IIRC. I'd imagine the pay is about the same now for a package handler. I remember a fast food worker in my area, would make about $8 an hour starting. I hear it's about $9 now.

Price of gas is higher now. Price of rent is higher now, insurance is higher now, food cost is higher now, heck my water bill used to be $35 every 2 months, now it's $80 every month. Same water service, same city.

I realize my perspective is antecdotal, but it sure seems like typical living costs rose faster than wages over those 17 years. What I'm saying is put yourself in the shoes of a 18 year old with a basic job -- it seems it would have been easier to buy a $400 card in the year 2000 than it is to buy a $700 card today.

When I was 18 I wasn't buying $400 GPUs. lol

 

[Archaea]

When I was 18 I wasn't buying $400 GPUs. lol

no, I wasn't either, but I bought plenty of $200 - $300 GPUs over the years since high school to now. (graduated 1997) I usually bought a new graphics card every generation or two.

Am I remembering correctly that a 3dfx voodoo card was about $300 in the mid/late 90s? That was lawn mowing money for me --- high school timeframe. I bought a Orchid Righteous 3d card, and it was a pretty penny, but it was obtainable. (yes checked -- $300 was launch price for both original voodoo cards and voodoo 2 cards) --- EVERYBODY who was into PC gaming that I knew had a 3dfx card back in those days. So yeah -- proves the point. By contrast very few people I know are willing to drop $700 on a card (even as adults), but nearly everyone back then was willing to drop $300. So the translation to today's wages doesn't work. (admittedly, at least with my subjective, small, anecdotal sample size of friends).

I usually bought mid-upper tier throughout the years and $200-$300 was always common outlay for those cards.

 

[Algrim]

Inflation figures don't cover all sources of higher costs. For instance, several years ago I paid less than 100 a month for employer-subsidized health care for my wife and myself with a 300 deductible and almost all additional services paid. Now it's 125 or so a month for an 80/20 plan with a 700 deductible each, co-pays for medications have doubled, etc.

So when we were told that cost of living was only 2% that ignores health care.

 

[CSI_PC]

As someone in engineering I can tell you that not all products appear in catalogs for competitive reasons or they are hidden secrets.

I'm not saying that is the case here and they may be only capable of producing a couple thousand per month. But it is a plausible explination.

Yeah if one has the cash.
I have worked for various tech engineering-manufacturers over 25 years and I do not remember parts/components being kept out of the catalogue/parts reference unless it is a bespoke component usually based upon tech with a more standard offering (Nvidia's 12N is one very recent example) or one buys the whole years manufacturing calendar and payment is mostly in advance and at a premium (unless your Apple it seems lol); even Apple who buys a lot of the manufacturing calendar order schedule in advance do not have said components hidden from the reference books from manufacturer when they buy all the very 1st batches - specific context is components such as memory/glass panels/capacitors/etc
Either way we are talking serious amount of current cash, which unfortunately AMD does not have.
What I am mulling over is this for SK Hynix; how much would they need to charge AMD to focus on such tech at the expense of being able to get the 1.6Gbps HBM2 out the door for orders that has now been delayed 6 months?
Samsung are now starting to get their HBM2 integrated with network processors, a segment SK Hynix will want very much themselves as they go into high margin network/backbone/backplane related products.

Cheers

 

[SighTurtle]

SK Hynix and AMD co-developed HBM together didn't they? Maybe part of the deal was first priority.

 

[kalston]

no, I wasn't either, but I bought plenty of $200 - $300 GPUs over the years since high school to now. (graduated 1997) I usually bought a new graphics card every generation or two.

Am I remembering correctly that a 3dfx voodoo 2 card was about $300 in the mid/late 90s? That was lawn mowing money for me --- high school timeframe. (yes checked -- $300 was launch price) --- EVERYBODY who was into PC gaming that I knew had a 3dfx card back in those days. So yeah -- proves the point. By contrast very few people I know are willing to drop $700 on a card (even as adults), but nearly everyone back then was willing to drop $300. So the translation to today's wages doesn't work. (admittedly, at least with my subjective, small, anecdotal sample size of friends).

I usually bought mid-upper tier throughout the years and $200-$300 was always common outlay for those cards.

Your case is interesting since my personal experience is the complete opposite lol. A $300 GPU was considered absolutely outrageous here in 2000, now I see those people complaining in the same manner about $600+ GPUs (and a few years back it was $400-500 that made people cry). But $300-400 has become "reasonable" and every serious gamer around me has a gtx 970 or 1070 (depending on when they were upgrading since they only do so every few years). Also a pretty good number of 1060s, especially students or people who game less. That's middle-class in France to put things into perspective - so not just a different country but also currency, heck we even had a different, weaker, currency in early 2000 compared to now.

 

[CSI_PC]

SK Hynix and AMD co-developed HBM together didn't they? Maybe part of the deal was first priority.

That was HBM rather than HBM2.
But the point is SK Hynix is suffering as they are not even launching 1.6Gbps and delayed now for consecutive quarters, for them AMD would need to pay a lot to not focus on this (if AMD is using different-bespoke product)
How much do you think Samsung is making from Nvidia with the P100 implemented in DGX-1 nodes and into large scale datacentres/HPC and the scheduled V100 massive projects even this year?
And then consider how much Samsung is going to make having their HBM2 in network processors (and creating the partnerships) or from major FPGA manufacturers such as Xilinx that is even larger scale sales, all the while SK Hynix is MIA.
Other major uses likely to see with solutions similar to Stratix 10.

Cheers

 

[Bahanime]

8Hi stacks of HBM running at 1.87Gbps will run extremely hot, it is quite natural to be skeptical of 16GB claims using only a 2048 interface. Supply will be ***extremely*** limited and it should even interfere with waterblock installation due to height. Its a surprise to everyone that 8Hi stacks exist because they do not appear on any catalogues at all.

Here's a thought for folks who depend on the public catalogs. AMD in Q4 financial claimed they spent ~80M on securing supply, before this, SK Hynix advertised 8Hi, full speed HBM2. After this, they removed the best HBM2 from their public catalog. SK Hynix + AMD worked closely on HBM2, it stands to reason there's a benefit for AMD here. You don't devote resources to a shared venture without benefits. LIkewise, NVIDIA gets exclusive GDDR5X with Micron, and likely, first dips for GDDR6 fastest modules too.

The point here is, sometimes what you see in the public domain, is not the 100% truth as there's hidden deals going on behind closed doors.

 

[razor1]

Here's a thought for folks who depend on the public catalogs. AMD in Q4 financial claimed they spent ~80M on securing supply, before this, SK Hynix advertised 8Hi, full speed HBM2. After this, they removed the best HBM2 from their public catalog. SK Hynix + AMD worked closely on HBM2, it stands to reason there's a benefit for AMD here. You don't devote resources to a shared venture without benefits. LIkewise, NVIDIA gets exclusive GDDR5X with Micro, and likely, first dips for GDDR6 fastest modules too.

The point here is, sometimes what you see in the public domain, is not the 100% truth as there's hidden deals going on behind closed doors.

It was never in the catalogue as a specific date it will be there, it was always TBD from what I remember and even now its still a TBD

Nothing has been removed from the catalogue, just that HBM2 hasn't been updated with a specific date. So I'm thinking they are using risk wafers for the first few initial batches, and that makes sense with the low availability rumors of Vega.

 

[TaintedSquirrel]

So it looks like the Q2 Vega release may have been referring to the Pro cards only, and the gaming chips will come later in Q3 or beyond. Lots of salt on Reddit today.
But despite all of this, I still find myself asking... # Where's Vega?

 

[NKD]

So it looks like the Q2 Vega release may have been referring to the Pro cards only, and the gaming chips will come later in Q3 or beyond. Lots of salt on Reddit today.
But despite all of this, I still find myself asking... # Where's Vega?

you are falling for the same crap. Why the hell is no one waiting until computex. If there is no vega announcement for gaming there then yea criticize all you want and we know for sure they fucked up big time throwing all their eggs in HBM2 bucket. I will reserve my opinion until computex. Until then I think this Q3 thing holds no truth and people are over reacting. Back it up with a firm statement from AMD and I call it a fact! But so far no one should be complaining about them not announcing any gaming cards at analyst day given they have already sent invite for computex event.

 

[CSI_PC]

It was never in the catalogue as a specific date it will be there, it was always TBD from what I remember and even now its still a TBD

Nothing has been removed from the catalogue, just that HBM2 hasn't been updated with a specific date. So I'm thinking they are using risk wafers for the first few initial batches, and that makes sense with the low availability rumors of Vega.

Well Ironically SK Hynix in Q4'16 removed the 1.6GBps HBM2 and kept 2Gbps in
And then to return Q1'17 with 1.6Gbps (keeps slipping since) but now 2Gbps HBM2 is removed and never appeard again.

However and importantly, like you say TBD was in place up to and including Q4'16.
When the Q1 and Q2'17 catalogue showed just 1.6Gbps it was with an actual avail estimate that matched the quarter of the catalogue, showing it keeps slipping its expected date.
But yeah only 1.6Gbps has ever had a scheduled date against it.

CHeers

 

[NKD]

Well Ironically SK Hynix in Q4'16 removed the 1.6GBps HBM2 and kept 2Gbps in
And then to return Q1'17 with 1.6Gbps (keeps slipping since) but now 2Gbps HBM2 is removed and never appeard again.
CHeers

That's the thing lol. Something is going on. It may be that AMD booked all the memory and got first dibs so they likely removed those. I mean it's unlikely they can't get their shit together every quarter. Lol

 

[Rizen]

no, I wasn't either, but I bought plenty of $200 - $300 GPUs over the years since high school to now. (graduated 1997) I usually bought a new graphics card every generation or two.

Am I remembering correctly that a 3dfx voodoo card was about $300 in the mid/late 90s? That was lawn mowing money for me --- high school timeframe. I bought a Orchid Righteous 3d card, and it was a pretty penny, but it was obtainable. (yes checked -- $300 was launch price for both original voodoo cards and voodoo 2 cards) --- EVERYBODY who was into PC gaming that I knew had a 3dfx card back in those days. So yeah -- proves the point. By contrast very few people I know are willing to drop $700 on a card (even as adults), but nearly everyone back then was willing to drop $300. So the translation to today's wages doesn't work. (admittedly, at least with my subjective, small, anecdotal sample size of friends).

I usually bought mid-upper tier throughout the years and $200-$300 was always common outlay for those cards.

It's still kind of apples and oranges. Today's GPUs are orders of magnitudes more complex than GPUs in the 90s and have expanded significantly both in scope (now they do GPGPU, physics, massively parallel HPC, etc) and die size.

To put this in perspective, NVIDIA's Volta is 815mm^2 and 21 billion transistors. Intel's $1600 Core i7-6950X is 3.4 billion transistors and 246mm^2. Volta is over 3x physically larger and packs about 7 times more transistors. By comparison, the 3dfx Voodoo2 had 4 million transistors, vs 7.5 million transistors on the Pentium 2 (release in 1997) and 9.5 million transistors on the Pentium 3 (released in 1999). As graphics cards have gotten more complicated, they've gotten more expensive to produce, and thus cost more to consumers.

Bear in mind that $200 GPU today basically allows you to max out games at 1080p, whereas a $200 GPU back in the day was definitely more of a compromise product.

 

[CSI_PC]

That's the thing lol. Something is going on. It may be that AMD booked all the memory and got first dibs so they likely removed those. I mean it's unlikely they can't get their shit together every quarter. Lol

I edited my post.
Remember the 1.6Gbps was the one that was removed in Q4, so are you saying they bought that?
Only the 1.6Gbps has ever had a date and that goes back to Q1'17, the 2Gbps was always TBD up to the point it never returned.
And with the way TBD was used and with the available dates slipping each quarter in the official book for 1.6Gbps suggests Sk Hynix is having problems.
Cheers

 

[razor1]

I edited my post.
Only the 1.6Gbps has ever had a date and that goes back to Q1'17, the 2Gbps was always TBD up to the point it never returned.
And with the way they changed while TBD and with the dates slipping each quarter in the official book for 1.6Gbps suggests Sk Hynix is having problems.
Cheers

Ah I was only thinking of the 2 Gbps lol. Damn keep thinking GBP, getting confused with Forex lol.

 

[Bahanime]

So it looks like the Q2 Vega release may have been referring to the Pro cards only, and the gaming chips will come later in Q3 or beyond. Lots of salt on Reddit today.
But despite all of this, I still find myself asking... # Where's Vega?

Computex in less than 2 weeks. E3 in June too.

Let's stop pretending with certainty, like you guys can read the future.

Doom & gloom after Computex & E3 all you like though.

 

[NKD]

I edited my post.
Only the 1.6Gbps has ever had a date and that goes back to Q1'17, the 2Gbps was always TBD up to the point it never returned.
And with the way TBD was used and with the available dates slipping each quarter in the official book for 1.6Gbps suggests Sk Hynix is having problems.
Cheers

Idk man. It's most likely amd is booking them and getting whatever they make. Idk Samsung has had it since 2016 or was it late 2015? It's shameful if Hynix can't get their shit together in 2017