What is wrong with AMD

[CharminUltraStrong]

Quick rant,

I was excited to see the Radeon VII launch and I was hoping that the card would force the Nvidia CEO to take his harsh words back. I, like everyone else, are beyond disappointed. How is it that AMD's GPU department with all the experience they have(ESPECIALLY with how widely used they are in consoles for christs sake), can BARELY compete with Nvidia? It's disappointment after disappointment. AMD tries to compete with current gen Nvidia but ends up competing with last gen. My jaw dropped when I saw the reported price of HBM early on in the cards leaks, somewhere around dec. I think? I get it, it's good for productivity but with all of the features Nvidia has going for their cards what the hell was AMD thinking? Honestly it feels like they're running out of innovation and everything they do is beginning to feel more like a rip of Nvidia. If anyone NEEDED that much HBM2 memory why don't they just make lower tier, quadro competitors? 10 gigs of HBM2 or 8gigs would've been fine if they could make the card faster. Then there's all this marketing fluff with the "First card to be produced with a 7nm process!), this is the same bull shit they pulled with Bulldozer. First consumer CPU with 8 cores or first CPU to hit 5Ghz whatever their claim to fame was with that(Of course the core count was "faked" but that's a discussion for another da-oh wait there's a class action...). They make these bold claims of advancement but they almost, NEVER deliver. The FX-9590 was slow as shit(no offense to anyone who owns one) and got absolutely creamed by the i7 4770k and even back to the i7 2600k it was STILL getting beat! Fortunately, AMD CPU's are now a viable  option and are HIGHLY competitive, to the point where Core i3's have been rendered(virtually) obsolete. The GPU department is what bugs me because they can say they're gonna compete but they always always always come up short. RX 590, 580, 570, 560 are FANTASTIC cards with a FANTASTIC value especially for people who aren't in the US. Countries like India have difficulty accessing reasonable pricing on decent cards and those cards are great options for them. Though AMD just can't get their shit together with high end cards. Remember the R9 Fury X, how that was supposed to be revolutionary? Then they had that laughable GPU launch that for the R9 Fury Nano that they were hyping like crazy(Don't get me wrong, I think the Fury Nano is a great little card for small form factor builds cough, even Linus had a project around them though for anyone else tiny zotac 980/980ti's would be the way to go). The Vega 64/56 launch was, basically, a disaster(the odds were stacked against them I'll give them that because of crypto miner hypebeasts). Sorry, had to get this off my chest because I'm really tired of seeing Nvidia getting 0 competition and I don't think there are going to be many VII themed builds this year which sucks big time.

(PS sorry if the grammar is shite, typing this out on my phone as I leave work)

[Sakkura]

3 hours ago, CharminUltraStrong said:

 If anyone NEEDED that much HBM2 memory why don't they just make lower tier, quadro competitors? 10 gigs of HBM2 or 8gigs would've been fine if they could make the card faster.

The Vega VII is just a consumer variant of exactly what you're talking about. It's called MI50.

Crippling the Vega VII to less memory would have wrecked performance for only a modest reduction in cost/price.

 

3 hours ago, PacketMan said:

I am disappointed too, I really expected a RTX 2080 competitor, or at least a 2070 competitor, but nothing. People are talking about bad launch drivers, but that's something fucked up due to their current situation

I mean, don't blame them, they do CPUs too, Nvidia doesn't and they just focus on one thing

Wish they launch a "lite" version with less VRAM, cheaper and at least the same price/performance than the 2070

Wat.

It is a 2080 competitor, just not a great one. It easily outclasses the 2070.

[NunoLava1998]

In teraflops it's amazing tho (as in, obliterates Quadro P6000 several times over)

In gaming, it's exactly what you'd expect (about a 2070 to 2080)

[BobVonBob]

I think you've got a skewed view of what AMD is doing with the Radeon VII. They ARE NOT trying to compete with Nvidia, they're trying to make money. The Radeon VII is first and foremost a datacenter card, the Instinct MI50, that has a higher power budget and an overclock. They did this so they didn't have to spend R&D money on developing an entirely new card. As for the HBM2, it is very possible that memory can't be removed without a major redesign of the chip since HBM2 is directly on the GPU, not a separate chip like the various GDDR types.

[DrMacintosh]

Nothing. Radeon VII is a hold over and was never intended to be mass marketed and is costs AMD a lot to produce and naturally they don't really want to sell it. 

[Zando_]

Not everyone else. I had reasonable expectations, they were met, mine just arrived today and I get to pick it up after work.

 

It is a good point that they really, really struggle to compete with Nvidia’s high end, but they eat them for snack in the midrange and now that Vega prices have come down those are excellent options as well.

 

As for competing with last gen Nvidia, that’s debatable. This is made to be a showcase first to market 7nm chip and they wanted to compete with the 2080. Just because Nvidia made the 2080 the same price and same performance (about) of the 1080 Ti doesn’t mean it only competes with last gen, it competes with this gen too (also people love to say "but much RTX" when there's uh... one major title that supports some of the features? Or is it 2?).

 

Like Sakkura said, this is just the consumer/gaming version of their quadro competitors. And as Tech Yes City explained, they needed the 16GB HBM2 to hit the 1TB/s memory bandwidth they wanted to, so it does have a practical purpose other than just being overkill (the headroom is nice though, I had 16GB VRAM on my Vega FE as well and there’s pretty much nothing that will get VRAM limited, even ROTTR which pushes 8.5GB VRAM usage at 1080p).

 

And yep, like I said it’s a marketing/showpiece card. And yeah, Vega launch was a disaster, this seems to be repeating the same low stock thing, hopefully they get more a heck of a lot faster or people will just right it off as another failed Vega. And again, now that prices are down they’re excellent options, but people tend to discount them because of the poor launch and they are unavailable/overpriced for so incredibly long. Hopefully that doesn’t happen to the Radeon VII.

[CharminUltraStrong]

3 minutes ago, DrMacintosh said:

Nothing. Radeon VII is a hold over and was never intended to be mass marketed and is costs AMD a lot to produce and naturally they don't really want to sell it. 

I understand this but time and time again they botch high end gpu releases. They don't need 16 gigs of hbm2 especially if it costs 350 or 250 USD that's ridiculous! It's basically a Vega 64ti. It's disappointing because they hyped it up to be much more than that and they said themselves it'd be a 2080 "competitor". If i could take a 2070 or vii I'd take the 2070 any day. Much more robust feature set and as a consumer, 16 gigs of hbm2 won't even cross my mind in day to day use. They're trying to be a titan and 2080 all at once. 

[mariushm]

The rumor is something went wrong with Navi during manufacturing and they had to make significant changes to that chip at least once ... such modifications require that they fabricate chips from scratch and test them and all that... the whole process takes around 3 months, maybe even more.

So basically, if the chip turns out to have a major flaw that requires redesign, they have to tweak the chip and then wait around 3 months to get the new chips and test them and within 1-2 weeks from then if there's no major faults, they can go full production mode.

 

So maybe AMD was in the position of not having Navi and therefore had to come up with something to place between RX 590 and high end nvidia cards, as a stop gap.

It's probably low volume because HBM2 memory is stupid expensive still, and it's difficult to put those memories and the chip on interposer... the process to glue those chips together is not perfect so they may lose some vega chips and some hbm2 memory chips just by trying to "glue" them on the interposer...

They may be selling the boards at a loss.

 

As for "they don't need 16 GB of memory" discussion ... it may be a limitation of the chip design, like for example the chip may need 4 memory stacks, may only work well with 4096 bits of bandwidth, maybe some cores would slow down if they don't have a direct stack near them ... so you have 4 x 4 GB stacks ... it could be hynix and samsung don't make 2 GB stacks, or those stacks would be lower height which would affect heat transfer, would have to resort to other things instead of that thermal pad...  could be it requires 4 stacks for simmetry, for better heat transfer etc etc

there's a nvidia card that uses hbm2 memory and that also has 4 stacks of memory on it, even though one is inactive.

 

[DrMacintosh]

2 minutes ago, CharminUltraStrong said:

I understand this but time and time again they botch high end gpu releases. They don't need 16 gigs of hbm2 especially if it costs 350 or 250 USD that's ridiculous! It's basically a Vega 64ti. It's disappointing because they hyped it up to be much more than that and they said themselves it'd be a 2080 "competitor". If i could take a 2070 or vii I'd take the 2070 any day. Much more robust feature set and as a consumer, 16 gigs of hbm2 won't even cross my mind in day to day use. They're trying to be a titan and 2080 all at once. 

Ok so you just have a fundamental misunderstanding how what Radeon VII is. Its literally a downscaled M150 workstation card. Also HBM doesn't work like GDDR memory where you can just pick how much of it you have. 

 

AMD doesn't care if you buy a 2070, if anything they would probably rather have you buy that 2070 so they wouldn't have to make another Radeon VII. Radeon VII is a label for AMD and isn't intended to knock Nvidia out of the game. They needed something to hold them over until Navi and to do that the took a card they already had and just started selling it to consumers. 

[TheRandomness]

18 minutes ago, Sakkura said:

The Vega VII is just a consumer variant of exactly what you're talking about. It's called MI50.

 

Crippling the Vega VII to less memory would have wrecked performance for only a modest reduction in cost/price.

Considering the memory alone on the card amounts to over $100, I wouldn't just call it a 'modest reduction in cost' to reduce the amount of memory. Though, yeah, having a bus half as big would be crippling the card, I think the relatively 'huge' reduction in price could make up for it.

[Sakkura]

2 minutes ago, TheRandomness said:

Considering the memory alone on the card amounts to over $100, I wouldn't just call it a 'modest reduction in cost' to reduce the amount of memory. Though, yeah, having a bus half as big would be crippling the card, I think the relatively 'huge' reduction in price could make up for it.

People are likely overestimating the memory cost. And we're only talking about half of it. And they would still be bearing the cost of the unnecessarily large interposer with room for the extra memory.

[mariushm]

3 minutes ago, Sakkura said:

People are likely overestimating the memory cost. And we're only talking about half of it. And they would still be bearing the cost of the unnecessarily large interposer with room for the extra memory.

Memory isn't that cheap, I think a big manufacturer gets their GDDR5 memory to around 5-7$ for 1 GB ... so around 50-60$ for 8 GB of GDDR5.

GDDR6 is around 70% more expensive.

HBM2 memory is still around 2-3x the price of regular GDDR5, rumors are the 16 GB of memory costs AMD around 320$ ....   you add to that the price of the interposer (let's say 10-30$), plus the cost of extra manufacturing step (placing the chip and memory chips on the interposer)...

 

At least AMD saves some money by reusing the circuit board from the instinct version and by dropping a couple of phases (around 20-30$ worth of mosfets and inductors and capacitors)... could be a simple microcode and firmware programming of the chips, to disable some "pro" features.

[TheRandomness]

51 minutes ago, Sakkura said:

People are likely overestimating the memory cost. And we're only talking about half of it. And they would still be bearing the cost of the unnecessarily large interposer with room for the extra memory.

To be fair, my only source is that GN video where they explained the HBM2 pricing for Vega cards, where it still did amount to a significant fraction of those card's pricing. I wouldn't be surprised at all if it's still the same story for these cards.

[Max_Settings]

Why did they only make 5000? Because at a $699 price they are loosing money or breaking even on every card and Radeon VII was just to say they had the first 7nm GPU. They can make a GPU that competes with 2080 but if Radeon VII had normal margins it would be at least a $1000 card. They just don't have the capability to make the card at a competitive price. At this point you might as well forget AMD and wait for Intel to come out with GPUs. IMO Intel will compete with Nvidia before AMD ever does again.

[WMGroomAK]

1 hour ago, CharminUltraStrong said:

I understand this but time and time again they botch high end gpu releases. They don't need 16 gigs of hbm2 especially if it costs 350 or 250 USD that's ridiculous! It's basically a Vega 64ti. It's disappointing because they hyped it up to be much more than that and they said themselves it'd be a 2080 "competitor". If i could take a 2070 or vii I'd take the 2070 any day. Much more robust feature set and as a consumer, 16 gigs of hbm2 won't even cross my mind in day to day use. They're trying to be a titan and 2080 all at once. 

While not necessary for gaming, I guess the 16 GB of HBM2 do have an impact on video editing in Adobe when dealing with long high resolution files...  That might make it worth while for someone getting into performing 4K gaming and video editing as it is a cheaper solution.

 

https://www.forbes.com/sites/jasonevangelho/2019/02/07/apparently-you-really-do-need-that-16gb-of-hbm2-in-amds-radeon-vii/#1b2544389da1 

[Nena Trinity]

I fail to see the issue except price...

[Drak3]

The only thing AMD is doing wrong is their marketing. They regularly have the better mainstream cards, they can easily compete on the workstation front, and they're no slouch for servers either.

 

The only area they don't compete is high end gaming, but they don't do any significant marketing that counteracts not having the card only a niche can realistically afford.

[Tristerin]

As a consumer of AMD products Im not sure the problem.  This card is not something I will buy, just like a Quadro, or well...any new GPU costing more than the rest of the rig.  The problem may be a lot of things, but those are each individual problems - you just don't meet the criteria of what this card is aimed at (same as me, for different criteria or not).  Or you aren't filthy rich enough to not care what you spend your money on (same), because if you were you know you would buy it to see what it can do!

[_Hustler_One_]

I'll just try to light this up a litlle bit.

This Radeon VII might will be one of the rarest retro PC parts in 2030-2035 worth scavenged.

[S w a t s o n]

2 hours ago, TheRandomness said:

Considering the memory alone on the card amounts to over $100, I wouldn't just call it a 'modest reduction in cost' to reduce the amount of memory. Though, yeah, having a bus half as big would be crippling the card, I think the relatively 'huge' reduction in price could make up for it.

 

1 hour ago, Sakkura said:

People are likely overestimating the memory cost. And we're only talking about half of it. And they would still be bearing the cost of the unnecessarily large interposer with room for the extra memory.

Yea I dont think so, the 16GB of HBM2 is supposed to be something like $300 and I believe it. They can't make a vega20 with less hbm stacks and this entire thread is a meme. Everyone knew (or should have known) what to expect

[Blademaster91]

2 hours ago, valdyrgramr said:

Well, the drivers were impacting performance and causing stability issues.  AMD always tunes their drivers over time.  I've had a hot mess of drivers from both teams, so I'm just used to it from both at this point.  From 1060 crashes, 1080 crashes, 280x crashes, and 390 stability issues.  Now, the Radeon VII is just a workstation card that can game and go toe to toe with the 2080.  The 2080 gains advantage from DLSS, but DLSS and real-time RT aren't even heavily implemented right now, so, for now, both are kinda disappointing to me.  The 2080 is an early adopter card that has potential, and the Radeon VII is a great workstation card that can game while having really bad driver issues.  You also need to understand that the Vega refresh thing was aimed at productivity mainly with gaming as maybe a secondary focus.  AMD stated this several times, but people keep treating Vega 20 as if it were Navi.  Unless you plan to do some rendering in Blender or Maya, for example, but you wanna game when you're not working then this card might not be for you.  You probably want either turing or navi as those are purely for gaming, not Vega 20.  Buying this for pure gaming would have been like buying the Vega FE over the Vega 64.

The drivers were so broken with the testing GamersNexus did, even overclocking in Wattman was pretty much broken. A consumer shouldn't have to be the beta tester for drivers,and it seems AMD rushed the card out on the 7th because of the whole Radeon 7 name and 7nm marketing thing, and maybe because of such low availability.  Yeah the card is great if you also need a workstation card, but for pure gaming it doesn't make much sense because the Radeon VII runs hotter,louder and is more power hungry than the competing 2080 and 1080Ti.

1 hour ago, valdyrgramr said:

I think people keep treating the Radeon VII like it is Navi because of AMD's advertising of it.  But, if they bothered to follow AMD since they announced the new Vega cards they clearly stated the primary point of them is for them to be workstation cards.  AdoredTV has even pointed this out for quite some time.  It might also have to do with consumers wanting a pure gaming card for competition.

IMO that is AMD's fault for pushing the marketing and advertising, and it certainly doesn't help that youtubers like AdoredTV overhype these cards.

[mr moose]

Why are people so hell bent on going to extremes with these companies,   Sorry OP you are not alone and I am not singling you out, we have so many people who are invested in either damning AMD or praising them beyond reality.  The fact of the matter is AMD are doing the best they can, just like Nvidia and Intel.  There is no overarching incompetence in these companies, they aren't failing miserably at the moment. Sure AMD came close a few years back, but they got a good CEO and restructured now they are back in the game.  just like Intel is having a few issues with their 10nm but it isn't the end of their success.

 

People:

Talk about the good stuff, enjoy the fact you have options,  But more importantly buy the best you can afford and enjoy the damn thing rather than getting wound up in who's the best and who's evil and who's doing what when and why.  We are likely wrong about most of it anyway.

[Chett_Manly]

I think it is because they lack the leadership, vision, and resources to be competing in more than one area. They are trying to match the best teams at Intel and Nvidia at the same time. I wonder if ATI was spun out of AMD if a renewed mission and fresh leadership couldn't right the ship.

[Sakkura]

9 hours ago, mariushm said:

Memory isn't that cheap, I think a big manufacturer gets their GDDR5 memory to around 5-7$ for 1 GB ... so around 50-60$ for 8 GB of GDDR5.

GDDR6 is around 70% more expensive.

HBM2 memory is still around 2-3x the price of regular GDDR5, rumors are the 16 GB of memory costs AMD around 320$ ....   you add to that the price of the interposer (let's say 10-30$), plus the cost of extra manufacturing step (placing the chip and memory chips on the interposer)...

 

At least AMD saves some money by reusing the circuit board from the instinct version and by dropping a couple of phases (around 20-30$ worth of mosfets and inductors and capacitors)... could be a simple microcode and firmware programming of the chips, to disable some "pro" features.

Those rumors are from last year. Memory prices are falling, and such drops can be more pronounced for the more expensive stuff.

 

As for the interposer, AMD wouldn't be able to save money by making it smaller. That would only be worth it for a bigger production run.

9 hours ago, TheRandomness said:

To be fair, my only source is that GN video where they explained the HBM2 pricing for Vega cards, where it still did amount to a significant fraction of those card's pricing. I wouldn't be surprised at all if it's still the same story for these cards.

Well, yeah it's still going to be expensive to buy 16GB of HBM2. I'm just not convinced it's quite as expensive as rumored.

8 hours ago, PacketMan said:

It's a 2080 competitor but not what we expected

From what I've seen it's not better than the 2080 in games, and the RTX in Europe is about 700€ (considering the Radeon VII is $700 in USA I'm afraid it'll be too much expensive here in Europe)

It outclasses the 2070 but it's much more expensive, that's what I meant by a RTX 2070 competitor: some GPU with the same price/performance

While the RTX 2080 came to the market with a $800 MSRP, now you can find ones for just $700 and that's what I expected from RVII, hope they launch a version with less VRAM and lower price

It's the same price as the 2080. It does have some downsides compared to that card though, so it's not like it's a great competitor.

8 hours ago, S w a t s o n said:

 

Yea I dont think so, the 16GB of HBM2 is supposed to be something like $300 and I believe it. They can't make a vega20 with less hbm stacks and this entire thread is a meme. Everyone knew (or should have known) what to expect

It might have cost that last year, but memory prices are dropping.

[S w a t s o n]

6 hours ago, Sakkura said:

It might have cost that last year, but memory prices are dropping.

Actually we have no reason to believe that it's dropping that much with Nvidia also using HBM2 en masse for it's highest end cards (not turing because G6, I expect Ampere quadros to use hbm2 again). Literally every source I can find is saying it's near $300. Even the rumors from within AMD are saying they are making about $0 on the card. That's not because of the vega20 die, it's the HBM2