Where the hell is NAVI???????????

[humbug]

I'm with Bru on this, if there was a 'significant' core count difference then i would agree a portion of the GPU is dud and salvaged, but they both have the same number of compute units actually clocked slightly higher on Vega VII, and they both have 16GB of memory.

 

[CAT-THE-FIFTH]

It could be the GPUs used for the Radeon 7 are also leakier,ie,you can push more voltage through them and they clock better,but at the expense of efficiency.

 

[bru]

Unfortunately it one of those thing we will never know for sure, all we know for certain is they are expensive to make, as when Buildzoid did his PCB layout he goes into some detail about the expensive components.

Anyway, this thread is supposed to be about Navi, which might not be all quite as rosy as it could be. Now whether this means not as fast, or not as efficient, or not as cheap, or just not ready in the expected timescale or any other issues that we haven't thought of, we just don't know.
Their might not be any thing wrong at all, all we have is a tweet from Jim at adoredTV that says

Well two phrases I've heard in the last week from two different sources were "Navi horror stories" and " Navi has been a nightmare".

Grain of salt, as usual.

 

[CuriousTomCat]

"been" meaning past tense so perhaps these issues have been resolved, I hope. Navi needs to be a success.

 

[ANDREW GREEN]

This is true but what you're not considering is where it sat in AMD's new 14nm line up not how it compared against it's namesake, The 480 slotted in where the 390 & 390x sat in the previous line up and at the time it was generally slower than those cards, Then you also need to consider how it took around a year to improve on it in the 14nm lineup. As for top level Navi I think you're right, from what I've seen I think it'll only offer 14nm Vega performance, It could be that there's a completely different high-end Navi chip coming to market as well, each with multiple skus but I've seen nothing to suggest this other than unsubstantiated rumours.

Didn't the 480 have a significantly lower SRP on par with the 380 though? 480 also had half the memory bus width and 20% less shaders so it was doing less with more. Obviously mining pushed up prices of both cards. I think instead of making a 490 and a Vega card, they went for 2 Vega Cards instead

If we compare on the top end Vega 64 was 40-50% faster than the Fury X.

With new architecture, Variable Rate Shading, faster RAM alongside 7nm and I'd expect 40-50% for replacement cards.

 

[LePhuronn]

So Ill just ask what part of the Radeon V11 is a failed chip?

Personally I have no idea, as far as I can tell, if it works then it could be used for either.

So you missed the bit in the specs that shows MI50 requires 8 pin and 6 pin PCI-E connectors and is passively cooled? Try doing that with a Radeon VII. I'd say that's a pretty clear indicator that the Vega+HBM2 package failed its power draw validation, that is what's broken.

Also, Radeon VII's double-precision performance is half that of MI50, likely another point of failure in the MI50 validation; there's no way that bit of GPU silicon can be disabled after the package has been assembled and tested.

And to me, the biggest suggestion that these are failed MI50 packages is this simple fact: why would AMD cripple their profit margin by redirecting a perfectly valid and working professional product into a gaming card? They wouldn't.

 

[Conor.]

uh oh

If this is true and it ends up with Navi not been able to compete with the 2070 I’ll be disappointed and just got with the best price/performance from AMD

 

[LePhuronn]

Anyway, back to Navi...

Personally I don't see AdoredTV Jim as some sort of guru, but a lot of his information makes logical sense and unless his sources are literally Lisa Su, Mark Papermaster and Uncle Jensen, he's going to get volatile, unconfirmed and outright duff information from time to time. At the very least his sources aren't going to be privy to every nuance and change, so they could feed Jim some information which really was the plan at the time, but not be aware of a different direction or technical issue which changes that plan, which then makes the leak look bogus.

Is that's what happened with Navi? Fudzilla said mid-October than their sources had Navi already testing in the lab and performing "better than expected" and ahead of schedule for the H2/Q3 launch. Good sign, but if that was the case why no CES demo? Ryzen is still 6 months away from release but we had a demo for that. AdoredTV then say there was a previously-unidentified issue with Navi which required a retape, but his sources said that was end of September. No way of corroborating one way or another, but seems reasonable to me given that it was a statement made in isolation.

But now AdoredTV is insinuating something's gone apocalyptic? Those phrases he quotes are very much past tense, so there's nothing to suggest Navi continues to be delayed by whatever bug apparently showed up. But unlike that bug suggestion, this latest tweet is made against the backdrop of significant price cuts to the Vega 56 at OcUK which actually don't make any sense.

Gibbo says that AMD requested the price drop, and cited a response to the 1660 Ti as the reason. But why? Better-performing cards cost more, plus these new prices really question getting a RX 590. But Gibbo's also cited RX Vega is EOL, and many times justified price gouging high-demand EOL stock as "maintaining sales expectations". So then, if he's selling 600+ Vega 56 units at the new crazy price but it's an EOL product, at what point is his stock going to run out and then bumps the price back up? Is Vega 56 actually EOL, or are AMD clearing stock as fast as possible (Lisa Su has made suggestions about having stock "left in the channel" when talking in vague terms about Navi)?

This says to me the following:

Vega 56 production has been ramped back up to cover the 1660Ti because Navi is broken and going to be late again.

Or

Vega 56 stock is being shifted fast because the RX 3070 is coming soon.

 

[Brazo]

I would like to think that the end result of the Radeon VII was due to bodging a failed instinct chip rather than actually any effort or R&D on AMD’s behalf, otherwise we should be very worried indeed.

 

[LePhuronn]

I would like to think that the end result of the Radeon VII was due to bodging a failed instinct chip, rather than actually any effort or R&D on AMD’s behalf...

MI50 and Radeon VII are just too similar to think anything else, unless it's a repeat of Vega 10 and a singular design was intended to be used in both compute and gaming cards. Which I don't see being the case given the suggested cost of manufacture and Mike Rayfield's 7nm Vega consumer project allegedly described as "unfeasible".

...otherwise we should be very worried indeed.

Why? Vega 20 is an evolution of Vega 10 coming from a company that has zero budget and facing bankruptcy for a long-ass time. You're not going to see miracles. And the same applies to Navi: the leaks make sense for re-establishing what mid-range performance and cost should be in 2019, but again it's intended to be a small and cheap, mid-range product to fully transition RTG into cost-effective 7nm mass production and close the door on the Koduri era, not the Second Coming of Christ.

Now if Arcturus in 2020 is still underpowered, overly hungry, hot and generally crap then we should be worried. Anything else is premature doom and gloom.

 

[Brazo]

MI50 and Radeon VII are just too similar to think anything else, unless it's a repeat of Vega 10 and a singular design was intended to be used in both compute and gaming cards. Which I don't see being the case given the suggested cost of manufacture and Mike Rayfield's 7nm Vega consumer project allegedly described as "unfeasible".


Why? Vega 20 is an evolution of Vega 10 coming from a company that has zero budget and facing bankruptcy for a long-ass time. You're not going to see miracles. And the same applies to Navi: the leaks make sense for re-establishing what mid-range performance and cost should be in 2019, but again it's intended to be a small and cheap, mid-range product to fully transition RTG into cost-effective 7nm mass production and close the door on the Koduri era, not the Second Coming of Christ.

Now if Arcturus in 2020 is still underpowered, overly hungry, hot and generally crap then we should be worried. Anything else is premature doom and gloom.

Well I think you answered your own question

Absolutely it’s the best they can do. That doesn’t make it ‘the best’ though. Upon release of Vega 1 I heard too much ‘wait for Navi’ and now it’s wait for Arcturus. People can only wait for so long given what they want already exists.

 

[Gibbo]

Anyway, back to Navi...

Personally I don't see AdoredTV Jim as some sort of guru, but a lot of his information makes logical sense and unless his sources are literally Lisa Su, Mark Papermaster and Uncle Jensen, he's going to get volatile, unconfirmed and outright duff information from time to time. At the very least his sources aren't going to be privy to every nuance and change, so they could feed Jim some information which really was the plan at the time, but not be aware of a different direction or technical issue which changes that plan, which then makes the leak look bogus.

Is that's what happened with Navi? Fudzilla said mid-October than their sources had Navi already testing in the lab and performing "better than expected" and ahead of schedule for the H2/Q3 launch. Good sign, but if that was the case why no CES demo? Ryzen is still 6 months away from release but we had a demo for that. AdoredTV then say there was a previously-unidentified issue with Navi which required a retape, but his sources said that was end of September. No way of corroborating one way or another, but seems reasonable to me given that it was a statement made in isolation.

But now AdoredTV is insinuating something's gone apocalyptic? Those phrases he quotes are very much past tense, so there's nothing to suggest Navi continues to be delayed by whatever bug apparently showed up. But unlike that bug suggestion, this latest tweet is made against the backdrop of significant price cuts to the Vega 56 at OcUK which actually don't make any sense.

Gibbo says that AMD requested the price drop, and cited a response to the 1660 Ti as the reason. But why? Better-performing cards cost more, plus these new prices really question getting a RX 590. But Gibbo's also cited RX Vega is EOL, and many times justified price gouging high-demand EOL stock as "maintaining sales expectations". So then, if he's selling 600+ Vega 56 units at the new crazy price but it's an EOL product, at what point is his stock going to run out and then bumps the price back up? Is Vega 56 actually EOL, or are AMD clearing stock as fast as possible (Lisa Su has made suggestions about having stock "left in the channel" when talking in vague terms about Navi)?

This says to me the following:

Vega 56 production has been ramped back up to cover the 1660Ti because Navi is broken and going to be late again.

Or

Vega 56 stock is being shifted fast because the RX 3070 is coming soon.

There is a lot of Vega 56 blower aged stock lurking around, think MSI have dumped most of theirs now as the £249 deal witnessed us sell hundreds in a matter of days, the price is now back up until I have concrete information on getting future shipments confirmed, right now its an unknown if they can pull stock from another region, as such those are now £269.99 until we can get another big volume of stock.

Sapphire is now £289.99, simply because £279 is cutting the margin very thin and were a business here to make a profit, we can't always operate like a charity.

I shall discuss with Sapphire on the possibility of more stock, right now its a bit vague but I've got a feeling they will have to manufacturer more Vega 56 otherwise come end of Spring early Summer AMD will have nothing to sell apart from RX 590 and below and as our Vega 56 sales have really exploded, the thousands we have are now down to just over 1000 units, which at current run rates is only around a months worth of stock, also hence the price increase to slow the sales down, though were still the cheapest in UK by typically £50 or so as such sales remain strong and we make better margin doing so.

 

[LePhuronn]

Well I think you answered your own question

Absolutely it’s the best they can do. That doesn’t make it ‘the best’ though. Upon release of Vega 1 I heard too much ‘wait for Navi’ and now it’s wait for Arcturus. People can only wait for so long given what they want already exists.

If anybody has been saying "wait for Navi" then their disappointment will be of their own making. It's always been wait for Arcturus to see AMD return to form, if they can. Navi has never, ever been discussed as more than a mid-range product, offering only a smidgen more performance than what we have already but at a more sensible cost - those AdoredTV leaks fuelled existing discussion, not spark a new one.

 

[Zeed]

@Gibbo this Vega 56 deal is so good managed to get My mate that not had other card than NV since ATI times to buy one while waiting for 7nm NV/AMD GPU's. Best deal on gpu out there by farrr.
I dont think NAVI will be better deal TBH not at this price point and how 7nm will cost more to produce. We will see sometime later this year.

 

[ANDREW GREEN]

Anyway, back to Navi...

Personally I don't see AdoredTV Jim as some sort of guru, but a lot of his information makes logical sense and unless his sources are literally Lisa Su, Mark Papermaster and Uncle Jensen, he's going to get volatile, unconfirmed and outright duff information from time to time. At the very least his sources aren't going to be privy to every nuance and change, so they could feed Jim some information which really was the plan at the time, but not be aware of a different direction or technical issue which changes that plan, which then makes the leak look bogus.

Is that's what happened with Navi? Fudzilla said mid-October than their sources had Navi already testing in the lab and performing "better than expected" and ahead of schedule for the H2/Q3 launch. Good sign, but if that was the case why no CES demo? Ryzen is still 6 months away from release but we had a demo for that. AdoredTV then say there was a previously-unidentified issue with Navi which required a retape, but his sources said that was end of September. No way of corroborating one way or another, but seems reasonable to me given that it was a statement made in isolation.

But now AdoredTV is insinuating something's gone apocalyptic? Those phrases he quotes are very much past tense, so there's nothing to suggest Navi continues to be delayed by whatever bug apparently showed up. But unlike that bug suggestion, this latest tweet is made against the backdrop of significant price cuts to the Vega 56 at OcUK which actually don't make any sense.

Gibbo says that AMD requested the price drop, and cited a response to the 1660 Ti as the reason. But why? Better-performing cards cost more, plus these new prices really question getting a RX 590. But Gibbo's also cited RX Vega is EOL, and many times justified price gouging high-demand EOL stock as "maintaining sales expectations". So then, if he's selling 600+ Vega 56 units at the new crazy price but it's an EOL product, at what point is his stock going to run out and then bumps the price back up? Is Vega 56 actually EOL, or are AMD clearing stock as fast as possible (Lisa Su has made suggestions about having stock "left in the channel" when talking in vague terms about Navi)?

This says to me the following:

Vega 56 production has been ramped back up to cover the 1660Ti because Navi is broken and going to be late again.

Or

Vega 56 stock is being shifted fast because the RX 3070 is coming soon.

Given they're making a loss on the product it's the latter. They have to shift X number of Vega Cards and X number of HBM2 memory they have in stock before Navi hits. But what that means is unknown. A= Days B=Units sold per day

So A x B = X. Once X is met Navi is released, so X/B= A or Navi Release date. Vega was already discounted and selling hand over fist but when 1660Ti launched, AMD had to drop the price to keep shifting the units.

The only issue is we don't know how much stock there is left so while it could be 5 months away in theory they could keep doing this for 9 months.

So according to Gibbo they only have enough Vega stock to reach early summer, in that case X=Earl Summer. But that may not be the case if they still have some stock of Vega components left that they need to sell.

And if they are managing their stock levels properly to cater for the £250-£400 market this has to be case, we just don't know how much stock they need to shift.

 

[ANDREW GREEN]

If anybody has been saying "wait for Navi" then their disappointment will be of their own making. It's always been wait for Arcturus to see AMD return to form, if they can. Navi has never, ever been discussed as more than a mid-range product, offering only a smidgen more performance than what we have already but at a more sensible cost - those AdoredTV leaks fuelled existing discussion, not spark a new one.

It depends what you mean by mid-range but like for like cards should be offering 35-50% jumps while being more expensive than their end of life equivalents. So the 580 replacements needs to be at least at 1070ti levels to compete with the 2060. The 570 replacement needs to be at 1070 levels to compete with the 1660TI. And really it should be beating all comfortably at 7nm

It's been revealed that AMD have a patent for variable rate shading and Nvidia claim this technology allows upto 50% increased performance, so if this is in Navi we should get a big improvement alone.

https://www.pcgamesn.com/amd/amd-variable-rate-shading-patent-navi-next-gen-consoles

Then there's 7nm process, the first new architecture in however many years and faster gddr6 vram.

 

[LePhuronn]

We have zero actual information regarding how Navi breaks down, but based on the AdoredTV leaked specs there is no real "like-for-like" in that Navi seems to actually reset the mid-range.

$250 RX 3080 = Vega 64 + 15% = GTX 1080/RTX 2070
$200 RX 3070 = Vega 56 = GTX 1070/RTX 2060
$130 RX 3060 = RX 580 = GTX 1060

Take that for what it is.

but like for like cards should be offering 35-50% jumps while being more expensive than their end of life equivalents.

But no, why should they be more expensive? That leads to ridiculous pricing as we're seeing now. Low, mid and high-end tiers are defined by price points, not performance, or at least they should be. So perhaps yes we should be getting a 35% jump in performance over the previous generation, but at the same price.

 

[ANDREW GREEN]

We have zero actual information regarding how Navi breaks down, but based on the AdoredTV leaked specs there is no real "like-for-like" in that Navi seems to actually reset the mid-range.

$250 RX 3080 = Vega 64 + 15% = GTX 1080/RTX 2070
$200 RX 3070 = Vega 56 = GTX 1070/RTX 2060
$130 RX 3060 = RX 580 = GTX 1060

Take that for what it is.

But no, why should they be more expensive? That leads to ridiculous pricing as we're seeing now. Low, mid and high-end tiers are defined by price points, not performance, or at least they should be. So perhaps yes we should be getting a 35% jump in performance over the previous generation, but at the same price.

Because End Of Life products are cheaper and will have to be sold alongside the new cards to an extent so you £150 580s and £130 570s on the market right now. If you could get a 3080 with the AdoredTV spec for £300, would you pay it? I would. Or a 3070 matching a Vega 56 for £250?

Those are the price points I'd anticipate and it depends how fast or slow Navi is. Let's say a 3080 matches a 2070, they could charge £400 and it'd be best of class. Or if a 3070 matched a 2060 why not price it at £300? But if 3080 only matches the 2060 and the 3070 only matches the 1060ti, they have to beat them in price because of Nvidia's superior brand image.

 

[LePhuronn]

Because End Of Life products are cheaper and will have to be sold alongside the new cards to an extent

Is that why the GTX 1080 costs £300+ more than the RTX 2080?

Now, I'm not trying to portray AMD as some kind of charity and saviour of all, but if AMD sell a 2070-matching RX 3080 for £400 then they're idiotic and just perpetuating all the price gouging greed going on in the GPU sector. They're not price-matching Intel for comparably-performing Ryzen 2000s, I don't see them exceeding Intel pricing when Ryzen 3000 crushes all that opposes it (gotta love hyperbole), so why would they match Nvidia prices with Navi? We've seen Apple and Nvidia take big hits in sales because they've gone a step too far in milking their customers, and you can make just as much profit from high-volume, low margin sales as you can low-volume, high margin sales (don't mention Radeon VII, that's a unique and messy situation).

AMD have a massive uphill battle in terms of mind share, not just market share, and if they can smash out something to match RTX 2070, undercut it by $100 and still make a healthy profit, then it will fly off the shelves, boosting opinion and market share in one hit. It's a long-term strategic game, and doing a short-term thing of "well, we've matched the 2070 so let's charge the same for it" just will not work.

Those are the price points I'd anticipate and it depends how fast or slow Navi is. Let's say a 3080 matches a 2070, they could charge £400 and it'd be best of class. Or if a 3070 matched a 2060 why not price it at £300? But if 3080 only matches the 2060 and the 3070 only matches the 1060ti, they have to beat them in price because of Nvidia's superior brand image.

Yes, and that applies across the board regardless of how Navi performs. Like I said above, they could charge £400 for a RTX 2070 competitor, but is that a sufficient undercut to swing sales? A RTX 2060 competitor for £300? That's only £40 cheaper; is it enough? It's not about just profit, it's about gaining market share.

So let's say an RX 3080 does indeed match the RTX 2070 and it costs AMD £150 to make. They could charge £400 and have a 267% profit margin for every sale. They could charge £300 for it and have a 200% profit margin (and I agree with you in that I'd snap one up in a heartbeat). Now yes, they'd make more money per sale on the former, but if they have double the total sales with the latter then that's simply more profit, and increase in market share and an increase in consumer confidence in the brand (i.e. fair pricing and not total rip-off merchants).

All the good stuff for shareholders. And seeing how Nvidia shareholders are selling in droves because of poor RTX sales (for one), I'm inclined to believe AMD won't make the same mistake themselves.

 

[nashathedog]

We all have had this assumption that the Radeon V11's are failed MI50 instinct cards, but Ill just point out that the MI50 instinct cards are listed with lower clocks on the core, the same clocks on the memory and consequently slightly lower peak throughputs.

So Ill just ask what part of the Radeon V11 is a failed chip?

Personally I have no idea, as far as I can tell, if it works then it could be used for either.

Anyway this is off topic for this thread, so apologies.

As I've posted previously I'd lean towards it being a case of Mi50 sales slowing and the VII being a way to sell more.