Vega/Navi Rumors (Updated)

[Glo.]

Then why not put it into the professional lineup and give it a professional name. You're grasping at straws but keep going.

If a certain competitor released a prosumer gpu and then topped that performance right away with a cheaper faster gpu people would be up in arms. Who knows, maybe one day ams will strip the pro parts from the future frontier edition.

This is only your problem, that you are refusing to look at things as they are. You are looking at it the way you want it to be.

Frontier Edition is professional GPU. AMD placed it in Professional lineup, even on their Raden.com site. Look in the mirror first, for biases, then turn to others.

 

[raghu78]

Then why not put it into the professional lineup and give it a professional name. You're grasping at straws but keep going.

If a certain competitor released a prosumer gpu and then topped that performance right away with a cheaper faster gpu people would be up in arms. Who knows, maybe one day ams will strip the pro parts from the future frontier edition.

Radeon Vega Frontier Edition is a Pro card.

http://pro.radeon.com/en-us/frontier/

AMD has 4 versions of Vega

1. Radeon Instinct (MI25) - For HPC, Deep learning and servers
2. Radeon Vega Frontier Edition - Professionals who require certified drivers for their professional applications
3. Radeon Vega Pro SSG - Specifically for 4k and 8K video editing , real time raytracing and very high fidelity rendering which have terabytes of data.
4. Radeon RX VEGA - Gaming.

btw Rageon Vega FE has 16GB HBM2 using two 8 Hi stacks while RX Vega has 8GB HBM2 using two 4 Hi stacks. Radeon Vega FE will come with certified Pro drivers and is designed for professional market. Radeon Vega FE is AMD's equivalent of Nvidia Quadro.

 

[CatMerc]

Dude, I'm not expecting amd to copy cat Nvidia tactics perfectly. You have to have some creativity I think it'd be breaking something to do EXACTLY what Nvidia did. It's similar lameness if you can't see that that's great. Myself and others do. I'll be actually interested to see who picks up this gpu and isn't just following it for fun.
Just like how everyone defended fury x but no one bought it new I doubt many of you defending Vega will buy top end Vega day 1.we have few Fiji owners who bought new gpus at release and we'll have few Vega owners too. Amd at the high end is lackluster in many manners and it's only getting worse. You're welcome to close your eyes to it. I'm not. Amd has lost my trust moving past Vega.

By that logic Quadro's are a disgusting way to milk people.
Oh wait, they come with drivers and support that aren't available in the gaming editions.

What do ya know.

 

[ElFenix]

This is getting a little too heated in here, everybody step back just a bit.

AT Moderator ElFenix

 

[StefanR5R]

Essentially, if we are to literally take "over the next 2 months" based on the date she made that comment, then yeah: End of July, there will be 4 Vega cards out there. FE is still 1 month away, so that leaves another month to release 3 different cards.

Kinda nuts, really. It almost sounds like a carpet-bombing campaign;

I think it's better to wait for further announcements (Computex presumably) before deciding on what "very soon thereafter" and "over the next couple of months" are supposed to mean in terms of calendar days.

Also, it should be at least 5 models: The two semi-pro cards (Frontier Edition with blower cooler and water cooler), and then "the enthusiast gaming platform, the machine learning platform and the professional graphics platform", whatever platform means in numbers of SKUs. But then, the differences between these models may be small (hardware/ firmware/ driver differences).

 

[Veradun]

I think it's better to wait for further announcements (Computex presumably) before deciding on what "very soon thereafter" and "over the next couple of months" are supposed to mean in terms of calendar days.

Also, it should be at least 5 models: The two semi-pro cards (Frontier Edition with blower cooler and water cooler), and then "the enthusiast gaming platform, the machine learning platform and the professional graphics platform", whatever platform means in numbers of SKUs. But then, the differences between these models may be small (hardware/ firmware/ driver differences).

The only big differences might be 8GB top consumer cards and cards with tiered memory (hbm2+ssd) for content creation.

 

[Mopetar]

I think the consumer cards will get the cut does as well. Vega is a pretty big die so I wouldn't be surprised to see consumer cards with 48/56 CU whereas the other categories just get the full dies or 2 of the 3 categories.

I think a 48 CU (~3k SP) Vega with 4 GB HBM2 could make for a sweet GPU if it used the Nano form factor and came in at around 150W TDP.

Otherwise the full does for RX Vega are the high-leakage parts that can clock a little higher and the professional parts are the low leakage dies with better power characteristics.

 

[CatMerc]

They'd need to be significantly faster clockspeed wise if Raja's claims of some gaming cards being faster than FE is true.

I don't think 14nm LPP yields at this point would warrant not having a full die at the top end of the gaming stack. Cut down variants will obviously exist, but at the top a full die will probably exist.

 

[antihelten]

I think the consumer cards will get the cut does as well. Vega is a pretty big die so I wouldn't be surprised to see consumer cards with 48/56 CU whereas the other categories just get the full dies or 2 of the 3 categories.

I think a 48 CU (~3k SP) Vega with 4 GB HBM2 could make for a sweet GPU if it used the Nano form factor and came in at around 150W TDP.

Otherwise the full does for RX Vega are the high-leakage parts that can clock a little higher and the professional parts are the low leakage dies with better power characteristics.

Since there is such a big gap in performance between Polaris 10 and Vega 10, I could definitely see AMD doing as much as 4 different SKUs off of Vega 10.

A 64 CU 1600 MHz version at $600 with water cooling. Performance roughly equal to 1080 Ti.
A 64 CU 1500 MHz version at $500 with air cooling (possibly without an official reference design cooler). Performance roughly 20% higher than 1080.
A 52 CU 1500 MHz version at $400. Performance roughly equal to 1080.
A 44 CU 1500 MHz version at $300. Performance roughly equal to 1070 (and thus 30-40% faster than RX 480).

This assumes that Vega 10 will have PPF (thanks Crisium) similar to Polaris. If instead it is similar to Hawaii (and Vega can thus fix the ROP/Bandwidth bottleneck of Polaris), then you could add 10-15% to the above performance numbers, and potentially ~$100 to the prices.

On problem with such a lineup is of course that it might not leave enough room for a smaller Vega 11 in the lineup.

I don't think any of them would have less than 8GB of HBM2 though. 4GB HBM2 would meant either two 2Hi stacks (I'm not sure SK Hynix even offers 2Hi stacks) or more likely a single 4Hi stack. In the case of the latter option bandwidth would be cut in half, which isn't really a viable option IMHO.

They'd need to be significantly faster clockspeed wise if Raja's claims of some gaming cards being faster than FE is true.

Or just drivers better tuned for gaming, which is essentially what Raja said would be the case:

RX will be fully optimized gaming drivers, as well as a few other goodies that I can't tell you about just yet....But you will like FE too, if you can't wait

 

[tential]

Since there is such a big gap in performance between Polaris 10 and Vega 10, I could definitely see AMD doing as much as 4 different SKUs off of Vega 10.

A 64 CU 1600 MHz version at $600 with water cooling. Performance roughly equal to 1080 Ti.
A 64 CU 1500 MHz version at $500 with air cooling (possibly without an official reference design cooler). Performance roughly 20% higher than 1080.
A 52 CU 1500 MHz version at $400. Performance roughly equal to 1080.
A 44 CU 1500 MHz version at $300. Performance roughly equal to 1070 (and thus 30-40% faster than RX 480).

This assumes that Vega 10 will have PPF (thanks Crisium) similar to Polaris. If instead it is similar to Hawaii (and Vega can thus fix the ROP/Bandwidth bottleneck of Polaris), then you could add 10-15% to the above performance numbers, and potentially ~$100 to the prices.

On problem with such a lineup is of course that it might not leave enough room for a smaller Vega 11 in the lineup.

I don't think any of them would have less than 8GB of HBM2 though. 4GB HBM2 would meant either two 2Hi stacks (I'm not sure SK Hynix even offers 2Hi stacks) or more likely a single 4Hi stack. In the case of the latter option bandwidth would be cut in half, which isn't really a viable option IMHO.



Or just drivers better tuned for gaming, which is essentially what Raja said would be the case:

If amd actually did what you suggested and had the wc cooled top end chip for $600 (significant discount from 1080ti) I'll break time to have it first. But I doubt it. I know I'll preorder this and be highly disappointed with the price /perf. I'm expecting $700 or even a full match to the 1080ti at $750.

 

[Head1985]

I think we will se same setup like with FIJI.
Full card 4096SP 20%faster than GTX1080
Cutdown 3584SP 5%faster/same as GTX1080
Nano at 1200Mhz GTX1070 performance.
Gap is not that big.RX580 with aftermarket cards is only 20-25% slower than GTX1070.

 

[zinfamous]

I just want a nano-type card (SFF) that can blast through the next ~4 years of 1440/100+ content (high to ultra) for </= $500.

Is that an impossible ask?

 

[tential]

I think we will se same setup like with FIJI.
Full card 4096SP 20%faster than GTX1080
Cutdown 3584SP 5%faster/same as GTX1080
Nano at 1200Mhz GTX1070 performance.
Gap is not that big.RX580 with aftermarket cards is only 20-25% slower than GTX1070.

Why would the nano card be slower this time around? Nano card was the same chip as the full card. The 1080ti just released a nano version.

There should be a high end and cut down nano chip.

 

[Qwertilot]

They didn't cut the nano but it definitely didn't run at the same clocks as the furyX either!

 

[Snarf Snarf]

Why would the nano card be slower this time around? Nano card was the same chip as the full card. The 1080ti just released a nano version.

There should be a high end and cut down nano chip.

That would entirely depend on the voltage/clock curve of the new iteration of GCN uarch. We all know that AMD always clocks things out of their ideal efficiency for max performance out of the box, if the top RX Vega hits around 1600 MHz boost I would fully expect the "nano" lower TDP card to be around 1350-1400 sustained clocks with maybe a bit higher burst in boost mode. The 1080ti mini card doesn't have to deal with all of the heat concentration being almost entirely on die like the RX Vega will with HBM, there is likely going to be a huge benefit to sustained clocks on the watercooled edition.

Edit: Forgot to mention that the mini 1080ti's aren't going up to 2GHz like the aftermarket cards are capable of doing. Do keep in mind even at that clock some of the best aftermarket coolers get into the 90 C realm.

 

[w3rd]

Vega is going to slide into the $399 and up.

AMD has nothing to fear from Pascal, in which it will most likely out muscle it. The GP102 does not match full Vega sku. Nvidia will have to wait for Volta for a try at that. Vega x2 is real...




lastly, HBM2 memory takes up almost no space, nearly all Vega boards will be Nanos.

 

[Mopetar]

They'd need to be significantly faster clockspeed wise if Raja's claims of some gaming cards being faster than FE is true.

Or just drivers optimized for gaming. The professional cards from both companies have historically been the same hardware, just different drivers. The Tesla cards from NVidia are a bit of a break from this with specialized hardware for various HPC workloads, but their high cost is what can justify the development of a separate chip.

I don't think 14nm LPP yields at this point would warrant not having a full die at the top end of the gaming stack. Cut down variants will obviously exist, but at the top a full die will probably exist.

I didn't mean to imply there wouldn't be a full chip for consumer Vega, merely that it appears that most or all cut chips are going towards consumer products as the FE card seemed to be a full die. There might be a cut Vega die offered as a lower tier HPC part, but I think most are being used for consumer GPUs.

lastly, HBM2 memory takes up almost no space, nearly all Vega boards will be Nanos.

If this were even remotely true, then the FE card would have been on a smaller board. Even if you don't need the actual board room, a bigger card allows for more fans and better cooling.

 

[Valantar]

Well, if the Zotac 1080Ti mini is the same size as the 1080 Mini (it's definitely not smaller, that's for sure!), it's not really comparable to the R9 Nano. The Nano was 15.8x11.1 cm - a true ITX GPU. The 1080 Mini is 21.1cm x 12.5cm. That's substantially bigger.

Still, the 1080 Ti mini has to dissipate 250W, not the 175 of the Nano - and it doesn't even have a vapor chamber to help it along. As for the Nano being handicapped due to concentrated heat, I disagree simply due to how much less power HBM uses compared to GDDR5(X). If your RAM is using half the power, and is close enough to be cooled by the same vapor chamber as the GPU itself, I'd count that as an advantage, not a handicap.

I do hope AMD follows it up. If conventional wisdom is anything to go by, wide and slow is more efficient than narrow and fast, which makes me hope for a full die or small cut at best. Other than that, I pretty much agree with the suggested SKUs above - Water cooled, air cooled, air cooled cut die, Nano. That is where what I hope for and what I see as likely intersects, at least. Now can Compute please get here? Not that I'm replacing my Fury X any time soon, but I want to know more

 

[Mopetar]

I don't think the 1080Ti has 50% of power going towards the memory. Probably only 25% at absolute maximum.

HBM2 will mean AMD has lower TDP, or more headroom for an OC, but I think you're overestimating its importance here.

 

[tential]

Well, if the Zotac 1080Ti mini is the same size as the 1080 Mini (it's definitely not smaller, that's for sure!), it's not really comparable to the R9 Nano. The Nano was 15.8x11.1 cm - a true ITX GPU. The 1080 Mini is 21.1cm x 12.5cm. That's substantially bigger.

Still, the 1080 Ti mini has to dissipate 250W, not the 175 of the Nano - and it doesn't even have a vapor chamber to help it along. As for the Nano being handicapped due to concentrated heat, I disagree simply due to how much less power HBM uses compared to GDDR5(X). If your RAM is using half the power, and is close enough to be cooled by the same vapor chamber as the GPU itself, I'd count that as an advantage, not a handicap.

I do hope AMD follows it up. If conventional wisdom is anything to go by, wide and slow is more efficient than narrow and fast, which makes me hope for a full die or small cut at best. Other than that, I pretty much agree with the suggested SKUs above - Water cooled, air cooled, air cooled cut die, Nano. That is where what I hope for and what I see as likely intersects, at least. Now can Compute please get here? Not that I'm replacing my Fury X any time soon, but I want to know more

nano wasn't a failure product as much as the underlying chip could have been better. You'd be insane to abandon the mini chip and wc top sku now.

Do what you did with Fiji, just don't suck at it.

And add in custom skus to use with the wc top sku for those who choose not to wc (non reference so your reference wc chip gets to top the review charts).

It feels so simple, but and is going to massively scree up Vega launch and make me pissed I spent the $700-1400 but unhappy since I didn't have a cost efficient alternative other than Vega due to monitor lock in now.

 

[tential]

They didn't cut the nano but it definitely didn't run at the same clocks as the furyX either!

Barely slower....
You could get fury x speed out of it if you wanted it just didn't make sense since you weren't gaining a lot of perf for a lot extra wattage.

Nano was a great idea hampered by the underlying chip, pricing, and being way too ahead of its time to have enough infrastructure in place to take full use of all the benefits of the card
Amd is either too late or too early, they're never on time.

 

[maddie]

I see guru3d has a post headlined as:

AMD RX Vega HBM 2 8GB Memory Stack Reportedly Costs $160
Reports have been showing that a 4 GB HBM2 stack costs $80. AMD Vega GPUs make use of 8 GB HBM2, two stacks used is $160.


To think they're stating the just the HBM2 modules cost almost as much as the entire Fiji card was estimated to cost and it's being seriously considered as accurate in the forum.

 

[Mopetar]

Yeah, Nano was good, but it was pretty much at the wall for Fury already. Nano would be more fondly remembered if Fury had another 25% headroom.

I think it would be funny for AMD to have a Vega Nano that has a lower TDP than the 580 cards you can get. Some of those things are the Hummers of GPUs.

Mayhaps small Vega is a 32 CU Polaris replacement that comes out next year. I don't see Navi until late 2018 at soonest and Polaris doesn't seem to have much of a future.

 

[Cookie Monster]

I see guru3d has a post headlined as:

AMD RX Vega HBM 2 8GB Memory Stack Reportedly Costs $160
Reports have been showing that a 4 GB HBM2 stack costs $80. AMD Vega GPUs make use of 8 GB HBM2, two stacks used is $160.


To think they're stating the just the HBM2 modules cost almost as much as the entire Fiji card was estimated to cost and it's being seriously considered as accurate in the forum.

Not sure about the accuracy but since HBM2 is newer and has alot more customers (nVIDIA, AMD etc).. the cost might have been driven up substantially given demand/supply compared to HBM1. Plus these are going into high margin products so perhaps the manufacturers like SK Hynix etc are looking to recoup the cost in investment also.

So there could be some truth in this. Doesn't bode well for the consumer versions i.e. the gaming SKUs.

 

[lobz]

Not sure about the accuracy but since HBM2 is newer and has alot more customers (nVIDIA, AMD etc).. the cost might have been driven up substantially given demand/supply compared to HBM1. Plus these are going into high margin products so perhaps the manufacturers like SK Hynix etc are looking to recoup the cost in investment also.

So there could be some truth in this. Doesn't bode well for the consumer versions i.e. the gaming SKUs.

Also keep in mind, that AMD surely has agreed at least with Hynix way beforehand on a fixed price for the first many functioning batches.

Sent from my VTR-L09 using Tapatalk