Poll: ** The AMD VEGA Thread **

[AllBodies]

Next gen memory can also mean anything after HBM2, such as LCHBM or HBM3.

True, but not for the 2018 product, or even 1H2019, since the only 'new' memory out in that timeframe is GDDR6.

 

[nashathedog]

So will there be proper benchmarks in 8 days then?

There will be AMD benchmarks by then but when you look at how different their Fury X v 980ti benchmarks were when compared to reviewers benchmarks 2 day's later I'm not sure how reliable they'll be.

 

[nashathedog]

Interesting post on Reddit -

"Reference models are already stockpiling at one company with which I have contacts. They apparently have "low thousands" ready for release. They usually have tens of thousands, though, but this is apparently a very expensive GPU. Supposedly priced above the 1080 at the top (my contact doesn't have pricing info outside of internal rumblings).

The company is a primary supplier, not a reseller."

Could be BS though.

Based on the price part I hope so.

There is nothing in the API's that allows control over the rasterisation of the scene. Even nvidia's TBR is all driver controlled, even down to a per app basis if required.

So you're saying it doesn't require any changes or additions to be made on the game developers side? I thought it did which is why I asked for a link correcting me.

EDIT: I'd written it wrong. Doesn't not does.

 

[arby1]

Based on the price part I hope so.

I know it's been said before, but as they were apparently comparing it vs the 1080, and instead of saying "it's cheaper than the nvidia alternative", they came up with a convoluted "system price including gsync/freesync monitors", I think we unfortunately know it's going to be near the top end of the price scale.

 

[nashathedog]

Im quite excited to see what Navi will be like. 7nm and its the architecture which Raja had most involvement in. Not to mention is using "Next gen memory". Also the potential to make a big GPU out of higher salvageable smaller dies could be quite interesting.

The Next Gen Memory is just GDDR6 FYI...

And likely 14000 MHz on 256-bit for the first Navi cards out in late-2018, which will likely be RX480 sized. So we're talking a ~13 Tflop GPU with 448 GB/s bandwidth.

What is Navi a replacement for?
The whole range or a split up like we're seeing with The Polaris + (500 series) and Vega cards?
If it's a split up are you presuming Navi will be a 400 or 500 series replacement?
If so what is coming to replace Vega?
I'm wondering because you mention the next gen memory being GDDR6 and,
If Navi is a whole range what is the plan for HBM?
I presume they won't develop the same range for both types of memory.

 

[r7slayer]

What is Navi a replacement for?
The whole range or a split up like we're seeing with The Polaris + (500 series) and Vega cards?
If it's a split up are you presuming Navi will be a 400 or 500 series replacement?
If so what is coming to replace Vega?
I'm wondering because you mention the next gen memory being GDDR6 and,
If Navi is a whole range what is the plan for HBM?
I presume they won't develop the same range for both types of memory.

Navi is the next gen architecture. It should replace both Vega and Polaris. AFAIK from the information AMD have given.

 

[AllBodies]

What is Navi a replacement for?
The whole range or a split up like we're seeing with The Polaris + (500 series) and Vega cards?
If it's a split up are you presuming Navi will be a 400 or 500 series replacement?
If so what is coming to replace Vega?
I'm wondering because you mention the next gen memory being GDDR6 and,
If Navi is a whole range what is the plan for HBM?
I presume they won't develop the same range for both types of memory.

I'm more referring to timescale, and then there's also some extra questions of if Navi is an MCM design.

But basically the part(s) which come out in 2018 will almost certainly be RX580 and RX560 replacements, because the 7nm process will be immature still. Therefore the dies will be small.

And these will use 256-bit and below interfaces, with 14000 MHz and below GDDR6.

Then if Navi is still monolithic they'll likely wait till HBM3 is ready (late 2019) to make anything bigger, i.e. Vega's replacement.

Luckily this time even the slowest HBM3 will still be 358GB/s per stack, so even super early HBM3 will be compelling.

If Navi is MCM it will be more interesting. Perhaps the RX580 replacement would be one ~200-250mm2 die, and the RX560 replacement would be still 1 die, but disabled salvaged ones (like Ryzen 3, 5, 7).

Then they could bring out a 2 die version, using 384-bit 14000-16000 MHz GDDR6 in early 2019 (that would give 672-768 GB/s). This would be pretty damn fast already, as you'd expect 400-500mm2 on 7nm to be in 2x1080 Ti territory.

And then bring out a 4 die version with HBM3 later on, likely with at least 400 GB/s per stack, and possibly 1 stack per die. As 4x250mm2 dies on 7nm would be hilariously powerful, bare minimum 3x max-clocked TitanXp speed.

Any MCM design would have to put a BIG emphasis on efficiency though. As realistically they wouldn't want to go over 100W per die. Even then, with 4 dies and ~50W for memory, your top chip would be 450W.

 

[nashathedog]

I know it's been said before, but as they were apparently comparing it vs the 1080, and instead of saying "it's cheaper than the nvidia alternative", they came up with a convoluted "system price including gsync/freesync monitors", I think we unfortunately know it's going to be near the top end of the price scale.

As you said having a high price had alway's been the likely scenario and mixing it up into a gpu + monitor bundle in order to say it's cheaper is diverting away from the fact that the price will be higher than most people are expecting.
AMD have sat watching gpu prices be artificially raised by Nvidia due to the lack of competition and they'll want a piece of the pie. When Polaris and Pascal were out AMD were acting like the 1080 price was disgraceful and showing how two 480's outperformed a 1080 for less money (which was just as disgraceful in my opinion considering how bad crossfire's become) but now they have such a card they want to do exactly the same thing and the saddest bit is it'll make those artificially raised prices become the norm.

 

[N19h7m4r3]

What is Navi a replacement for?

Navi can replace what ever it wants. You can't judge Navi! Navi will judge you!

 

[nashathedog]

Navi is the next gen architecture. It should replace both Vega and Polaris. AFAIK from the information AMD have given.

Like the first iteration of GCN which gave us the whole 7000 series range, In that case it may be as big a game changer as GCN was.

I'm more referring to timescale, and then there's also some extra questions of if Navi is an MCM design.

But basically the part(s) which come out in 2018 will almost certainly be RX580 and RX560 replacements, because the 7nm process will be immature still. Therefore the dies will be small.

And these will use 256-bit and below interfaces, with 14000 MHz and below GDDR6.

Then if Navi is still monolithic they'll likely wait till HBM3 is ready (late 2019) to make anything bigger, i.e. Vega's replacement.

Luckily this time even the slowest HBM3 will still be 358GB/s per stack, so even super early HBM3 will be compelling.

If Navi is MCM it will be more interesting. Perhaps the RX580 replacement would be one ~200-250mm2 die, and the RX560 replacement would be still 1 die, but disabled salvaged ones (like Ryzen 3, 5, 7).

Then they could bring out a 2 die version, using 384-bit 14000-16000 MHz GDDR6 in early 2019 (that would give 672-768 GB/s). This would be pretty damn fast already, as you'd expect 400-500mm2 on 7nm to be in 2x1080 Ti territory.

And then bring out a 4 die version with HBM3 later on, likely with at least 400 GB/s per stack, and possibly 1 stack per die. As 4x250mm2 dies on 7nm would be hilariously powerful, bare minimum 3x max-clocked TitanXp speed.

Any MCM design would have to put a BIG emphasis on efficiency though. As realistically they wouldn't want to go over 100W per die. Even then, with 4 dies and ~50W for memory, your top chip would be 450W.

I see what you mean, I alway's presumed that making a single architecture with 2 different types of memory was a no go, To be honest I'd be more than happy to see HBM get dropped and see GDDR6 take it's place with AMD as so far it seems to have done nothing well.

Navi can replace what ever it wants. You can't judge Navi! Navi will judge you!

Now you're just getting scary.

 

[Steampunk]

I know it's been said before, but as they were apparently comparing it vs the 1080, and instead of saying "it's cheaper than the nvidia alternative", they came up with a convoluted "system price including gsync/freesync monitors", I think we unfortunately know it's going to be near the top end of the price scale.

Which is nonsense, because how many people have budgeted for a new monitor, or even need one? It would be like selling a small car at the same price as a Ferrari, and then claiming it's cheaper because you only need to buy a smaller garage to put it in.

 

[Steampunk]

As you said having a high price had alway's been the likely scenario and mixing it up into a gpu + monitor bundle in order to say it's cheaper is diverting away from the fact that the price will be higher than most people are expecting.
AMD have sat watching gpu prices be artificially raised by Nvidia due to the lack of competition and they'll want a piece of the pie. When Polaris and Pascal were out AMD were acting like the 1080 price was disgraceful and showing how two 480's outperformed a 1080 for less money (which was just as disgraceful in my opinion considering how bad crossfire's become) but now they have such a card they want to do exactly the same thing and the saddest bit is it'll make those artificially raised prices become the norm.

AMD have to be better and cheaper, because they are the underdog and a year late. Unless Vega is a spectacular performer, why would anyone buy it in favour of a cheaper 1080 that is from the incumbent and most supported company? And Nvidia will cut prices if Vega is anywhere near them in top performance.

 

[AlamoX]

Why sandbag though? What have they got to gain? Nvidia will already know the performance of the RX. They're not going to release a competing card but they can reduce prices very quickly.
AMD have had lots of bad publicity from the FE release and they've done almost nothing to counter it - a few twitter / reddit comments, the world tour

Why haven't they released snippets of information, benchmarks etc to generate positive interest. It just strikes me as incredibly naive marketing.
If they are sandbagging it's a joke. They've lost sales to Nvidia that they potentially could've had.

well, mostly because of these slides :
this one should solve the major issue Fiji had, fully utilize it's SPs

Spoiler

this one, should significantly save power draw and boost performance

Spoiler

this one, means the card wont be relying on brute force and large workloads, but should be also good at smaller workloads, this is a major change from radeon graphics since GCN introduction, also depanding on what glofo's node allow(power), the RX card could clock really high (2k or so )

Spoiler

and none of this seem to be enabled or optimized for vega FE, and with the potential to make quite a large difference.
ofc all this is hopefull thinking, because if Vega is barely matching a 1080 with almost double the TDP, and 50% bigger Die size, then AMD screwd up something very badly, and i don't believe AMD is that bad, because this is not lack of money, this is pure engineering, do you believe AMD's engineers are this bad ?

 

[MichaelM]

I know it's been said before, but as they were apparently comparing it vs the 1080, and instead of saying "it's cheaper than the nvidia alternative", they came up with a convoluted "system price including gsync/freesync monitors", I think we unfortunately know it's going to be near the top end of the price scale.

This really annoys me. I'm not looking for a cheaper monitor from AMD, i'm looking for a GPU to drive it, a GPU that is competitively priced for its performance. A GPU that should be out already considering how long AMD have been talking about it for.

As someone who does not consider Nvidia an option I've gone from "I need Vega" to "I'll see what it's like first".

 

[opethdisciple]

I'm also hanging on to see what Vega can do as i own a Ryzen. Hoping there is some sort of efficiency bonus if you have both in your system.

 

[MichaelM]

I'm also hanging on to see what Vega can do as i own a Ryzen. Hoping there is some sort of efficiency bonus if you have both in your system.

Hopefully not

 

[DarrenM343]

This really annoys me. I'm not looking for a cheaper monitor from AMD, i'm looking for a GPU to drive it, a GPU that is competitively priced for its performance. A GPU that should be out already considering how long AMD have been talking about it for.

As someone who does not consider Nvidia an option I've gone from "I need Vega" to "I'll see what it's like first".

Are you low end gaming only? If not then I don't understand that personally. If you want to game, at least out of the low end, you have to consider Nvidia.
I'm not overly fond of AMD past products due to past experience but if they were well ahead of the game with GPU's, I would pick up an AMD GPU in an instant. I wouldn't wait and wait for my preferred competitor to come out with something. Life's too short for that.
Fine if you wait it out now as we're nearly there but if AMD don't come up with the goods, just look elsewhere dude

 

[Devrij]

Are you low end gaming only? If not then I don't understand that personally. If you want to game, at least out of the low end, you have to consider Nvidia.
I'm not overly fond of AMD past products due to past experience but if they were well ahead of the game with GPU's, I would pick up an AMD GPU in an instant. I wouldn't wait and wait for my preferred competitor to come out with something. Life's too short for that.
Fine if you wait it out now as we're nearly there but if AMD don't come up with the goods, just look elsewhere dude

I think the point is that nvidia gpu's are expensive, which you are supporting if you buy one. Also, some people object to their business practices and would prefer not to buy nvidia.

I used to buy amd because they were the best bang for buck, but now that I'm older and have a decent job, the only way upwards is nvidia. So we need Vega to be good because if nvidia have a stranglehold on the top end they will continue to stay expensive.

So I understand the concept of holding out to support a business because you want more market competition.

 

[JediFragger]

I'm also hanging on to see what Vega can do as i own a Ryzen. Hoping there is some sort of efficiency bonus if you have both in your system.

FE performance better on Intel chips from what I've seen, so prob not. More BS.

 

[JediFragger]

I think the point is that nvidia gpu's are expensive, which you are supporting if you buy one.

I think 99% of us support whoever has the best product out at the time inc me. Problem is AMD haven't been competing for a long time, and we're still holding out hope that Vega will compete in some form to give us options!!