AMD unveils new Radeon Pro Card

[cj09beira]

if this is indeed using 2.4gbps vram it would be cool if anyone tried to oc it, though samsungs aquabolt probab.y clocks better

[Taf the Ghost]

7 minutes ago, cj09beira said:

if this is indeed using 2.4gbps vram it would be cool if anyone tried to oc it, though samsungs aquabolt probab.y clocks better

Seen some discussion that suggest those modules probably get all of the benefits any memory OC can provide with the Architecture. 

[cj09beira]

2 minutes ago, Taf the Ghost said:

Seen some discussion that suggest those modules probably get all of the benefits any memory OC can provide with the Architecture. 

if they are running at 2.4 gbps then probably yes, but they might not as someone said before. though the overclocking part was mostly because i love to know how far things can be pushed 

[Taf the Ghost]

Just now, cj09beira said:

if they are running at 2.4 gbps then probably yes, but they might not as someone said before. though the overclocking part was mostly because i love to know how far things can be pushed 

I'd still love to know what the strange bottleneck seems to be within Vega, but I have a feeling some AMD Engineers were in the same place a few years ago after they taped it out.

[cj09beira]

1 minute ago, Taf the Ghost said:

I'd still love to know what the strange bottleneck seems to be within Vega, but I have a feeling some AMD Engineers were in the same place a few years ago after they taped it out.

maybe IF in not good enough and might have added some latency where it makes a big difference

[Taf the Ghost]

24 minutes ago, cj09beira said:

maybe IF in not good enough and might have added some latency where it makes a big difference

I've seen a few high level discussions of it, and it really comes down to limitations of the GCN structure. Though it gets a lot more complicated than that, obviously. 

[Okjoek]

Cool.

[Trixanity]

3 hours ago, schwellmo92 said:

The memory will at least hit 2.4Gbps with overclocking as that’s what it’s rated for, how far past that is lottery.

If what you're saying is true why would AMD not clock it at 2.4? Either they're crippling it deliberately or SK Hynix is. 

 

Granted, AMD has no intention of bumping clocks on older products so the disparity would be pretty big between Vega products but if I recall memory manufacturers do sell different SKUs with different speeds in order to have different price points just like anything else.

[GOTSpectrum]

2 minutes ago, Trixanity said:

If what you're saying is true why would AMD not clock it at 2.4? Either they're crippling it deliberately or SK Hynix is. 

It could be for any number of reasons, power, reliability, heat, some kind of limitation in vega we don't know about etc etc

 

[cj09beira]

5 minutes ago, Ben Quigley said:

It could be for any number of reasons, power, reliability, heat, some kind of limitation in vega we don't know about etc etc

 

even though most of a hbm memory controller is in the memory chips part of it is still in the silicon, so there is a good chance that vega 10 simply cant coupe with the the added bandwidth without memory errors

[schwellmo92]

10 minutes ago, Trixanity said:

If what you're saying is true why would AMD not clock it at 2.4? Either they're crippling it deliberately or SK Hynix is. 

 

Granted, AMD has no intention of bumping clocks on older products so the disparity would be pretty big between Vega products but if I recall memory manufacturers do sell different SKUs with different speeds in order to have different price points just like anything else.

They wont clock it that high because it'll eat too much in to their higher tier card.

 

https://www.anandtech.com/show/13210/amd-announces-radeon-pro-wx-8200

 

"AMD for their part isn’t drawing a whole lot of attention to the matter. But when poked about it, they’re mentioning that the WX 8200 is using SK Hynix’s “Gen 2” HBM2, which unveiled earlier this year, offers memory speeds up to 2.4Gbps."

[Trixanity]

17 minutes ago, schwellmo92 said:

They wont clock it that high because it'll eat too much in to their higher tier card.

 

https://www.anandtech.com/show/13210/amd-announces-radeon-pro-wx-8200

 

"AMD for their part isn’t drawing a whole lot of attention to the matter. But when poked about it, they’re mentioning that the WX 8200 is using SK Hynix’s “Gen 2” HBM2, which unveiled earlier this year, offers memory speeds up to 2.4Gbps."

"Up to" being the keyword. GDDR5X goes up to 14 Gbps but we all saw how that went.

 

I'm not saying you can't OC many or all of them to go way beyond 2 Gbps but I'm skeptical of the notion you'll be able to take any card and just OC it to 2.4 as it it was artificially limited across the board.

[schwellmo92]

1 hour ago, Trixanity said:

"Up to" being the keyword. GDDR5X goes up to 14 Gbps but we all saw how that went.

 

I'm not saying you can't OC many or all of them to go way beyond 2 Gbps but I'm skeptical of the notion you'll be able to take any card and just OC it to 2.4 as it it was artificially limited across the board.

The difference is the GDDR5X JEDEC specification is up to 14Gbps (actual 10-14Gbps), whereas this is the actual HBM2 product being designed for 2.4Gbps.

[Trixanity]

Just now, schwellmo92 said:

The difference is the GDDR5X JEDEC specification is up to 14Gbps (actual 10-14Gbps), whereas this is the actual HBM2 product being designed for 2.4Gbps.

And first gen HBM2 was designed for 2 Gbps; yet none of the graphics cards I'm aware of hit that target until now with second gen.

[ARikozuM]

"A number of software partners - from Autodesk to Adobe - ... "

 

Someone really needs to learn their alphabet.

 

[schwellmo92]

1 hour ago, Trixanity said:

And first gen HBM2 was designed for 2 Gbps; yet none of the graphics cards I'm aware of hit that target until now with second gen.

You’re misunderstanding the difference between the HBM2 products specification and what AMD/AIB’s implemented it as. All 2Gbps HBM2 can run at 2Gbps, if AMD/AIB’s run it out of the box at a lower frequency that’s completely up to them.

[Trixanity]

1 hour ago, schwellmo92 said:

You’re misunderstanding the difference between the HBM2 products specification and what AMD/AIB’s implemented it as. All 2Gbps HBM2 can run at 2Gbps, if AMD/AIB’s run it out of the box at a lower frequency that’s completely up to them.

You think AMD implemented HBM2 at 1.89 Gbps for funsies? Seems quite arbitrary. Not to mention it making their card look silly when it has lower memory bandwidth than Fiji.

 

No, it's more likely that the memory couldn't consistently hit 2 Gbps at the time and only later improved but by then it's too late to change unless you want a PR problem.

 

Of course if you have a source to explain the deliberate underclocking of HBM2 then I'm all ears but the Anandtech article on Vega says HBM2 didn't reach its intended speed. Figured I might as well put it out there preemptively.

[Vordreller]

Just slightly under the 9100 in performance but under half the price(The 9100 is at 2199 USD MSRP right now I believe).

 

Should be a nice price point.

[ZacoAttaco]

So this card would have applications in networking virtualization? Any other applications for the virtualization specifically of this card?

[Stefan Payne]

13 hours ago, timl132 said:

I think this could be a pretty good card, in the benchmarks of the vega 64 and such, it was already seen that these cards perform good at things like cad and rendering.

And you don't necessarily need CUDA cores if the app supports OpenCL too, an most do.

AMD has a CUDA Wrapper or works on it. So that isn't really a problem...

[orbitalbuzzsaw]

Neat, but get back to me when they have something that performs north of the 1080 Ti for a reasonable price

[schwellmo92]

7 hours ago, Trixanity said:

You think AMD implemented HBM2 at 1.89 Gbps for funsies? Seems quite arbitrary. Not to mention it making their card look silly when it has lower memory bandwidth than Fiji.

 

No, it's more likely that the memory couldn't consistently hit 2 Gbps at the time and only later improved but by then it's too late to change unless you want a PR problem.

 

Of course if you have a source to explain the deliberate underclocking of HBM2 then I'm all ears but the Anandtech article on Vega says HBM2 didn't reach its intended speed. Figured I might as well put it out there preemptively.

That HBM2 was 1.6Gbps product from Hynix, that’s why it was implemented at 1.89Gbps not 2Gbps..

[Trixanity]

6 hours ago, schwellmo92 said:

That HBM2 was 1.6Gbps product from Hynix, that’s why it was implemented at 1.89Gbps not 2Gbps..

So you're saying Anandtech is wrong? Got any evidence?

 

Edit: just to clarify

Do you have evidence that current graphics cards used the lower SKU and why would they do that beyond a yield or price reason (or both)? It's poor marketing so there had to be a pretty good reason to lose memory bandwidth.

 

What evidence do you have that there aren't multiple SKUs in this new generation of memory (I know Samsung has only published one but this isn't Samsung) and why would they presumably pay extra money when it would make sense for 2 Gbps to be cheaper unless the yields are somehow better on 2.4 Gbps SKUs and are therefore cheaper? 

 

Something doesn't add up. That's all I'm saying.

[schwellmo92]

44 minutes ago, Trixanity said:

So you're saying Anandtech is wrong? Got any evidence?

 

Edit: just to clarify

Do you have evidence that current graphics cards used the lower SKU and why would they do that beyond a yield or price reason (or both)? It's poor marketing so there had to be a pretty good reason to lose memory bandwidth.

 

What evidence do you have that there aren't multiple SKUs in this new generation of memory (I know Samsung has only published one but this isn't Samsung) and why would they presumably pay extra money when it would make sense for 2 Gbps to be cheaper unless the yields are somehow better on 2.4 Gbps SKUs and are therefore cheaper? 

 

Something doesn't add up. That's all I'm saying.

The RX Vega 56 and RX Vega 64 used SK Hynix 1.6Gbps 1.2v HBM2, this new WX 8200 uses SK Hynix 2.4Gbps 1.2v HBM2. The SK Hynix 2Gbps product that they had slated for RX Vega wasn't ready in time and instead all SK Hynix had to offer was 1.6Gbps (which is what Vega 56 runs at) and for Vega 64 they pushed more voltage in to the HBM2 to get the faster 1.89Gbps speeds (that's why people flash the 64 bios - it allows for higher HBM2 voltage). As to why the WX 8200 uses the 2.4Gbps chips, I have no idea, but I can tell you they aren't clocked at 2.4Gbps out of the box because that would cause some pressure on their more expensive WX 9100 SKU which is already only marginally faster than the WX 8200 SKU.

 

This is all common knowledge you can find with a quick Google, so instead of arguing with do your own research and you'll find everything I am telling you is correct.

[Trixanity]

58 minutes ago, schwellmo92 said:

The RX Vega 56 and RX Vega 64 used SK Hynix 1.6Gbps 1.2v HBM2, this new WX 8200 uses SK Hynix 2.4Gbps 1.2v HBM2. The SK Hynix 2Gbps product that they had slated for RX Vega wasn't ready in time and instead all SK Hynix had to offer was 1.6Gbps (which is what Vega 56 runs at) and for Vega 64 they pushed more voltage in to the HBM2 to get the faster 1.89Gbps speeds (that's why people flash the 64 bios - it allows for higher HBM2 voltage). As to why the WX 8200 uses the 2.4Gbps chips, I have no idea, but I can tell you they aren't clocked at 2.4Gbps out of the box because that would cause some pressure on their more expensive WX 9100 SKU which is already only marginally faster than the WX 8200 SKU.

 

This is all common knowledge you can find with a quick Google, so instead of arguing with do your own research and you'll find everything I am telling you is correct.

'Do your own research' is probably the worst argument possible. It's used by flat earthers and anti vaxxers alike. Doesn't really help your case. I simply asked for documentation of some kind.

 

Some of your arguments make sense but some seem to be assumptions with no reasoning behind it except that it just is. That's why I'm not simply saying 'okay' and moving on.