[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

[dacostafilipe]

Wasnt there 9 months between 8000 and 200 series? (january 2013 vs october 2013).

Yes, presentation was at CES.

 

[3DVagabond]

You are the one missing the point.

After talking about the 380 why on earth would they talk about a "future GPU" and not directly 390X if what's released in June has that name?

They did the exact same thing just before introducing the R9 2xx series with the HD 8000. Maybe they will introduce a new naming again?

I had already heard they might be changing their naming convention. Possibly doing away with numbers all together.

 

[Glo.]

If the rumor about R9 390X single card being almost as fast as R9 295X2 is true, then whole information from Cloudfire is getting more real, than before.

The performance and efficiency gains are phenomenal in this case, for every single card. Lets look at it:

If 370X is 2048 GCN core and performs around GTX780-R9 290, then bigger cards...

Now Im starting to be curious how well will be 3072 GCN core chip performing here.

 

[R0H1T]

If the rumor about R9 390X single card being almost as fast as R9 295X2 is true, then whole information from Cloudfire is getting more real, than before.

The performance and efficiency gains are phenomenal in this case, for every single card. Lets look at it:

If 370X is 2048 GCN core and performs around GTX780-R9 290, then bigger cards...

Now Im starting to be curious how well will be 3072 GCN core chip performing here.

Only & I repeat if & only if it's at 20nm, otherwise I don't see how a single GPU will be better than last gen's (actually the same gen, sort of) dual GPU solution. The only other possibility being that this is a super massive die, in which case no 395x2

 

[Vesku]

If it's just a top end HBM SKU or two they'll need to be first with the node shrink if they want to claw back market share.

 

[ShintaiDK]

If the rumor about R9 390X single card being almost as fast as R9 295X2 is true, then whole information from Cloudfire is getting more real, than before.

The performance and efficiency gains are phenomenal in this case, for every single card. Lets look at it:

If 370X is 2048 GCN core and performs around GTX780-R9 290, then bigger cards...

Now Im starting to be curious how well will be 3072 GCN core chip performing here.

The 380 is a 1792SP part, and you expect the 370X to be a 2048SP?

I think AMD debunked that quite sufficiently yesterday.

 

[Erenhardt]

The 380 is a 1792SP part, and you expect the 370X to be a 2048SP?

I think AMD debunked that quite sufficiently yesterday.

How many SP 285 had compared to 280X?

 

[ShintaiDK]

How many SP 285 had compared to 280X?

285 was positioned lower than the 280X.

How much do the 270X got?

380X however may be a full 2048SP Tonga part.

 

[Glo.]

The 380 is a 1792SP part, and you expect the 370X to be a 2048SP?

I think AMD debunked that quite sufficiently yesterday.

Yes, because simply, OEM parts are different than consumer parts. Thats My opinion of course.

 

[nvgpu]

http://www.extremetech.com/gaming/2...map-update-2016-in-mobile-desktop-and-servers

Also, Lisa Su noted that while AMD had started multiple 20nm designs, those projects aren’t going to come to market, due to minimal profitability on that silicon.

Anyone that still mentions 20nm should be ignored.

 

[ocre]

If the rumor about R9 390X single card being almost as fast as R9 295X2 is true, then whole information from Cloudfire is getting more real, than before.

The performance and efficiency gains are phenomenal in this case, for every single card. Lets look at it:

If 370X is 2048 GCN core and performs around GTX780-R9 290, then bigger cards...

Now Im starting to be curious how well will be 3072 GCN core chip performing here.

Cloudfire info....

What about all this information that has been coming out left and right that completely goes against it.

There is already mountains of piles....
Its looking to be a bunch of bull. Just some guy was
trolling him, that's my guess

 

[Maxima1]

What the heck!? Why would anyone think they're releasing a much better 400 series just after launching the 300 series?

 

[utahraptor]

Did these 300 series specs get posted yet?

http://www.amd.com/en-us/products/graphics/desktop/oem#

 

[Silverforce11]

What the heck!? Why would anyone think they're releasing a much better 400 series just after launching the 300 series?

We don't even know what the retail SKUs are gonna be called. 390X is an assumption.

It's actually amazing nothing concrete has leaked and we're a month away from the launch.

 

[Erenhardt]

If 380X is such superior GPU, why on earth would they rebrand tonga, pitcairn and possibly tahiti once again. They could just cut that behemoth 5 times and fill all spots with unfunctional dies.

Also mobile rebrand is just sad. I know it is low end stuff but Cape Verde? C'mon...


But I guess it makes sense to concentrate R&D efforts on new and shiny 14nm than the old-to-be-replaced 28nm process.

It is very likely we will see 380X and 380 as a pipe cleaner for HBM, first driver improvements and market polishing. Then they will have less problems new poducts on 14nm.

 

[Enigmoid]

I think you completely miss the point. Do you really think AMD is launching 300 OEM series now and forward to june. If a new 400 series is coming right away?

While I don't agree often with your other posts I agree here.

This kind of nonsense (300 OEM then 400 retail) does nothing but piss OEMs off. Why is there are re-badges? Because OEMs want new names, they want something new and more interesting, something different to sell to the consumer. Launching 400 series a month later undoes that. All of a sudden they are behind with the old and 'boring' products. There is no point releasing 300 OEM then releasing 400 a month later. Just wait a month and release the entire 400 series for retail and OEM.

http://www.extremetech.com/gaming/2...map-update-2016-in-mobile-desktop-and-servers

Anyone that still mentions 20nm should be ignored.

This was obvious from the start.

What is also obvious is that unless perf/mm^2 can be improved significantly AMD will not make new 28 nm dies for the low/mid range. Unless more profit can be made off the new chips after masks, manufacturing, validation, distribution, etc. compared with the old gcn 1.0, 1.1, 1.2 products and the opportunity cost is sufficiently low (ie respinning mid range 28nm doesn't take valuable personnel and resources that would be better spent elsewhere) then it makes little sense to respin.

 

[csbin]

http://www.chiphell.com/thread-1287686-1-1.html

 

[Silverforce11]

VRM + GPU water cooled out of box, all the benefits of a full loop with non of the hassles.

If its got the performance to back it up, AMD can have my money!

 

[Shehriazad]

http://www.chiphell.com/thread-1287686-1-1.html

Mind = blown. That is SHORT. This would easily fit in almost any ITX case. Like damn... I had no idea that it was gonna be this short...I think it's going to fit into all of my small cases. This card just got a lot more interesting.


If the performance really comes ANYWHERE close to a 295x2...I will grab one for my higher end box. Especially with the water already included it will definitely be cooler and more silent than a Titan X out of the box(Not a fan of 83°C during gaming with stock cooling)...I don't even care if it beats the Titan X...THAT size + Water + High end = VERY interesting.

 

[poofyhairguy]

380X however may be a full 2048SP Tonga part.

If that means a full Tonga for around $200 AR then sign me up! That would be a great value in the midrange I would think.

 

[twjr]

If that means a full Tonga for around $200 AR then sign me up! That would be a great value in the midrange I would think.

It also doesn't make much sense if the 390x turns out to be a 4096sp water cooled monster. The gulf in price and performance is too silly.

Maybe as a 370x it would make sense to have a 2048sp chip at around $200 but not the 380x.

 

[maddie]

It also doesn't make much sense if the 390x turns out to be a 4096sp water cooled monster. The gulf in price and performance is too silly.

Maybe as a 370x it would make sense to have a 2048sp chip at around $200 but not the 380x.

I agree.
Which would imply that we [and everyone else here] don't know whats happening.

 

[Kenmitch]

http://www.chiphell.com/thread-1287686-1-1.html

Birth of a new GPU metric if true. FPS vs board square mm or something similar?

If performance matches the alleged leaks one could have a beastly sff rig.

 

[digitaldurandal]

Yep.

And looks like GCN development is more or less stopped. 2x Performance/Watt in 2016 going from 28nm to 14nm is simply due to a shrink(And not even impressive considering all the marketing talk for the node), while they had to box 2013-2015 in the same box to even list something useful.

We still have to see how much difference HBM does in the 390 series and what cards actually materialize. But I wouldnt get my hopes up since all other series are just rebrands and some even going backwards in the name of performance/watt. Aka 265=370. I cant imagine a massive delta between 380 and 390. And 380 is Tonga.

Sooo 8GB per techpower up.. starting to seem more and more like you are wrong at HBM1. How long until you are wrong about the performance?

 

[Azix]

something new for people to complain about. Tubes not long enough, tubes too long.