[Videocardz rumour] Vega pushed forward to October

[xpea]

I don't believe one second this rumor. It was launched by an AMD fanboy. The kind of guy who thinks that 512bit 320GB/s memory is faster than 256bit 320GB/s

 

[Krteq]

Whaa?

 

[ShintaiDK]

Welp, no

There is a slide from eSilicon presentation for 14nm LPP HBM2 integration - test chip tepe out at March 2016 - Final Models availability at September 2016

That's the HBM2 controller only. And HBM2 is already in production and on a card today, P100.

 

[Ancalagon44]

The thing is the improved perf/W compared to Polaris. I doubt we will see many architectural improvements so soon to account for this, implying HBM2.

Small Vega can use 2 stacks of HBM2 and still have the bandwidth of FuryX. More than enough. Also allows a small interposer and lower assembly costs. 4 components vs 6.

Chicken and egg. For HBM2 based products to drop in price, you have to increase production and vice versa.

Bear in mind that while they said that Vega will have improved perf/W compared to Polaris, this doesn't necessarily mean that every single Vega will be more efficient than every single Polaris.

It could just mean that the Vegas that come with HBM2 will be more efficient than the Polaris cards that come with GDDR5 or GDDR5X

 

[airfathaaaaa]

That's the HBM2 controller only. And HBM2 is already in production and on a card today, P100.

"today" is kinda hard when they gonna actually start giving those out on 2017

 

[ShintaiDK]

"today" is kinda hard when they gonna actually start giving those out on 2017

You mean June 2016.

 

[railven]

Unfortunately for AMD, using the far superior cooling system brought too much negative publicity due to poor launch execution. With a properly working AIO CLC, most people would be amazed at how well it works on 250-300W cards, or even 500W cards.

And this is exactly why I don't think they'd risk it. Just let their AIB's take care of something fancy.

 

[airfathaaaaa]

You mean June 2016.

there is really a common missunderstanding...

nvidia never said they will launch on june what they said is that they are gonna start shipping boards to manufac first and by mid june to "some" (3) hpc partners..
the rest is due to q1 2017 hbm isnt ready for anything of that magnitude so early everyone knows it

 

[ShintaiDK]

there is really a common missunderstanding...

nvidia never said they will launch on june what they said is that they are gonna start shipping boards to manufac first and by mid june to "some" (3) hpc partners..
the rest is due to q1 2017 hbm isnt ready for anything of that magnitude so early everyone knows it

You can buy DGX-1 systems in June. And that's not 2017.

 

[PPB]

I don't believe one second this rumor. It was launched by an AMD fanboy. The kind of guy who thinks that 512bit 320GB/s memory is faster than 256bit 320GB/s

It is, assets with data widths higher than 256b needs to be decomposed in smaller fractions, thus needing more cycles than a 512b data chunk being processed by a 512b width bus. That is why going wide with HBM/2 pay dividends in 4K res and the whole Maxwell lineup besides the 980ti take such a serious hit going abive 1440p

Sent from my XT1040 using Tapatalk

 

[Ma_Deuce]

You can buy DGX-1 systems in June. And that's not 2017.

It's also not today.

 

[dark zero]

That's the HBM2 controller only. And HBM2 is already in production and on a card today, P100.

And that is a rumor too like nVILINK since it got banned from Intel X86 and are forced to be with Power Architecture.

 

[airfathaaaaa]

You can buy DGX-1 systems in June. And that's not 2017.

they have quota in place for oem and consumers meaning that they wont launch at all this year its really very simple

 

[moonbogg]

AMD will release Vega early and make the 1080 look ridiculously overpriced and expose it for the mid range card that it is. This will force Nvidia to react early with big pascal, which will be a punch right in the nose to those who spent seven hundred dead presidents on their new mid range 1080.

TAKE IT TO THE BANK. Like I'm wrong. Puh-lease

 

[MrTeal]

AMD will release Vega early and make the 1080 look ridiculously overpriced and expose it for the mid range card that it is. This will force Nvidia to react early with big pascal, which will be a punch right in the nose to those who spent seven hundred dead presidents on their new mid range 1080.

TAKE IT TO THE BANK. Like I'm wrong. Puh-lease

You forgot the inevitable conclusion to the drama, when AMD drops a 16384 SP quad die Vega 11 on interposer chip with 32GB HBM2 in early 2017.

 

[Stuka87]

I don't believe one second this rumor. It was launched by an AMD fanboy. The kind of guy who thinks that 512bit 320GB/s memory is faster than 256bit 320GB/s

Do you understand how a bus works? The bus is the interface between the GPU and the memory. The wider the bus, the more data can be processed in a single cycle. This is a big deal if a datagram is longer than the bus width, as it then requires that memory to be broken up into multiple cycles. Meaning the GPU is having to wait longer to process that memory.

Sure you can have a highly clocked 256bit bus and get decent performance, but it will suffer at higher resolutions in comparison to a 512bit bus.

 

[airfathaaaaa]

seems like that serious site once more proved that it has become worse than wccft....

http://www.forum-3dcenter.org/vbulletin/showthread.php?p=11027263#post11027263
all the rumor is based on this and
http://www.forum-3dcenter.org/vbulletin/showthread.php?p=11027319#post11027319

seriously a random guy on a random forum is enough to make news now

 

[MrTeal]

Do you understand how a bus works? The bus is the interface between the GPU and the memory. The wider the bus, the more data can be processed in a single cycle. This is a big deal if a datagram is longer than the bus width, as it then requires that memory to be broken up into multiple cycles. Meaning the GPU is having to wait longer to process that memory.

Sure you can have a highly clocked 256bit bus and get decent performance, but it will suffer at higher resolutions in comparison to a 512bit bus.

Uhh....
Ignoring the larger complexities of multiple data rate buses, if you have a 2GHz bus a transaction will take 0.5ns to complete. If you need 1024 bits of data, it will take four cycles or 2ns. If you have a 512bit bus you can do it in two cycles, but if that bus is clocked at 1GHz the whole transfer will take... 2ns.

 

[railven]

You forgot the inevitable conclusion to the drama, when AMD drops a 16384 SP quad die Vega 11 on interposer chip with 32GB HBM2 in early 2017.

For $500!!!! Haha.

EDIT:

seems like that serious site once more proved that it has become worse than wccft....

http://www.forum-3dcenter.org/vbulletin/showthread.php?p=11027263#post11027263
all the rumor is based on this and
http://www.forum-3dcenter.org/vbulletin/showthread.php?p=11027319#post11027319

seriously a random guy on a random forum is enough to make news now

Tis the new age of "journalism" If one of us posts something on reddit and add "legit, I work for AMD/NV, here my Employee ID is 145034" you can bet your read end there will be an article about it.

 

[Riek]

Uhh....
Ignoring the larger complexities of multiple data rate buses, if you have a 2GHz bus a transaction will take 0.5ns to complete. If you need 1024 bits of data, it will take four cycles or 2ns. If you have a 512bit bus you can do it in two cycles, but if that bus is clocked at 1GHz the whole transfer will take... 2ns.

FYI: the 512bit bus on hawai was smaller than the 256bit controller on Tahiti. So I don't think saying it is more complex is false as running at very high frequencies is also a pain in the bottom for the memory controller.

 

[MrTeal]

FYI: the 512bit bus on hawai was smaller than the 256bit controller on Tahiti. So I don't think saying it is more complex is false as running at very high frequencies is also a pain in the bottom for the memory controller.

Couple things.
1. I never said anything about area, just talked about performance.
2. Never said getting a memory controller to run at high frequencies wasn't a challenge.
3. The 512-bit controller on Hawaii isn't smaller than the 256-bit controller on Tahiti, as Tahiti doesn't have a 256-bit controller.

My point of contention was that you don't have to wait longer for a chunk of data with a bus half the size of a large one if that smaller bus sends data twice as often, all else being equal.

 

[Aristotelian]

AMD will release Vega early and make the 1080 look ridiculously overpriced and expose it for the mid range card that it is. This will force Nvidia to react early with big pascal, which will be a punch right in the nose to those who spent seven hundred dead presidents on their new mid range 1080.

TAKE IT TO THE BANK. Like I'm wrong. Puh-lease

This is becoming a difficult time for people like me (and I think there are a few) that want to build a nice PC this year. Sooner rather than later. If I went 1080 Tri SLI watercooled and a top Vega with 16 HBM2 came out where say, 1 Vega was as fast as 2 1080s (roughly), I'd be an idiot buying the 1080s in June with such a performance increase coming soon after. And yet, for the past while, it's always (in my mind) "The 980Ti is a great card, but Broadwell-E isn't out. Wait for Broadwell-E. The 1080s will be out when the 6950X is, so get those and watercool them, get a new monitor - but there's no HDR 4k monitors available yet. And worse, the 1080s will be stomped either by a 1080Ti or Vega, or whatever the name of Nvidia's HBM 2 monster Titan will be, so...wait".

And I sit here on a 2600k overclocked with 580s in SLI playing at a peasant 1440p as I have been since 2010 or so, and in modern games I dial it down to 1080p :'(

Please, to the gods that run these companies, put out the blistering performance parts now! Didn't any of them want to catch those of us building new broadwell-E systems with top end parts that will not make us wonder why we dropped thousands of dollars on mid range parts NOW? Please!

 

[railven]

This is becoming a difficult time for people like me (and I think there are a few) that want to build a nice PC this year. Sooner rather than later. If I went 1080 Tri SLI watercooled and a top Vega with 16 HBM2 came out where say, 1 Vega was as fast as 2 1080s (roughly), I'd be an idiot buying the 1080s in June with such a performance increase coming soon after. And yet, for the past while, it's always (in my mind) "The 980Ti is a great card, but Broadwell-E isn't out. Wait for Broadwell-E. The 1080s will be out when the 6950X is, so get those and watercool them, get a new monitor - but there's no HDR 4k monitors available yet. And worse, the 1080s will be stomped either by a 1080Ti or Vega, or whatever the name of Nvidia's HBM 2 monster Titan will be, so...wait".

And I sit here on a 2600k overclocked with 580s in SLI playing at a peasant 1440p as I have been since 2010 or so, and in modern games I dial it down to 1080p :'(

Please, to the gods that run these companies, put out the blistering performance parts now! Didn't any of them want to catch those of us building new broadwell-E systems with top end parts that will not make us wonder why we dropped thousands of dollars on mid range parts NOW? Please!

If you're going to wait for something better, you'll be stuck waiting. Now, I'm not advocating you buying a particular brand/product, but if this post is any kind of reflection of your train of thought - you'll regret whatever you do, to some degree.

Just buy something. Anything. Enjoy it. And don't let the "well if I waited x-months I could have gotten this" because it's just going to keep repeating. I mean, even if get 1080 Ti - the 980 Ti itself was not a full GM200 chip, there could have always been a 980 Ultra. Just food for thought.

 

[Arachnotronic]

You forgot the inevitable conclusion to the drama, when AMD drops a 16384 SP quad die Vega 11 on interposer chip with 32GB HBM2 in early 2017.

And sells it for $549

 

[S.H.O.D.A.N.]

Given the lack of developments on the HBM2 front, the only way I can see AMD going for an October release with Vega is with GDDR5X based SKUs. Might have something to do with hynix hitting mass production a bit earlier than anticipated.