[Micron]GDDR5X has Arrived

[ShintaiDK]

https://www.micron.com/about/blogs/2016/february/gddr5x-has-arrived

Speeds already hitting 13Gbps. Looks to be 8GB for 256Bit cards as standard.

Positive Manufacturing Progress

There have been a variety of rumors in the industry about GDDR5X availability timeframes, so I’d like to clear that up from the beginning: Micron’s GDDR5X program is in full swing and first components have already completed manufacturing. We plan to hit mass production this summer. The team at our Graphics DRAM Design Center in Munich, Germany is doing a fantastic job, too. Not only do we have functional devices earlier than expected, these early components are performing at data rates of more than 13Gb/s! Memory components mature as they move through the development and manufacturing process, so to see first silicon performing at nearly full performance specs was a pleasant surprise—these early results are incredibly promising. Our first generation GDDR5X is an 8Gb (1GB) density manufactured on our 20-nanometer process technology.

 

[thilanliyan]

If by "arrived" you mean "planned" for the summer, then yes it has
Cheap(er) vram for all!!

 

[DooKey]

If HBM2 is still in short supply along with the interposer then any cards designed with this tech in mind may be what we get in the September/October timeframe.

 

[NTMBK]

If the memory is only hitting mass production in summer, don't expect GDDR5X cards before Christmas.

 

[DeathReborn]

We have only heard from Micron afaik, who knows how far Hynix, Samsung etc are with their production.

 

[JDG1980]

The rumors are that Nvidia's GP100 will be using HBM2, but GP104 will be using GDDR5X. If that's so, then based on the timeline here, we're probably nearly 8 months away from a GP104 release. Micron won't hit mass production until "this summer", so even getting it out in time for the back-to-school season would be a stretch. October-November, maybe.

 

[Despoiler]

If HBM2 is still in short supply along with the interposer then any cards designed with this tech in mind may be what we get in the September/October timeframe.

The interposer is the easiest part of HBM. It was never the cause of HBM supply issues. The TSVs were what was causing the issues. As all of that stuff is reused for HBM2, the only thing that can hold up HBM2 are the modules themselves. At least for SK Hynix and AMD.

As for GDDR5X. 1.35v assuming 256gbit interface assuming this is 13gb/s divided by 2 for a clock of 6.5gb/s. Base specs make it 15% higher power and 20% slower than 4GB HBM1. If it's 384bit that would be 15% higher power for 20% more bandwidth than HBM1. Did I get that right?

 

[NTMBK]

For reference, Samsung announced that they had begun mass production of HBM2 in January.

 

[raghu78]

The rumors are that Nvidia's GP100 will be using HBM2, but GP104 will be using GDDR5X. If that's so, then based on the timeline here, we're probably nearly 8 months away from a GP104 release. Micron won't hit mass production until "this summer", so even getting it out in time for the back-to-school season would be a stretch. October-November, maybe.

yeah. Given that GDDR5X volume production does not get started till June/July I would say a Oct/Nov launch seems possible. I think GP104 is most likely a Q4 product. Polaris 11 with HBM2 could launch as early as July as Samsung has already started HBM2 production in Jan.

 

[jpiniero]

That makes me wonder if Polaris 10 and 11 are using HBM1. No wonder why AMD is going to charge 950 prices for 950 performance on Polaris 10.

 

[Silverforce11]

That makes me wonder if Polaris 10 and 11 are using HBM1. No wonder why AMD is going to charge 950 prices for 950 performance on Polaris 10.

AFAIK, AMD reps have said Polaris 10 is a GDDR5 SKU, and HBM2 will be reserved for higher end parts due to "cost issues".

This suggests Polaris 11 is HBM2.

I'm still seeing a huge gap in their stack if Polaris 11 is high-end, Polaris 10 is low-end... where's the mid-range?!

Fury Nano = mid-range?

 

[raghu78]

AFAIK, AMD reps have said Polaris 10 is a GDDR5 SKU, and HBM2 will be reserved for higher end parts due to "cost issues".

This suggests Polaris 11 is HBM2.

I'm still seeing a huge gap in their stack if Polaris 11 is high-end, Polaris 10 is low-end... where's the mid-range?!

Fury Nano = mid-range?

I think AMD may have 3 GPUs now that we have seen the zauba listings. I think Polaris 10 refers to GDDR5 based Polaris and Polaris 11 refers to HBM2 based Polaris. We might see 2 Polaris 10 GPUs and 1 Polaris 11 GPU.

http://videocardz.com/58101/nvidia-gp104-and-first-polaris-gpu-supposedly-spotted-on-zauba

http://wccftech.com/amd-polaris-gpus-spotted/

Polaris 10 -

Entry level
C913 (102-C91301-00)
C924 (102-C92401-00)

Mid range
C980 (102-C98001-00)
C981 (102-C98101-00)

Polaris 11-

Flagship
C993 (102-C99398-00)

AMD launched GCN in 2012 with Tahiti, Pitcairn, Cape Verde. I don't think they will change that strategy of 3 GPUs to address the entire price range.

 

[tviceman]

AFAIK, AMD reps have said Polaris 10 is a GDDR5 SKU, and HBM2 will be reserved for higher end parts due to "cost issues".

This suggests Polaris 11 is HBM2.

I'm still seeing a huge gap in their stack if Polaris 11 is high-end, Polaris 10 is low-end... where's the mid-range?!

Fury Nano = mid-range?

Probably their best option if they're only going small die and large/huge die with no medium dies in between.

 

[tviceman]

I think AMD may have 3 GPUs now that we have seen the zauba listings. I think Polaris 10 refers to GDDR5 based Polaris and Polaris 11 refers to HBM2 based Polaris. We might see 2 Polaris 10 GPUs and 1 Polaris 11 GPU.

http://videocardz.com/58101/nvidia-gp104-and-first-polaris-gpu-supposedly-spotted-on-zauba

http://wccftech.com/amd-polaris-gpus-spotted/

Polaris 10 -

Entry level
C913 (102-C91301-00)
C924 (102-C92401-00)

Mid range
C980 (102-C98001-00)
C981 (102-C98101-00)

Polaris 11-

Flagship
C993 (102-C99398-00)

AMD launched GCN in 2012 with Tahiti, Pitcairn, Cape Verde. I don't think they will change that strategy of 3 GPUs to address the entire price range.

Having 3 entirely new GPUs would entirely contradict what Raja and other AMD execs have said about GPU releases in 2016. There would be no point to lie to the public about it either, given how low their stock is and Nvidia operating on their own R&D / timetable regardless of what AMD does.

 

[tviceman]

If the memory is only hitting mass production in summer, don't expect GDDR5X cards before Christmas.

That's a bit pessimistic, don't you think? Memory is not nearly as complex as an SoC, CPU, or GPU. They're undoubtedly sharing samples with various customers right now, and if they go into full production in June or July, given the node maturity and tech familiarity, isn't it reasonable to look for Q3 availability?

 

[Silverforce11]

That's a bit pessimistic, don't you think? Memory is not nearly as complex as an SoC, CPU, or GPU. They're undoubtedly sharing samples with various customers right now, and if they go into full production in June or July, given the node maturity and tech familiarity, isn't it reasonable to look for Q3 availability?

Full production in June. First volume shipped to graphics card makers a month later, assembled onto PCBs, tested, building of volumes, release for retail.. definitely Q4 time-frames.

Samsung volume production of HBM2 already means there's a potential release in late Q2 or early Q3 of those parts.

The question here is how bandwidth starved would a mid-range (~300mm2) Pascal be with normal GDDR5 on a 256 bit bus? Do they absolutely need GDDR5X? If so, then it's very likely we won't see it until Q4.

 

[R0H1T]

Having 3 entirely new GPUs would entirely contradict what Raja and other AMD execs have said about GPU releases in 2016. There would be no point to lie to the public about it either, given how low their stock is and Nvidia operating on their own R&D / timetable regardless of what AMD does.

Not necessarily, they might release a Polaris 10 with GDDR5 & then another one with GDDR5x OR the cut down version of Polaris 10 may have GDDR5 with the full chip being equipped with GDDR5x :\

With three types of memory available to them there's a ton a possibilities, not to forget that the first gen of GDDR5x will also be more expensive than the current GDDR5 & I don't see the entry level or even mid tier abandoning GDDR5 anytime soon.

Edit ~ forgot to add, the 3 new GPU theory works if all three of them use different kind of memory otherwise it'll likely be just the two.

 

[Silverforce11]

@R0H1T

Polaris 10 is a very small chip. There's no need for the big increase in bandwidth of GDDR5X (which uses more power than normal GDDR5) for such a product. Low-end, low-cost, GDDR5 is effective enough.

Polaris 11 has to be either HBM2 or GDDR5X (not and/or), since going for GDDR5X means it need a MC, as the MC on HBM2 is on the memory stack. That would double research & development of two separate chip. I don't think AMD is in a financial position to do that and they don't need to. They already have experience with HBM.

So AMD's stack doesn't involve GDDR5X, because they are missing the mid-range chip that needs that bandwidth gains.

 

[R0H1T]

Polaris 10 is a very small chip. There's no need for the big increase in bandwidth of GDDR5X (which uses more power than normal GDDR5) for such a product. Low-end, low-cost, GDDR5 is effective enough.

Don't think that's the case, since as per AT (PIPELINE STORIES) ~

Thanks to reduction of Vdd/Vddq by 10% as well as new features, such as per-bank self refresh, hibernate self refresh, partial array self refresh and other, Micron’s 13 Gbps GDDR5X chips do not consume more energy than GDDR5 ICs (integrated circuits) — 2–2.5W per component (i.e., 10–30W per graphics card), just like the company promised several weeks ago. Since not all applications need maximum bandwidth, in certain cases usage of GDDR5X instead of its predecessor will help to reduce power consumption.

Also, the size of the chip will ultimately decide whether it warrants GDDR5x even if sometime into the future, not necessarily at the time of release. The size of Polaris 10 (correct me if I'm wrong here) hasn't been revealed yet so there's a small chance that it may need GDDR5x.

Polaris 11 has to be either HBM2 or GDDR5X (not and/or), since going for GDDR5X means it need a MC, as the MC on HBM2 is on the memory stack. That would double research & development of two separate chip. I don't think AMD is in a financial position to do that and they don't need to. They already have experience with HBM.

Definitely, I expect it to be HBM2 & anything less would be a step backwards. So unless there's some major hurdle in getting HBM2 (plus interposer) for Polaris 11, there's no need for going that route.

So AMD's stack doesn't involve GDDR5X, because they are missing the mid-range chip that needs that bandwidth gains.

This I don't agree with, as they don't need to go with a 512 bit wide (290/x) GPU & can save some power & money if they get GDDR5x & just a 256 bit wide memory bus, if ever the need arises.

 

[Silverforce11]

Polaris 10 is ~110-120mm2 by accounts of tech journalists who saw it first hand. It is a very small chip.

 

[3DVagabond]

Not necessarily, they might release a Polaris 10 with GDDR5 & then another one with GDDR5x OR the cut down version of Polaris 10 may have GDDR5 with the full chip being equipped with GDDR5x :\

With three types of memory available to them there's a ton a possibilities, not to forget that the first gen of GDDR5x will also be more expensive than the current GDDR5 & I don't see the entry level or even mid tier abandoning GDDR5 anytime soon.

Edit ~ forgot to add, the 3 new GPU theory works if all three of them use different kind of memory otherwise it'll likely be just the two.

Or if the 3rd one isn't called Polaris?

 

[R0H1T]

Or if the 3rd one isn't called Polaris?

Yeah perhaps, but it'll be a risk worth taking IMO only if they have a lead of at least a quarter (or two) over Nvidia.

Polaris 10 is ~110-120mm2 by accounts of tech journalists who saw it first hand. It is a very small chip.

What about the performance? If it's closer to a 380x, instead of GTX 960, then the price bracket (200$ or above) justifies GDDR5x. This, since I don't expect we'd get massive perf/W & per/$ gains simultaneously i.e. we'll pay for the efficiency first & that the GPU costs won't come down as quickly as we'd seen with the first gen 28nm parts.

 

[AtenRa]

I dont see AMD using GDDR5X. GDDR5 will be fine for entry/middle GPUs and HBM2 for high-end GPUs.

 

[Silverforce11]

Polaris 10 Performance would be at max equivalent to a ~220-240mm2 GCN part on 28nm. Well below 380X levels.

It is LOW-end, targeting the 950, not even the 960.

 

[R0H1T]

I dont see AMD using GDDR5X. GDDR5 will be fine for entry/middle GPUs and HBM2 for high-end GPUs.

This year i.e. first gen Arctic Islands?

Polaris 10 Performance would be at max equivalent to a ~220-240mm2 GCN part on 28nm. Well below 380X levels.

It is LOW-end, targeting the 950, not even the 960.

Personally I'm hoping it'd be Tonga level, if not better than that, & anything <960 would be a major disappointment. Have you also taken into account the DX12 gains?