Upcoming Maxwell GPU pictured!!!!

[toyota]

Wonder if the GM204 will be the absolute highest numbered 800 series GeForce GPU and GM200 will be professional/Titan only. Along with Eyrines at Beyond3D, apparently Charlie wrote in Semi-Accurate (only an article reply says this) that the GM200 will be on 28nm: https://semiaccurate.com/forums/showpost.php?p=216233&postcount=4, http://semiaccurate.com/2014/07/01/nvidia-updates-gm204-maxwell-roadmap/ (subscription)

Is there a possible chance that the GM204 will have a 384-bit or 512-bit memory bit bus? The 8GB configuration suggets 256-bit bus or maybe even 512-bit bus but I am not sure if that 8GB is the highest theortical number that can be put on but offically they'll be 3 and 6GB versions. I also think Nvidia could release the GM204 sooner (mid to late september?) if AMD tonga is a surprise.

there is probably no chance Nvidia will use mixed memory sizes on high end card.

 

[dangerman1337]

there is probably no chance Nvidia will use mixed memory sizes on high end card.

Yeah but I was more wondering just because the board can take up to 8GB would mean they would fully use it all ?

 

[f1sherman]

256 bit or 512 bit due to VRAM amount, but since 512 bit and 7 Gbps would be an overkill AND in order to differentiate from GM200 (can you really imagine xx4 part with 512 bit??) - > 256 bit

also 4GB is a nobrainer here

 

[OCGuy]

Get ready for predictions and speculation to be proven or disproven....... :biggrin:

 

[RussianSensation]

I think the key for this card is going to be if it overclocks like GM107. If its stock 15% faster than 780ti but has the huge headroom like GM107 it will be fantastic

But....either way im waiting for big daddy maxwell

But 780Ti OC headroom is also 20-25% so unless it overclocks 35-40%, 15% faster is not enough for these owners. I suppose if it is 15% faster for $549 then it isn't necessarily targeting 780ti owners as upgraders for GM204.
http://www.hardwarecanucks.com/foru...asus-gtx-780-ti-matrix-platinum-review-3.html

I also think Nvidia could release the GM204 sooner (mid to late september?) if AMD tonga is a surprise.

Tonga is supposed to come in between R9 280 and 280X in performance, not the target market for 880. Don't see why NV would care to release 880 earlier because of Tonga. Also, they need time to clear all those GK110 chips. Since the prices for 780/Ti are still very high on average it doesn't seem like 880 is close to launch yet or we would have started to see price drops to clear inventory on these cards.

 

[tviceman]

15% faster than gk110, while a great achievement for being on the same node and a much smaller die, will be very boring and not a worthwhile upgrade for many here.

 

[Galatian]

15% faster than gk110, while a great achievement for being on the same node and a much smaller die, will be very boring and not a worthwhile upgrade for many here.

Well for someone with a small Kepler GPU the switch might be worth it. GTX 680 to GTX 780 was like 20%, put another 15% on it and you got a nice jump.

 

[SlickR12345]

What will make or break these cards are the prices. If they are priced good they will sell, if they have ridiculous prices like the previous generation, I don't see them as worthwhile.

 

[witeken]

What will make or break these cards are the prices. If they are priced good they will sell, if they have ridiculous prices like the previous generation, I don't see them as worthwhile.

They will probably have ridiculous prices.

 

[cytg111]

Yes.. 400 Watts of compute on one card, one chip. Suck it Intel & AMD and your hippie "save the world, emit less co2" - crap.

 

[tviceman]

Tonga is supposed to come in between R9 280 and 280X in performance, not the target market for 880. Don't see why NV would care to release 880 earlier because of Tonga. Also, they need time to clear all those GK110 chips. Since the prices for 780/Ti are still very high on average it doesn't seem like 880 is close to launch yet or we would have started to see price drops to clear inventory on these cards.

I agree that the 880 launch isn't pending, but Nvidia never did any significant price drops on gtx580 or gtx680 when their replacements came. The gtx480 had some huge discounts though, but probably in part to having some serious trade offs.

 

[Ajay]

Fully agree. If they go 28nm now, its basicly a complete long term disaster for dGPU makers. I assume AMD will do the e3xact same due to cost. Then we may end up with a 4 year period before 20nm is even an option.

It totally depends on wafer pricing and yeilds. I would expect that AMD & Nvidia wil find a window of opportunity in the next two years or less. If not, it will be because TSMC is still having a problem with yields. If the same cycle repeats on 20nm as is happening on 28nm then dGPU will be under heavy pressure @ 14/16FF. I think @ 14/16 could be the last hurrah for one of the two dGPU developers. IMHO.

 

[Subyman]

I'm sure that's a test board and likely not the layout we will see. I'd be surprised if they put PSU connectors on the end of the card; they haven't for years.

 

[ShintaiDK]

It totally depends on wafer pricing and yeilds. I would expect that AMD & Nvidia wil find a window of opportunity in the next two years or less. If not, it will be because TSMC is still having a problem with yields. If the same cycle repeats on 20nm as is happening on 28nm then dGPU will be under heavy pressure @ 14/16FF. I think @ 14/16 could be the last hurrah for one of the two dGPU developers. IMHO.

Remember, its not just wafer pricing. Its also IC design cost and time. And that increases rapidly as you shrink.

And its really painful when cost doesnt decrease and volume shinks as well.

 

[tviceman]

Well I don't think it's all doom and gloom. Supposedly 20nm is going to be very short lived, and Nvidia (as well as AMD) may be completely skipping 20nm. 16nm / 20nmFF is going to be in mass production by this time next year. http://focustaiwan.tw/news/aeco/201405180015.aspx

The density jump moving from 28nm to 20nm was quite significant, but the power savings was ho-hum. Being able to go from 28nm to 16nm will be quite a leap.

 

[ShintaiDK]

Just remember its not a true 16nm, its 20nm with FF basicly. The same applies to Samsungs "14nm".

 

[witeken]

Intel formulated it great:

Intel is Continuing to Scale While Others are Pausing to do FinFETs

 

[Ajay]

Just remember its not a true 16nm, its 20nm with FF basicly. The same applies to Samsungs "14nm".

Intel formulated it great:

Intel is Continuing to Scale While Others are Pausing to do FinFETs

Intel decoupled scaling of the metal layers from xtor scaling sometime ago (I think is was either 90 or 65 nm, but I'm not sure). It was less of a problem for Intel at the time because frequency and IPC were scaling better. This should be a one time hit for Samsung, TSMC, etc. - 10nm (just a label now) will be their second pass at FinFet and will likely be much closer to Intel in terms of density. I think that graphic (from Intel) will prove to be overly optimistic. As usual, Intel will likely do better with xtor performance. That said, none of the foundries can afford to make the mistakes they made with 20nm again - fortunately, Intel fumbled a bit @14nm.

So I think the foundries (at least Samsung and TSMC) have learned a good lesson and are making positive adjustments to their R&D structure including spending and hiring. There is one foundry who appears to have responded to late - they need IBM's IP and talent to stop them from becoming the new 'Chartered Semi' (how ironic) - even then, it will be and uphill battle for GF.

Intel does have a lead, but these other companies are literally fighting for their survival - so I wouldn't say that Intel's complete domination is assured. There are nationalistic implications and vendor choice issues which will likely favor the Pure Fabs' survival.

IMHO. May you live in interesting times

 

[tviceman]

Just remember its not a true 16nm, its 20nm with FF basicly. The same applies to Samsungs "14nm".

Yes I am fully aware of what it is, but Intel has absolutely nothing to do with this thread.

 

[HurleyBird]

I'm going to repost what I just wrote on the XS forums here.

Btw... did anyone else notice this?

Well, Maxwell has been long rumored to have integrated arm cores. Seems like Maxwell is a lot more than that though -- it's an SOC! I bet you can run standalone using the extra power connection, or perhaps via a PCI-e interface that only supplies power.