The fact that you can OC some card to x MHz (probably custom, improved PCB too) doesn't mean it would ever be feasible as reference model
AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread
sheepdogexpress
Newcomer
The die is the main bottleneck though and the most costly/time consuming to fix. But go to any gtx 780/titan thread and they are getting to the same clocks as a 7970 average wise.
Add in that most of the asics they use on these gtx 780/Titans are pretty low(nothing in the 80s) mostly in the 60's range and it doesn't take much of a leap of faith that Nvidia could make a gtx 780/titan at much higher clocks when they overclock just just as well as 7970s. The fact that Gtx 780/titans can top performance per watt chart, particularly because they are a big chip shows that these things are majorly underclocked.
For how much Nvidia has been charging for their big chips, they can easily afford to improve the PCB.
A open air but better perfoming cooler than blower gtx titan cooler is probably cheaper too.
Plus I am saying set the clocks at 1050-1100 mhz not 1300ghz.
UniversalTruth
Veteran
Who cares about nvidia finance in this thread? Please, stop spamming and go to the appropriate one!
A1xLLcqAgt0qc2RyMz0y
Veteran
This is the appropriate one because the question was first raised here in post #315. It is just one page back so you shouldn't have a hard time finding it.
And it matters because in post #315 a quote in it from DigiTimes stated that Nvidia would not be able to counter AMD's Hawaii release with a price cut because of a claim by unnamed sources that "Nvidia is unlikely to counter with a price cut because of "weaker than expected profits".
I responded with post #316 stating that Nvidia has the financial revenues and profits to counter AMD if they want to and that the source(s) had no understanding of Nvidia's finances.
Two other response posts (#317 & #326) needed further follow-up. If that counts as spamming then there are many many other offenders here.
There are two sides to every story and if one side brings up an point the other side should have a chance to counter it.
I haven't seen that feature in their product roadmap in a long time now.
I think they have canned that feature.
hUMA ( heterogeneous Unified Memory Access ) have been presented around the 30th April 2013.. Its part of HSA anyway, but i dont know if this will be integrated or not looking the poor information we have about VI ( outside codenames ) ( Do we have a recent "roadmap" for VI anyway ? ).... single memory access space was allready in the roadmap, and HSA integration too anyway. just the name change ( or i do error and there's more change on it ? ).
I have rarely be so confused about "a AMD new series", we have allways tons of rumors, fake and other, with additional real leaks, but theres not so much things about VI outside what seems to be just speculation, and there's even articles speaking about Curucao XT as "HD9970". ( who was initially on the roadmap of the HD8000 but not VI ).
I have rarely be so confused about "a AMD new series", we have allways tons of rumors, fake and other, with additional real leaks, but theres not so much things about VI outside what seems to be just speculation, and there's even articles speaking about Curucao XT as "HD9970". ( who was initially on the roadmap of the HD8000 but not VI ).
Last edited by a moderator:
So what you're saying is AMD has done a good job obfuscating details of their future product roadmaps.
Considering the not so recent news that several high level ex-employees made off with thousands of sensitive documents and files (including licensing agreements, internal confidential work database, etc) before going to work for a direct competitor (Nvidia) the lack of details should come as no surprise.
Considering the not so recent news that several high level ex-employees made off with thousands of sensitive documents and files (including licensing agreements, internal confidential work database, etc) before going to work for a direct competitor (Nvidia) the lack of details should come as no surprise.
It seems yes... But im not sure this is directly linked to Nvidia, because, when you are using the same fundry, the same AIB, i can imagine the " informations" are somewhat not so hard to get for product who should be launched soon when they implied so much peoples ( I say product who should be launched soon, this is for dont enter technology spying question, but we all know how it work ) ... but strangely i think, even on the professional side (FirePro, workstation, Autodesk and 3D professionals softwares developpers, i have never see so much closed mouths, so i imagine, they just lack the informations ( or peoples who get them, are really well choose )...
Even on the rumors side, what we have `? nothing.. I dont know how AMD have sucess to do that, but really we have nothing other than speculation at maybe 1-2 months of the launch. In general at this point, we should get , tons of false leak, but there we dont even got a roadmap, for tell you if this card use a GCN 1.1 or a new generation of GCN (2.0 ) .. I cant tell, if this is a good sign or not.
This said, the affairs, you are talking about, could have a big impact, for predict what your opponnent can do, on a marketing, and developpement side.. and could have an impact on the reason of AMD to have completely change their plan for the 2013-2014.. ( but there, i speculate myself too ) .
Last edited by a moderator:
UniversalTruth
Veteran
WOW!
That's certainly quite intriguing. Are those ex-employees now being sued? What are the procedures?
My bet is that they lost interest/ their friendly attitude to share with the public.
It's a mess confirmed by this sentence as well:
AMD is expected to debut its first "Volcanic Islands" GPUs in Q4-2013, when foundry partner TSMC's swanky new 20 nm node is expected to take flight.
http://www.techpowerup.com/188551/n...islands-and-pirate-islands.html?cp=2#comments
Here, it doesn't necessarily mean that both have connection.
But we still have that leaked result from 3D Mark, probably with an overclocked Hawaii.
Good thing is that risk production of 16 nm will begin THIS YEAR, if we can trust those comments underneath the article.
There is a limit to how "bad" it can be so I wouldn't be worried about that. Hell if they just doubled Pitcairn they'd beat Titan with some ease, and I'd consider anything less than beating Titan a failure - but even if they end up 10% slower it'll be priced reasonably and likely make both Titan and the 780 look as overpriced as they undoubtedly are.
They'd have to go some for it to be really bad, I can't really see how that is possible given the GHz Edition is only 30% behind as it is.
UniversalTruth
Veteran
I think he is being more concerned about the general condition of AMD, and not Hawaii's performance specifically
Well it is pretty much confirmed that it is a relatively big overhaul of GCN. Implementing and improving on the efficiency gains from Sea Islands, removing or reducing the major bottlenecks from GCN1.0, etc.
Don't forget about AMD Radeon Sky!
Edit- Forgot to add that lack of leaks is probably partially due to how they have been handling leaks the last couple of years. People/sources are leery of giving out their info since there has been so much misinformation spread about by AMD on past releases.
Re edit- AMD Rep joined the Anandtech forums recently, Warsam71, and one of the first questions he answered was that his favorite number is 28...
Last edited by a moderator:
I seen that, i'll quote them so people understand better.
His response:Welcome, Sam.
So, what's your favorite number, 20 or 28?
AMD can easily deny this comment means anything so hopefully people do not run away with it.Hi guys, thank you for the welcome
It's 28
Edit: bwahaha the Anand forum thread is great.
Last edited by a moderator:
He response to Steamroller I think. Steamroller is 28nm.
The question he responded to was basically asking him if the new GPU series would be 20nm or 28nm, we already know SR is 28nm.
But like i said, in no way can you use his answer as confirmation because he can easily deny it in many various ways("i like the number 28 more than 20, there is no deeper meaning to my choice" Etc).
Here's another vague and unusable quote:
Response:
My interpretation: Question asked if new 9970 chip is 48CU or 32CU, his reply is that it's niether.LOL - I'm starting to think of the Queen Mary and how it didn't hit anything (iirc)...
So it could potentially be a 40CU chip.
Last edited by a moderator:
My interpretation is that he is a community manager that hasn't confirmed any architectural detail.
Given the latest events in console land, AMD might be better served if their public-facing staff don't know what a silicon whatchamgatchet is.
Also, if he was referencing the RMS Queen Mary, she did hit something. In that incident, it was the other thing that sank.
According to Wikipedia, the Queen Mary collided with an escorting cruiser in WWII, leading to 239 lives lost.
Granted, he should be a community manager and not AMD's official maritime historian.
Given the latest events in console land, AMD might be better served if their public-facing staff don't know what a silicon whatchamgatchet is.
Also, if he was referencing the RMS Queen Mary, she did hit something. In that incident, it was the other thing that sank.
According to Wikipedia, the Queen Mary collided with an escorting cruiser in WWII, leading to 239 lives lost.
Granted, he should be a community manager and not AMD's official maritime historian.
I doubt that was in reference to the CUs. No way they are staying with 32CUs, unless they double up units per CU like G80->GT200.
I have a feeling that was more in reference to the number of ROPs.
Unless they go 40ROPs or decrease the ROPs dependancy on memory bandwidth, i.e. more bandwidth efficient, I don't see why they would go to 48ROPs w/o a 512bit bus.
My reasoning is that info from Tahiti reviews stated that Cayman's ROP's were starving for bandwidth so keeping the number of units the same but increasing the bandwidth by going to a 384b interface allowed them to increase ROP performance. Since going from 32-48ROPs, +50%, and not increasing efficiency or bandwidth by ~50% they would be back where they started.
Edit- Forgot to add Titan-ic. AMD should call Hawaii XT/XTX Iceberg, if they are able to overtake Nvidia in single GPU performance.
You better be listening Dave.