AMD refresh Volcanic Islands

This is the main reason I bought a 7870 for a $199 back in January. It will hold me over until 20nm parts come. If those specs are legit, then sign me up fro an 8970 come Christmas time.
 
AMD do not will launch their next gen on the HD8000 series.. Will be on the HD9000 or whatever they call it... The HD 8000 its just a rehash of the HD 7000 series with the memory a bit faster nothing more that a rebrand.. Also yes, the hawaii architecture will just be awesome if all the reports and leaks are true.. but im just worried about how much power hungry could be that architecture... Also the price.. Those specs are just bestials to think on they'll keep lower price vs Nvidia..
 
Isn't the next fab process .22nm?

I say bull,
512bit memory interface?
Going against the 700 series which is pretty much a rebrand + castrated titan?
 
Yeah the next its 22nm... Thats main reason of rebrand 7000 series just gain some time until they can just skip .22 and go directly to .20 or that was people said.. Nothing confirmed by AMD. Not only the 512 bit interface but the 64 ROPS,256 TMU, and 4096 stream processor... That just crazy.. I guess that was the main reason of they launch only HD 8000 series OEM...
 
Stoly said:
I say bull,
512bit memory interface?
Click to expand...

Why does 512bit strike you as odd? It's been rumored as having that since before Christmas. They're currently on 384bit, so I'm sure that took some by surprised when they went with it, and it might have been a stepping stone for 512bit. Was kinda funny to see nV and AMD swap bus widths this last gen though since it meant nV stepping backwards to 256 heh

At any rate, the 20nm is more curious than ramping up the memory bus. Only reasons I can think of for that are either they are hoping it'll help out for manufacturing the CPUs on that process, or so they can cram an even more powerful GPU into a 24nm APU. Time will tell!
 
My guess: 8gb onboard GDDR5 and hardware -based frame metering for CFX latency issues.
 
FaRKle0079 said:
TSMC is going to 20nm. Not that there's a huge difference between it and 22nm...
Click to expand...

That 2nm difference could add up to a bit of power savings the more cores/transistors there are. And that's never a bad thing. :cool:
 
hmm if ATI could cram 1600 cores at .40, why not 4096 at .20 + ultra wide bus?

NOW... Stable & Optimized drivers on release PLEASE! :D
 
After reading the whole TPU thread that went along with that article, I think I have to agree with some: It's this pic but with inverted/negative (whichever) colors. On top of that, looks like PS4's chip. The article's "specs" indicate 16 serial cores, but there are only 8 individuals, coincidentally the same number PS4 has.

Oh well, it's fun to dream, right? ^_^
 
Stoly said:
Isn't the next fab process .22nm?

I say bull,
512bit memory interface?
Going against the 700 series which is pretty much a rebrand + castrated titan?
Click to expand...

Not sure why it would be so hard to believe instead of updating tahiti it seems as if they are going to try and go to 20nm earlier than nvidia. Also it seems as if it would be the sam kind of jump from the 4870 to the 5870.

I wonder if AMD will try to decrease board length to 10.5 " to be able to fit inside sff cases.
 
KazeoHin said:
My guess: 8gb onboard GDDR5 and hardware -based frame metering for CFX latency issues.
Click to expand...

That 8GB seems to be a lot... I think maybe in a minimun of 4GB up to 6GB wich i think will be more adecuate but only in the higher-end hypotetical HD 9790 or whatever they name the card... To the moment in the way that 4k resolution become more viable and popular (just like the moment of 2560x1600).. More cards will need that amount of vram... I do not think even with the next consoles gen the games need more than 4GB vram... The XFire latency well that its a must to do...
 
DejaWiz said:
That 2nm difference could add up to a bit of power savings the more cores/transistors there are. And that's never a bad thing. :cool:
Click to expand...

Practically you won't really see a difference between 22nm and 20nm. The power savings of going from one process to another in the same node (high-power, high-leakage to low-power, low-leakage) will be significantly greater than 22nm to 20nm. The material stack going from 28nm to 20nm is much more interesting than purely shrinking IMO. Beyond N20 we get even more interesting materials...
 
Araxie said:
That 8GB seems to be a lot... I think maybe in a minimun of 4GB up to 6GB wich i think will be more adecuate but only in the higher-end hypotetical HD 9790 or whatever they name the card... To the moment in the way that 4k resolution become more viable and popular (just like the moment of 2560x1600).. More cards will need that amount of vram... I do not think even with the next consoles gen the games need more than 4GB vram... The XFire latency well that its a must to do...
Click to expand...

you can't put 6GB on a 512bit bus

hence the reason bus sizes have "set" amounts of memory

GTX 580 = 1.5GB, 3GB, 384bit bus
7970 = 3GB, 6GB, 384bit bus
GTX 680 = 2GB, 4GB, 256bit bus
 
DejaWiz said:
That 2nm difference could add up to a bit of power savings the more cores/transistors there are. And that's never a bad thing. :cool:
Click to expand...

20nm and 22nm are just labels. The actual size of the transistors will vary and for the most part not be what you'd expect. Kind of like Intel's 45nm and TSMC's 40nm offering nearly identical density despite the theoretical 5nm difference.
 
Last edited:
refraxion said:
Will be keeping an eye out for this!
Click to expand...

Pick up a lot of eye drops because you'll be keeping that eye out for a long time and I fear it'll begin to look a lot like a raisin :D [See my previous post as to why]

Not saying I don't have a hope of it being real, or that even if the "die" image is fake that I don't wish those specs to be true, just not likely and is nothing but "fan fiction" :(

Still, one doesn't know. The biggest and most interesting aspect of the PS4 [in my opinion] is its shared memory architecture for GPU-CPU, something a lot of us figured was one of those "Sony customizations" to the APU that Cerny spoke of, which AMD just announced is coming to their consumer APU lines. So if those 8000-series specs are real but mainly based on PS4 GPU tweaks and so forth, it could very well be they end up in PC's integrated and dedicated offerings... after all the PS4 is 512bit.
 
Reality said:
you can't put 6GB on a 512bit bus

hence the reason bus sizes have "set" amounts of memory

GTX 580 = 1.5GB, 3GB, 384bit bus
7970 = 3GB, 6GB, 384bit bus
GTX 680 = 2GB, 4GB, 256bit bus
Click to expand...


Citation, please.
 
Solomutt said:
Citation, please.
Click to expand...

Are you just unfamiliar with that, or are you doubting him? :confused:

Nevertheless, here's a good explanation for you:
http://www.techpowerup.com/forums/showpost.php?p=2699537&postcount=4

Boils down to the basics of how computers have been designed, everything pretty much revolves around 8bit and multiples of that. Which I think is what is mentioned later in that thread, called "Power of 2". Anyways, in the example it shows you CAN, but it's just not smart [ideally efficient]. We could probably see DIMMs in 3GB modules with 2GB on one side and 1GB on the other, except it'd complicate things when addressing the available space.
(note: this is just my understanding, but I'm pretty confident it's the case)
 
one might point out the 2GB 660 on a 192bit bus, surely just the same process in reverse?
 
Formula.350 said:
Are you just unfamiliar with that, or are you doubting him? :confused:

Nevertheless, here's a good explanation for you:
http://www.techpowerup.com/forums/showpost.php?p=2699537&postcount=4

Boils down to the basics of how computers have been designed, everything pretty much revolves around 8bit and multiples of that. Which I think is what is mentioned later in that thread, called "Power of 2". Anyways, in the example it shows you CAN, but it's just not smart [ideally efficient]. We could probably see DIMMs in 3GB modules with 2GB on one side and 1GB on the other, except it'd complicate things when addressing the available space.
(note: this is just my understanding, but I'm pretty confident it's the case)
Click to expand...

Yes, it's technically possible, but requires a lot of work to the memory controller that AMD and Nvidia would rather avoid having to do. It's hard enough for them to ramp the speed up as it is and adding more complexity would be counter-intuitive.
 
R3MF said:
one might point out the 2GB 660 on a 192bit bus, surely just the same process in reverse?
Click to expand...

I'm not sure I'm following you...?

But...
The way they got 2GB on a 192bit bus was by (inefficiently) cramming two more ICs onto a channel. Which lets use an Intel X58 for example, it'd be like running it and using DDR3 in 512MB | 512MB | 1024MB. Ideally, not something anyone would do, but people who buy prebuilt systems aren't generally enough in-the-know to realize that. That's why we see laptops and desktops with 3GB, 6GB, 12GB. The latter isn't as much of a problem since they usually have 4 DIMM slots and then 3 are filled with 4GB modules; however, the first two you usually see mismatched with a 2GB+1GB and a 4GB+2GB. Granted, 3GB config was generally seen years ago and I doubt you'd find one with that amount

But I digress, as mentioned in that thread, it's all basically a marketing ploy. In the real world it may not effect performance too much, but given the right circumstances it could. Though a computer with mismatched modules running in dual-channel will probably be impacted more than a graphics card would be. The benefit of a computing being configed that way is you can rectify it, a graphics card... not so much (at least not easily).

I'll admit though, I had always thought it'd be cool to see memory sold in oddball packages like that lol 1.5GB, 3GB, 6GB modules :p

EDIT: Ooohh I think I gotcha. You mean REMOVING some memory ICs from a card. Sure, you could, but again it'd impact performance. You'd be better off doing what AMD did (I think) with GCN, by lasering or whatever part of the memory controller and turning the 384bit into a 256bit for the 7800s. Or as this Wiki snippet says, might be more akin to modular memory controller made up of 6 units at 64bit each: "While each bigger die has two additional memory controllers widening its bus by 128 bits" ("two additional" 64bit controllers = 128bit). But yea I still wouldn't want to buy a card that is packaged like that. I'd rather it be intentionally crippled at the memory controller and be confident that my performance won't scale oddly when texture storage starts utilizing the additional, or in this case absent, memory ICs.
 
Last edited:
Araxie said:
I guess that was the main reason of they launch only HD 8000 series OEM...
Click to expand...

There is no "Only OEM 8000 series", they're just doing the usual AMD shindig and relaunching a few low end cards. There should be a Hawaii HD8xxx series.

Araxie said:
AMD do not will launch their next gen on the HD8000 series.. Will be on the HD9000 or whatever they call it... The HD 8000 its just a rehash of the HD 7000 series with the memory a bit faster nothing more that a rebrand..
Click to expand...

Again, there are a few OEM only low-end cards but the plan is for Hawaii to the the upper end of the 8000 series. Hell, almost all the 8000 series will be new silicon, there's just a few rebrands in the LOW end, and ONLY for OEMs. Hawaii should still be 8000 series.
 
sushiwarrior said:
There is no "Only OEM 8000 series", they're just doing the usual AMD shindig and relaunching a few low end cards. There should be a Hawaii HD8xxx series.

Again, there are a few OEM only low-end cards but the plan is for Hawaii to the the upper end of the 8000 series. Hell, almost all the 8000 series will be new silicon, there's just a few rebrands in the LOW end, and ONLY for OEMs. Hawaii should still be 8000 series.
Click to expand...

He refers to this.
http://www.amd.com/us/products/desktop/graphics/8000/pages/8000-series.aspx#2
 
sushiwarrior said:
Yes, there is an OEM only part of HD8000, it doesn't mean HD8000 is exclusively rebranding though. Hawaii should still end up as the "actual" HD8970 or what have you.
Click to expand...

This. AMD is bringing their mobile chip rebranding strategy to their dGPU line-up...but only for OEM white-box or brown-box parts. The consumer level offerings of the HD8000 series will be the new architecture. Thought this had already been beat to death.
 
folks its not VI it cold however be a dead ringer for kavari or an engineers wild fantasy time will tell.

"Although the image of the system-on-chip is extremely blurry, it is evident (if enlarge, sharpen, filter it and then add some logic) that it has all the blocks that the next-generation accelerated processing units are supposed to have. The SoC has security co-processor (ARM Cortex-A5), AMD-V, AMD-P blocks, unified northbridge (UNB), system controller hub (SCH), HyperTransport links, PCI Express 3.0 lanes, CrossFireX interface, eight 72-bit ECC DDR3 memory controllers, display controller, eight Steamroller-class x86 modules (with two integer/SPU cores, one FPU and L2 cache), unified L3 cache as well as a decent GCN architecture-based graphics processing unit."

http://www.xbitlabs.com/news/other/...024_Stream_Processors_512_Bit_Memory_Bus.html
 
Last edited:
Geforcepat said:
That's nice but im gonna keep both of my eyes open:D

And no Hawaii will not be an 8000 series...sorry
Click to expand...

Ya, some people still waiting HD 8000 to be hawaii.. And the fact its even that HD8000 are available.. But only OEM as a rebrand of 7000 series..
 
Araxie said:
Ya, some people still waiting HD 8000 to be hawaii.. And the fact its even that HD8000 are available.. But only OEM as a rebrand of 7000 series..
Click to expand...

I can't tell if you are just a troll in every subforum you post in, or just so uneducated that you believe the stuff you spew out..You are the first official entry to my ignore list, and I have tolerated a lot (of your troll like kind) that most people couldn't...:eek:
 
ccityinstaller said:
I can't tell if you are just a troll in every subforum you post in, or just so uneducated that you believe the stuff you spew out..You are the first official entry to my ignore list, and I have tolerated a lot (of your troll like kind) that most people couldn't...:eek:
Click to expand...

yeah bud of course:

http://www.amd.com/us/products/desktop/graphics/8000/pages/8000-series.aspx#2

http://www.amd.com/us/Documents/AMD_Radeon_HD_8970_Feature_Summary.pdf

http://www.amd.com/us/Documents/AMD_Radeon_HD_8950_Feature_Summary.pdf

http://www.techpowerup.com/gpudb/1972/radeon-hd-8970-oem.html

i love to troll..!!
 


From xbit what is in the picture is not vi folks. it is however a future amd apu most likely on a 22nm/20nm process with VI gfx

picture
 
You must log in or register to reply here.
Back
Top