AMD Radeon HD 9970 Specifications Speculations

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

kog91

Honorable
Nov 2, 2012
24
0
10,510
This seriously needs to not be circulated anymore.
1) This is not a GPU design, it is an APU design.
The SPU's or Serial Processing Units are simply Bulldozer-esque "Modules".
It also has a northbridge and southbridge.
The memory units are for DDR3, not GDDR5 (which is what you would normally use for a high power GPU).
It has a HyperTransport bus. There is no useful purpose to have this on a GPU.
2) This is an Opteron APU.
The RAM is ECC which is not yet a mainstream consumer part.
It includes an ARM security module, which has been a part of Opteron CPUs for a little while.
It has platform management capabilities, only found in servers.
It has Advanced Server RAS (see http://www.intel.com/content/www/us/en/servers/reliability-availability-and-serviceability-for-the-always-on-enterprise-paper.html)
 

bowzef

Distinguished
Oct 18, 2010
543
1
19,010
yeah if you read and look at diagram it does in fact say DDR3, no were to be seen as GDDR5, also this is all just BS, this is def APU
 

InvalidError

Titan
Moderator

I wonder how much of recent alleged Haswell overclock and new "GPU" announcements are manufactured/astroturfed news.

The picture in this article clearly does not match the story so either at least one side is all made-up or the truth may lie somewhere in-between.
 

bowzef

Distinguished
Oct 18, 2010
543
1
19,010

i personally think it's APU and that current spec's that were listed are not going be true, reason why is as i said before diagram shows DDR3 memory and having such spec's to run on DDR3 is pointless and huge bottle neck to those spec's. next gen graphics will not be double it will be minor performance like we seen in past it's going be 10 to 15% more GPU power over 7000 cards from AMD. back to the diagram 72bit DDR3 x8 now if this is remotely correct this would mean 576bit, now we all know GDDR5 is not same as DDR3, we all know GDDR5 is high memory bandwidth, more Bits u has on memory don't mean anything if Bandwidth is not there theoretically its like having HD7970 with 6-8gbs VRAm using DDR3 memory, just does not make sense
 

somebodyspecial

Honorable
Sep 20, 2012
1,459
0
11,310
NV doing the same thing with maxwell on 20nm...This should be no surprise and why I skipped current gen. No die shrink means small increases. You get real gains at the shrinks where they have a lot more watts and room to play with.
I'll be more interested if it doesn't take them a year+ to get their drivers right again (which you're still waiting on working enduro for notebooks and working prototype driver for CF/runts/stutter people on desktop). Unless this card is pounding NV perf wise (doubtful, they're doing the same making them close to equals already) I will go with the best drivers. All the hardware in the world doesn't matter if it takes a year+ to use it all. At this point it hasn't been proven the current stuff will work totally until they fix stuff...proving my point - Prototype after Q2 is NOT specific and a year late already, likely a year and a half before prototype hits and should have been in the box. I won't buy a product that doesn't work as advertised, and I'm not your dang beta tester for a year+.
Queue down ratings from AMD lovers who miss the point.
 

tigger888

Distinguished
Nov 16, 2011
57
0
18,640
wow.. this was pulled from semiaccurates forums. when they asked the guy what it was about he said “I thought I made myself clear that I was just illustrating a point in transistor scaling from 28 nm to 20 nm when I first posted the image on the forums"
talk about reposting stuff without any verification.
 
Wow. These are big numbers. If these are real and AMD can get a handle on microstutter and get it well-documented, there will definitely be real competition in the marketplace again.
I wonder if they'll release a 9700 Pro...?
 
Status
Not open for further replies.