• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Hawaii memory configurations

4 GB memory with a 384-bit interface would be weird (but possible). 256-bit are unrealistic for the top dog. Leaves 512-bit.
 
Yeah, Box brings up a good point. Typically the memory bus width is in ratio with the amount of memory, so 4 GB of RAM would be tied to a 256-bit or a 512-bit memory bus. 256-bit is on the small side, especially if the 9970 is to take on the 780/Titan, which leaves 512-bit as the probable outcome.

If 9970 is indeed 512-bit, we can expect a large chip from AMD, which I think is good news considering we could always benefit from some competition.
 
Why would they release a mid size card with more memory then their high end offerings currently.

Because this is just a picture of one whiteboard. How do you know this is the high end part they have sketched out?

Don't you think it's possible that a 384bit 6gb card is probable for the high end?

Nowhere on that picture does it say Hawaii XT. The fact that it's paired with an obvious low end GPU further leads me to believe that this is a mid range memory configuration we are seeing. There's probably another board with Hawaii XT and Hawaii Pro on it.
 
we know who is going to look silly come mid Oct. and thats you. I would like to see what you post after Hawaii launches. :biggrin:

Well if AMD brings out any new uarch I will definitely look silly since I didn't think they would do so on 28nm in the first place.

I will probably say what I already do, it won't be cost effective to move from 28nm to 28nm, just like it wasn't last gen, just like it wasn't this gen with 7970 to 780.

The only thing this can possibly change is AMDs steady decline into the sub $300 market with their best card.
 
Well if AMD brings out any new uarch I will definitely look silly since I didn't think they would do so on 28nm in the first place.

I will probably say what I already do, it won't be cost effective to move from 28nm to 28nm, just like it wasn't last gen, just like it wasn't this gen with 7970 to 780.

The only thing this can possibly change is AMDs steady decline into the sub $300 market with their best card.

you were dreaming that AMD will give up the high end space. AMD has made a very sensible decision to compete with Nvidia at the high end. Thats the only way AMD can improve their GPU market share in consumer and professional / HPC market. Nvidia can kiss their record GPU margins goodbye for the next 12 months especially in consumer market. Tesla / Quadro is still dominant and is more of a long term fight for AMD. It will take years but eventually Nvidia's strangle hold will fall primarily because of rise of OpenCL above CUDA.

Also TSMC 20nm is nowhere near ready. the indications are towards a Q3 2014 volume ramp with Apple gobbling up initial supply leaving very less for Nvidia, AMD and Qualcomm. So AMD's decision is well thought out and their strategy to compete for single GPU flagship is the best strategy in the long run.
 
Last edited:
Considering the fact it took NV several goes to get big die working correctly within thermal envelopes, I cant wait to see how AMD does......new architecture, new drivers....its going to be fun...hope they dont forget about all the multi-GPU and monitor boys and girls, with the frame pacing driver now they have a new product coming out.
 
Well if AMD brings out any new uarch I will definitely look silly since I didn't think they would do so on 28nm in the first place.

I will probably say what I already do, it won't be cost effective to move from 28nm to 28nm, just like it wasn't last gen, just like it wasn't this gen with 7970 to 780.

The only thing this can possibly change is AMDs steady decline into the sub $300 market with their best card.

From what we were told VLIW4 was intended for 32nm, before it was scrapped. They released a cut down version on 40nm. IIRC the original specs were 1820sp, or something like that (Sorry, working from memory. Just best recollection here.). This could be something similar and these 4096sp specs we see floating around (Not realistic on 28nm, IMO.) could be from the original 20nm specs.

Not sure what you are getting at with the part of your post I bolded. AMD made the 7970 and nVidia the 780. How can you go from one to the other. 780 (GK110) is just nVidia's big chip this gen. It simply took longer to get out like the last gen GF110.
 
Remember sushiwarrior commenting on one of my speculations that I should "add some bandwidth"? Given that I already went with 7 Gbps and 384-bit, this and the new speculation (4GB, 512-bit, 1:2 DP) fit well together. You cannot have enough bandwidth for compute, and apparently, AMD is making a Tesla crusher.
 
Remember sushiwarrior commenting on one of my speculations that I should "add some bandwidth"? Given that I already went with 7 Gbps and 384-bit, this and the new speculation (4GB, 512-bit, 1:2 DP) fit well together. You cannot have enough bandwidth for compute, and apparently, AMD is making a Tesla crusher.

If they can do all of that for <$600 (and make money) I'll be impressed. That chip will be ginormous. 😉
 
Remember sushiwarrior commenting on one of my speculations that I should "add some bandwidth"? Given that I already went with 7 Gbps and 384-bit, this and the new speculation (4GB, 512-bit, 1:2 DP) fit well together. You cannot have enough bandwidth for compute, and apparently, AMD is making a Tesla crusher.

yeah I remember him saying AMD did something weird with the memory. I didn't think it would be so freaking weird. :biggrin:

http://hardforum.com/showpost.php?p=1040092133&postcount=143

"It performs like a top card should perform, but I can't say too much. Expect some weird news with the memory system, it's less like recent cards and more like good 'ol Nvidia big-die method. "

the only issue is that massive 512 bit memory bus going to draw a lot of power. that leaves less power for the stream processors. if they clock a 512 bit memory bus at 5500 Mhz for a bandwidth of 352 Gb/s will it draw less power than a 384 bit memory bus at 7000 Mhz for a bandwidth of 336 Gb/s. my guess is 512 bit at 5500 Mhz will draw more power than 384 bit at 7000 Mhz. anyway AMD will have a lot of explaining for their decisions.
 
I remember one of the first leaks I read was that the "9970" would be 512 but the "9950" would not be. Really interested to see how this plays out. With big dies this round and an expensive die shrink the next, it may be awhile before a a good bang for buck upgrade happens to 7950 buyers.
 
These rumor threads are always fun,hoping a Titan destroyer no matter the memory configuration from Amd comes out and while my sig might suggest i have a loyalty to nvidia someone needs to put Nvidia in check before we get some $1,200 Titan Ultra or a next generation $650 gtx880....
 
Back
Top