AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Raja saying that the RX part will run faster for games can mean a number of things. Less HBM2 (8GB) at faster frequency offering higher b/w? Higher boost clocks? High overclock headroom? At least the part about Volta(ge) means to me that they won't be limiting the GPUs as much as Nvidia does, doesn't say much about how these will overclock but it's a good sign for the people that want to put in the effort.
 
Before Vega you mean?
You would have known that AMD's engineering groups have been calling every project SoC internally since a few years ago, if you manage to triage through piles of LinkedIn profiles. They just don't call things SoC for marketing, because SoC has a more specific and perceived meaning when it comes to branding.
 
Raja saying that the RX part will run faster for games can mean a number of things. Less HBM2 (8GB) at faster frequency offering higher b/w? Higher boost clocks? High overclock headroom? At least the part about Volta(ge) means to me that they won't be limiting the GPUs as much as Nvidia does, doesn't say much about how these will overclock but it's a good sign for the people that want to put in the effort.

Or just higher clockspeed, and that the drivers are the "professional one" ( Same as for FirePro ), and that the catalyst driver will be more optimized so for gaming.
 
Last edited:
You would have known that AMD's engineering groups have been calling every project SoC internally since a few years ago, if you manage to triage through piles of LinkedIn profiles. They just don't call things SoC for marketing, because SoC has a more specific and perceived meaning when it comes to branding.

Oddly enough, in this case there are specific additions for SOC15 in the latest patches. This was also a code name that came up for the IP integration portion of Project Skybridge in 2012. Whether this is an instantiation of that (rather late?), or a naming collision is unclear.
 

That link seems to take me to a slightly off portion off the thread. However, if it's the post concerning the public definition of SOC, I'm not sure how that clarifies the specific usage SOC15 across AMD's disclosures via its executives in 2012 concerning Skybridge and what is now showing up associated with Vega and recently Raven Ridge.
 
I'm reaching just a bit with this one but this just popped up: http://forums.guru3d.com/showpost.php?p=5431939&postcount=367
MSI AB 4.4.0 beta 9 is available:

http://office.guru3d.com/afterburner...up440Beta9.rar

New version introduces new Quad Overvoltage(tm) architecture support for upcoming custom design MSI graphics cards, which means that you'll be able to control up to 4 independent votages on some hi-end MSI graphics cards (core voltage, memory voltage and 2 auxiliary voltages).

Nvidia is known to restrict AIBs on allowing voltage controls so I don't see how this is not about some sort of MSI custom Vega variant.
 
I'm reaching just a bit with this one but this just popped up: http://forums.guru3d.com/showpost.php?p=5431939&postcount=367


Nvidia is known to restrict AIBs on allowing voltage controls so I don't see how this is not about some sort of MSI custom Vega variant.
Not entirely true and depends on the AIB and card characteristics. Here is one case where the normal power limit of 120% was exceeded using Evga's tool PrecisionXOC.

Edit: When I had my Classifieds they provided additional hardware (EvBot) to control voltages. Later on they provided software for anyone to change the voltages. I think if the ABI got the nod from Nvidia then they can go ahead with voltage customizations, but in this case if it's Afterburner I don't know who is calling the shots .. MSI or UnWinder.

NVVDD GPU Voltage
FBVDD Memory Voltage
PEXVDD1 PCIe Voltage #1
PEXVDD2 PCIe Voltage #2
OCP OverCurrent Protection
.
 
Last edited by a moderator:
Not entirely true and depends on the AIB and card characteristics. Here is one case where the normal power limit of 120% was exceeded using Evga's tool PrecisionXOC.

Edit: When I had my Classifieds they provided additional hardware (EvBot) to control voltages. Later on they provided software for anyone to change the voltages. I think if the ABI got the nod from Nvidia then they can go ahead with voltage customizations, but in this case if it's Afterburner I don't know who is calling the shots .. MSI or UnWinder.

NVVDD GPU Voltage
FBVDD Memory Voltage
PEXVDD1 PCIe Voltage #1
PEXVDD2 PCIe Voltage #2
OCP OverCurrent Protection
.

And EVGA have got a lot of problem with Nvidia after that ( who have put the Greenlight program under the public spotlight ) ..

But for be honest, 4 independant voltage dont mean much lol .
 
Could go either way on the changes to MSI Afterburner as both have logical reasons.
For Nvidia they do quietly allow for extreme OC-voltage and this is because of the professional benchmark record events/competitions market but these are usually only the very top model in the 1080 and 1080ti lineup by a couple of manufacturers.
As an example Galax with HoF in the 1080 have gone with International Rectifier IR3595A, with the 1080ti HoF not sure what they are using from IR but with software that is available to certain owners of these card this can be configured (including exceeding the standard voltage ceiling).
You could also get signed drivers for the top model 1080 from Asus and probably will again with their very top 1080 ti that were over 1.25V, again though this is not easily accessible for the average consumer.
And their specific drivers for the higher voltage do not seem to do much on other 1080 models (and I would think same applies to 1080ti as well) - worth noting those more unique 1080 cards were still within the same or similar power limit though.
MSI may be joining in this as well.

But then with AMD launching Vega some point for consumer, and everyone knows AMD is happy to let their owners OC like mad in terms of pushing voltages.
Cheers
 
Dont rule out 16 hi + 0 hi.

:mrgreen:

Yeah, it's not like 3.75 Gbps stacks show up in SK Hynix's product catalog either, so who knows. We can't rule this out yet.

I bet they disable one of the stacks for yield purposes. Those 16-hi 3.75 Gbps stacks probably aren't getting very good yields.
 
Dont rule out 16 hi + 0 hi.

:mrgreen:
Thermoelectric coolers, where you stack em as deep as you wish and screw the power consumption!

Kind of curious if they actually tried that, although IBM patented it already. Sandwiching TECs between the stacked dice to facilitate heat transfer could provide really interesting results. Just imagine the clocks, capacity, and form factor of that!
 
Only way you get 16 GB with 2 stacks is 8-Hi stacks

My original post said 8 GB HBM2 with 8 GB GDDR and was ment to be largely tongue-in-cheek, although that scenario would possibly account for dramatic PCB size increase over Fury X as well as shed some light onto HBCC concept.
 
Knowing AMD, custom variants for Vega would take another 2-3 months after Vega release, so I doubt he's releasing new AB for something 6 months down the line. Could be a custom 580 if nvidia are that loath to voltage control on their cards.
 
Knowing AMD, custom variants for Vega would take another 2-3 months after Vega release, so I doubt he's releasing new AB for something 6 months down the line. Could be a custom 580 if nvidia are that loath to voltage control on their cards.

AIB are set to present them the 31th on Computex.. dont know if this will be OC variants.. but well.

In fact if we have first hear the date of 31th it is due to some AIBs who have been a little bit too much verbals about what they will show there. ( thanks MSI )
 
Back
Top