AMD Vega Speculation Questions

FoxKitsune

Honorable
Apr 25, 2015
68
0
10,630
So with Vega on the Horizon and myself planning to build a new system come time around Black Friday I figured I'd ask a few of my questions here as elsewhere on the internet I have found conflicting answers and the such. Now before all the questions I just want to make a couple things known:

-I am no computer expert here
-Yes I know all we really know about Vega is that it has HBM2 with 4GB and 8GB models and not much else
-Questions 1 and 2 are complete speculation
-I am dead beat tired as I'm typing this so please correct me on grammar, terminology, and the such

1. It has been officially announced that Vega will be sporting HBM2 VRAM in both 4GB and 8GB varients to compete with Nvidia's 10 Series cards that are sporting GDDR5 and GDDR5X memory Now to my understanding the difference between HBM and GDDR memory is that GDDR memory favors core clock speed over memory bandwith/speed and HBM favors bandwith/speed over core clock speed (which is no surprise considering it's called High Bandwith Memory so it's kinda in the name). I have also heard that AMD has supposedly made a claim that their 4GB models will give the 1080 Ti some competition done to the increased bandwith. So does this mean that the extra bandwith makes up for the lack of total memory and lower clock speed? Is this theoretically true?

2. Another thing that I've read up on is that another major difference between the two is that for GDDR its 1GB per chip (ex: 2GB=2Chips, 4GB=4chips, etc.) whereas HBM the memory (specifically HBM2) can have either 4GB or 8GB per stack (ex: 4GB=1stack or 8GB=1stack/2stacks). Would this mean that at the very least that the 4GB variants of Vega could possibly be ITX friendly like the Nano was? I know this is something that can't be out right answered since there are no pictures of the PCB (i think that's what it's called please correct me if I'm wrong) but I figured I'de ask since Vega's architect is going to be half the size of Fiji. And if they do decide to make a ITX friendly version of the 8GB models doing a single stack of 8GB rather than two stacks of 4GB I assume that I'd be correct in assuming that thermals would be a nightmare unless they did a closed loop deal like they did with the Fury X.

3. Now this last question doesn't necessarily pertain specifically to Vega but more so all GPUs in general. What do games favor typically when it comes to Core Clock Speed versus Memory Bandwith Speed? I've heard it depends on the resolution that you are choosing to play at so for example 1080p would favor the Core Clock Speed whereas 1440p and (especially) 4K favor Memory Bandwith due to all the pixels that need to be pushed to the screen. Is this true?
 
Solution
1. my understanding is vega will give gtx 1080s non ti a run for the money not the GTX 1080Ti. The bandwidth issues are moot at this point between gddr5x and hbm because even gddr5x can increase its bus just not as wide as HBM and the clock speed makes up the difference. So both AMD and Nvidia will have similar VRAM bandwidth for this generation on their high end cards at least (gtx 1080 Ti that is the non Ti has much less) but in theory yes HBM can smoke GDDR5X when implemented properly. amd will likely have a slight overall edge in bandwidth but my guess is it will not be very large.

2. both 4GB and 8GB should be itx friendly in theory due to the fact that using HBM allows for a smaller PCB then Gddr5/5x. HBM uses an interposer to...
1. my understanding is vega will give gtx 1080s non ti a run for the money not the GTX 1080Ti. The bandwidth issues are moot at this point between gddr5x and hbm because even gddr5x can increase its bus just not as wide as HBM and the clock speed makes up the difference. So both AMD and Nvidia will have similar VRAM bandwidth for this generation on their high end cards at least (gtx 1080 Ti that is the non Ti has much less) but in theory yes HBM can smoke GDDR5X when implemented properly. amd will likely have a slight overall edge in bandwidth but my guess is it will not be very large.

2. both 4GB and 8GB should be itx friendly in theory due to the fact that using HBM allows for a smaller PCB then Gddr5/5x. HBM uses an interposer to connect to ram which is much smaller then the PCB buses GDDR uses. In theory a 4Gb and 8Gb HBM card could be the same size PCB wise.

3. the higher the resolution the more VRAM bandwidth matters, not to say core clock doesn't count either. Both need to be high for 1440p/2160P gaming. But again this generation the high end cards should have similar overall bandwidth. My understanding is AMD will hit 512GB while the GTX 1080 has much less at 320GB and the GTX 1080Ti has 480GB which as stated is close to AMD. Point being if vega is indeed meant to compete against the GTX 1080 (non ti) then you can see memory bandwidth is not everything. Plus nvidia uses some cleaver compression techniques to help mitigate the difference. Regardless the GTX 1080 Ti will be very close to vega in the memory bandwidth category and should out perform in gaming if the leaks on vega are true.
 
Solution