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

sebbbi said:
Is it reasonable to assume that 8 GB of HBM2 could run at slightly higher clocks than 16 GB? Aren't those 16 GB stacks pretty high? What about HBM2 power consumption?
Click to expand...

ToTTenTranz said:
Even if they overclock those 4-Hi stacks to death, if RX Vega keeps showing a performance-per-CU-per-clock that is lower than Polaris then this chip is a total clusterfuck for gaming.
Final RX Vega gaming drivers need to show at least a 20% boost in games compared to whatever the FE is running right now.
Either that or AMD will have to be selling RX Vega for $350, because at these performance and power levels the miners won't touch the thing.
Click to expand...
DavidGraham said:
RX Vega has already been revealed running 3DMarK11, it has the same memory clock as Vega FE. With slightly better gaming performance (still lower than 1080)
https://videocardz.com/70777/amd-radeon-rx-vega-3dmark11-performance
Click to expand...

¯\_(ツ)_/¯ I didn't get much of an answer before.
 
sebbbi said:
Is it reasonable to assume that 8 GB of HBM2 could run at slightly higher clocks than 16 GB? Aren't those 16 GB stacks pretty high? What about HBM2 power consumption?
Click to expand...

There may be fewer thermal concerns for the bottom die if there are four fewer powered dies in the stack above. That might help somewhat with avoiding throttling. I wouldn't know if the extra client on the channel could influence signal integrity.

I have not seen a cross section of an 8-hi HBM2 stack, but the lower elements of the stack are all die-thinned. For 4-hi HBM stacks, the topmost DRAM die is the one they don't bother to thin, and it's about as thick as the rest of the DRAM layers combined.

An 8-hi would be taller, which could pose a mechanical concern if it cannot match the GPU's height even if the difference amounts to one or two non-thinned DRAM dies.

HBM2's signalling power relative to HBM would most likely track with difference in their bit rates. HBM2's arrays would be the major DRAM-side consumers, and that may scale with their activity (scales with bandwidth and cost of array access) although pseudo-channel mode may make HBM2 more efficient with sub-page granularity accesses.
The voltages some think HBM2 is running at may be high, and the power cost of array accesses and refresh can be modified by what generation of DRAM arrays is being used and what process is being used.
I have not seen what node HBM2 is currently on.
 
Kaotik said:
AMD never said anything about how much more performance there would be with RX.
Click to expand...
Raja did say the RX would be faster, but that's generally assumed to be higher clocks from better cooling and not a huge delta. Driver performance should affect both equally. Memory pool likely not playing much of a part save for some extreme cases.

BacBeyond said:
The switch is not meant to be just the UI. It is there to swap between two sets of drivers without having to restart/reboot.
Click to expand...
Sure would make regression testing a lot easier if there was a dropdown for a year's worth of driver versions.

sebbbi said:
Is it reasonable to assume that 8 GB of HBM2 could run at slightly higher clocks than 16 GB? Aren't those 16 GB stacks pretty high? What about HBM2 power consumption?
Click to expand...
Probably reasonable and the 8-Hi's were hitting 1100MHz fairly consistently, but as mentioned above, they may perform better even at identical clocks due to less thermal issues. As for power, obviously less static energy costs refreshing the DRAM, but likely the same energy during transmission. I'd ballpark a third less power by cutting the memory in half from what's probably in the 40-50W range at 16GB.
 
Malo said:
I'm not sure where any of that came from. I think you're just making up stuff. There was certainly some basic discussion about whether or not FE and RX were different in any way other than the HBM2 stacks and speculation regarding certain advertised features working/not working but I don't think anyone here said anything about massive performance deltas between the 2 products.
Click to expand...

I think most of that discussion centered on the software side of things ala "illegitimate results; Fury drivers!; wait for the real drivers, you can't judge Vega by Vega FE, etc etc"
 
So RX Vega is first seen on Linus LTX show (what a strange way to show a product), The system has ThreadRipper and Limited Edition of RX Vega

AMD-Radoen-RX-Vega-64-LE-2-1000x750.jpg

https://videocardz.com/71325/first-amd-radeon-rx-vega-64-limited-edition-shown
 
Bound to be a weird launch with the GPU already in the wild. And whatever (driver) improvements come with rx vega will likely come to vega fe. So "faster than vega fe" is bound to be down to clocks.
 
So Vega FX won't actually be launched at Siggraph because it was launched to the press at this Linus Tech show..?


 
Interesting. Was. Yesterday. Feeling kind of oversaturated by constant trickling of µ-info. Maybe, for good measure, AMD could announce price digit-wise? My money is on a 9 as the last digit. Maybe even for second-to-last, although that might be as well a 7. And next week, we'll get to know the 3rd-to-last digit. Whether or not that's the final one will be revealed later in an upcoming exklusive LTT-show.
 
Why does hardware canuck said competition is back ? And the Linus video was embarassing. If you have nothing to show, stay quiet ffs...
Some PR ppl at AMD need to find a new job.
 
You must log in or register to reply here.
Back
Top