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

BacBeyond said:
Thats 111w more for a lower clock and score. So clearly his OC settings are causing the massive power draw, not the card's normal functions.
Click to expand...
It's not really an overclock, the guy is trying to force the chip to stay @1600MHz. He had to increase power limit, because otherwise the chip will downclock no matter what he did. There is a reason AMD stated 375w for the water cooled Vega, they know it consumes this much when pushed to it's max 1600MHz.
 
Just to be clear: This is with power limit set to +50 % in wattman? Or in default operation. OC is a realm of it's own, I'd say. Default operation with clock drops toward what seems more like the sweet spot of the chip definetly is not that high.
 
DavidGraham said:
It's not really an overclock, the guy is trying to force the chip to stay @1600MHz. He had to increase power limit, because otherwise the chip will downclock no matter what he did. There is a reason AMD stated 375w for the water cooled Vega, they know it consumes this much when pushed to it's max 1600MHz.
Click to expand...

I'm talking about the 1400 vs stock.

His 1400 clocks test score lower than stock (1440?) while using 111watts more.

That shows that his OCing is causing the voltage to be way too high for what is normally required. So don't look at the power draw numbers and think it is normal.
 
BacBeyond said:
His 1400 clocks test score lower than stock (1440?) while using 111watts more.
Click to expand...
That's just manual OC with no power limit increase, the chip throttles more, hence the lower clocks and score.
When he pushed the power limit higher, it managed to stick to the 1600MHz and scored 7273, which is ~10% faster.

EDIT: Sorry, scrap that. He clears the confusion in the video, that 6701 score was a faulty one, he was never able to replicate it again, In normal operation that card downclocks to even lower than 1400MHz and scores 6500 ish. So by forcing it @1400MHz he is increasing power draw and performance as well (to 6650 ish).
 
Last edited:
Pressure said:
Can you give some context to the video?
Click to expand...
  • Buildzoid = LN2 overclocker, primarily covers AMD hardware
  • He got the Vega FE and began testing for power/performance/OC
  • The GPU throttles heavily as the reference cooler doesn't do a good enough job cooling the GPU (like all blower style coolers)
  • When pushed to its stock clocks continuous the GPU draws 375 watts from the two 8 pins, PCIe power draw excluded, safe to say it's in the 400+ watt range
  • The GPU doesn't perform up to par currently, especially when considering the power it draws
And just to be clear, "1600" core clock with power limit and throttling is not exactly 1600 core clock.
 
Last edited:
DavidGraham said:
That's just manual OC with no power limit increase, the chip throttles more, hence the lower clocks and score.
When he pushed the power limit higher, it managed to stick to the 1600MHz and scored 7273, which is ~10% faster.

EDIT: Sorry, scrap that. He clears the confusion in the video, that 6701 score was a faulty one, he was never able to replicate it again, In normal operation that card downclocks to even lower than 1400MHz and scores 6500 ish. So by forcing it @1400MHz he is increasing power draw and performance as well (to 6650 ish).
Click to expand...
The performance at stock is 6550 ish, drawing 235W from the 8pins.

To match the power draw, he needs to set the card between 900MHz and 1000MHz. The score then is 4600 to 5000. Still drawing the same 235W.

Clearly manually setting clockspeeds incurs a power penalty for the same performance level.
 
Cat Merc said:
Clearly manually setting clockspeeds incurs a power penalty for the same performance level.
Click to expand...
And why would that happen exactly? selecting a clock speed (especially one that happens to be under the official max clocks) and sticking with it is something that any GPU should be able to do without much penalty.
 
DavidGraham said:
And why would that happen exactly? selecting a clock speed (especially one that happens to be under the official max clocks) and sticking with it is something that any GPU should be able to do without much penalty.
Click to expand...
The simple reason is that it overrides power saving techniques and artificially raises voltages. It could be as simple as Wattman applying incorrect voltages or forcing the card into the highest power state. Not all that different from the memory overclocks driving down clocks, cutting bandwidth in half.
 
DavidGraham said:
And why would that happen exactly? selecting a clock speed (especially one that happens to be under the official max clocks) and sticking with it is something that any GPU should be able to do without much penalty.
Click to expand...
I'll bet on various power control features to be working incorrectly when manually adjusting clocks.
 
duderandom84 does great youtube comparisons of graphics card and he has tamed the beast, 1600Mhz battlefield 1, the best results for Vega yet with 10% lead over 1080, rising to 20% in one area,

 
Kaotik said:
The "boardpartner cards" sounds like AIB Custom models, could it be that we actually get reference hardlaunch during SIGGRAPH and customs couple weeks later?
Click to expand...

The fact that he mentions "launchday" on week 33 makes me think that's the actual date for public availability.
I'm thinking during Siggraph they'll finally spill the beans on specs for all RX Vega SKUs, showing some performance numbers and distributing cards to reviewers. One week later the review embargoes get lifted and a week after that the cards become available to the public.

I have no idea when we'll see 3rd-party coolers though. The XTX is seemingly coming with an AiO cooler, like the Fury X. If they follow the same launch timings, the XT and XL versions may come 3-4 weeks later with a different PCB and only be available with 3rd-party coolers. A blower cooler wouldn't be a great option for 300W gaming cards.
 
Last edited by a moderator:
DavidGraham said:
And why would that happen exactly? selecting a clock speed (especially one that happens to be under the official max clocks) and sticking with it is something that any GPU should be able to do without much penalty.
Click to expand...

He says in the video, setting the power management manually disables the auto voltage of the GPU, causing it to always pull 1.2v.
 
BacBeyond said:
He says in the video, setting the power management manually disables the auto voltage of the GPU, causing it to always pull 1.2v.
Click to expand...

This seems to parallel what happens with Ryzen once a user tries to modify settings and engages overclock mode.
Perhaps this is an artifact of AMD's current version of the DVFS method being run over the control side of the Infinity Fabric?
It seems wandering past the edge of a limited envelope takes the system into the portion of the power/clock table populated by Do Not Care values.
I'm curious now how this compares to prior GPUs and AMD's fast-reacting DVFS.
 
Digidi said:
Beyond3D Suite on Vega:
http://www.pcgameshardware.de/Vega-...ase-AMD-Radeon-Frontier-Edition-1232684/3/#a1
Click to expand...


So anyone is welcome to correct me here, but as a layman I see at least two major culprits:

1 - Texel fillrate (per TMU per clock) is pretty terrible compared to Polaris and even Fiji (maybe connected to new ROPs as L2 cache clients?)
2 - Effective bandwidth is actually lower than Fiji

Something strange going on with geometry performance too, as the promised 2.6x geometry performance boost due to the new primitive shader simply isn't there. It was supposed to be 11 triangles/clock when in fact we're seeing the same 4 triangles/clock as Fiji. At 1050MHz Vega is hitting close to 4000MTriangles/s, when the slides suggested it should reach up to 11000MTri/s at ~1GHz, putting it above the Pascal cards if it was at the default clocks.



Combine this with current Vega FE clocks barely going above Polaris 20's, when slides from January suggested a 2x clock increase, and it's pretty much the perfect storm of anticipated vs. current performance.
 
Last edited by a moderator:
You must log in or register to reply here.
Back
Top