Amd tdp has always been conservative before Hawaii. Probably only made the exception with Hawaii because they had no other choice. Tdp is mostly there so you can tell if a cooling solution is inadequate if it falls below the tdp, amd just likes that conservative and never reach tdp in gaming and nvidia likes to have cards that draw more than the tdp at times.
AMD: Speculation, Rumors, and Discussion (Archive)
- Thread starter iMacmatician
- Start date
- Status
Anarchist4000
Veteran
My only other thought on that is if AMD is using really low voltages where that thermal inversion phenomenon was in play. Run the chips around half a volt, which would give really low power, then run up the frequencies/power with the water cooler providing a much more stable temperature. Just seems really odd a partner would design water coolers when the worst case was 150W and no indication the chips are getting near that number.Going out on a limb with completely made up hypotheticals. But if water cooling allowed for say, a 50% overclock (unreasonable and unlikely) that would likely justify a 300 USD price point and the use of a water cooler. Depending on the consumer if a water cooled solution offered 10% (or more) higher clocks than the best AIB overclocked on air card, they may find it compelling.
It still seems excessive for a mid-range card, but then overclockers are weird beasts.
Regards,
SB
I'd really have to look into the specs for the memory controllers more, but in theory it would be advantageous to develop a controller that could support them all. Likely using those GMI links they developed. Either way they are dealing with 128/256 bit data paths at significantly different speeds and voltages. The control logic could probably be a superset of both memory types or even programmable. The base die for HBM could also integrate part of the controller or possibly a minimally active interposer. Design it so 256 data pins out of the chip, which should make it easier to connect an interposer, can hook to GDDR5 or multiplex a multiple of HBM chips. 256@8GHz or 1024@2GHz. Power/latency of the HBM should be far easier to drive than the GDDR5. In theory they did some faster interconnect just to integrate GCN with Zen for their APUs beyond using PCIE.
AMD has gone to typical board power, which has a margin of error of hundreds of watts for Fury X.
TDP would be more rigorous, since it would be a physical problem if they violated it.
Polaris might be interesting in that the power connector doesn't give AMD that kind of room.
TDP would be more rigorous, since it would be a physical problem if they violated it.
Polaris might be interesting in that the power connector doesn't give AMD that kind of room.
D
Deleted member 13524
Guest
Ok people until we have like one thread for Polaris 10 and another for Polaris 11 (or maybe just one for rumors/leaks and another for post release Polaris 10+11) I'm just going to post everything in here.
RX 480M is Polaris 11 at 35W, offering that promised "Console Performance":
It's basically GM107 performance at GM108 power consumption levels.
As anandtech mentioned, AMD claims that Polaris 11 has decreased z-height for thin laptops. GM108 was used as Surface Book's discrete GPU, and I'm thinking AMD is going for that kind of product with Polaris 11.
RX 480M is Polaris 11 at 35W, offering that promised "Console Performance":
It's basically GM107 performance at GM108 power consumption levels.
As anandtech mentioned, AMD claims that Polaris 11 has decreased z-height for thin laptops. GM108 was used as Surface Book's discrete GPU, and I'm thinking AMD is going for that kind of product with Polaris 11.
D
Deleted member 13524
Guest
Here are the leaked RX 470 (Polaris 10 "Pro") benchmarks:
Credit from:
https://tweakers.net/nieuws/112519/...-van-rx-480-en-rx-470-en-onthult-rx-480m.html
Though they were smartasses by putting the graphs in javascript, so kudos to wccftech too for putting it into forum-pasteable format:
http://wccftech.com/amd-radeon-rx-polaris-10-polaris-11-specs-performance/
Credit from:
https://tweakers.net/nieuws/112519/...-van-rx-480-en-rx-470-en-onthult-rx-480m.html
Though they were smartasses by putting the graphs in javascript, so kudos to wccftech too for putting it into forum-pasteable format:
http://wccftech.com/amd-radeon-rx-polaris-10-polaris-11-specs-performance/
D
Deleted member 13524
Guest
In the end, all results are being taken from this slide, which has a lot more stuff like system specs and more detailed test configurations:
So that "280M" is actually the R9 M280X, which is a 14 CU Bonaire at 1GHz.
So that "280M" is actually the R9 M280X, which is a 14 CU Bonaire at 1GHz.
Erm.. what?
Jawed
Legend
I'd forgotten about these tests:
http://www.lesnumeriques.com/carte-...ectrique-plus-40-cartes-graphiques-a2733.html
R9 270X hits 145W there.
Interestingly it's a "middle of the pack" performance per watt card (for AMD GCN).
Anyway, I think we can be fairly sure that RX 480 has worse performance per watt.
http://www.lesnumeriques.com/carte-...ectrique-plus-40-cartes-graphiques-a2733.html
R9 270X hits 145W there.
Interestingly it's a "middle of the pack" performance per watt card (for AMD GCN).
Anyway, I think we can be fairly sure that RX 480 has worse performance per watt.
Spikes aside (which happen to break TDP on most if not all cards), Fury X's TBP is 250W and average gaming power consumption 220-230W (Tom's, TPU)
Furmark is another thing, obviously, as AMD doesn't throttle their performance/power limit there the way NVIDIA does, if they did it would fit quite nicely to TBP
I thought Pitcairn was AMDs top perf/W GPU.
D
Deleted member 13524
Guest
It says under embargo until June 13. So I guess this are not really leaks
Well I guess they're not leaks, but those are some pretty explicit benchmark numbers in that page.
I do wonder what made the enthusiast websites skip over this, though.
Spikes alone do not violate TDP if they are not for a thermally significant period.
If there is a need to waffle on TDP, it's because there are chips in that SKU or workloads that can spend non-trivial amounts of time above what marketing considers typical.
A non-serious(?) answer would be that none of them got the slides they now have permission to disseminate.
D
Deleted member 13524
Guest
And it was.. in 2012..
And likely in 2014 if not later. Is there really anything better in perf/W? I'm excluding Fiji for obvious reasons.
According to TPU Tonga pushes slightly past Pitcairn at 4K
(but they've only tested one 3xx Tonga, and no 3xx Pitcairns)
D
Deleted member 13524
Guest
Yup. Bonaire, which came in 2013.
- Status
