Undervolting Vega

hardwareluxx.de/index.php/artikel/hardware/grafikkarten/44084-amd-radeon-rx-vega-56-und-vega-64-im-undervolting-test.html
Vega 56 coming close to 1070 power draw with
near 1080 performance.

Sorry OP, I don't speak nazi.

Gee, if only someone could have predicted that Vega would perform much better with undervolting, and that an undervolted Vega 56 at the same clocks as Vega 64 would perform 100% the same since both cards remain front end bottlenecked until primitive shaders are implemented in driver.

>$500+
>hot
>14 months late

what else is new
Overclocking memory and tweaking timings on Polaris could get you ~10% more performance. When you underclock and undervolt to get the same performance as stock you can make it use like 20-40% less power with the same exact performance as the stock card.

Why is AMD retarded like that and not tweaking shit in factory is beyond me.

>Why is AMD retarded like that and not tweaking shit in factory is beyond me.

It's a yield issue, the voltage target they set, the more good chips they can get.

>Why is AMD retarded like that and not tweaking shit in factory is beyond me.
It's called the Nano.

The rest are all mass market products.

>hot
Not anymore :^)

>Why is AMD retarded like that and not tweaking shit in factory is beyond me.
because it's silicon lottery. You can't get all the chips to be mantain the target clock at that voltage. The same reason not all 7700K can reach 5Ghz at somehow acceptable volts.

/Thread

Even if it matches (18 month old, lol) Pascal performance, mining prices will keep it out of the competition until Volta comes out. This shit is just too little, too late

>Vega64
>same performance as 1080
>10w more than a 1080ti when undervolted, or 90w more than a 1080
Yrev very goodsame power draw

How is this different for any GCN chip? It's literally the same thing has been doing for 6 years, AMD cards are diamond in the rough, Nvidia since Maxwell has optimized their manufacturing and binning to the max at the expense of yields, so they're better out of the box.

Because everyone buys their hardcore gaming gear the day it's released, right?

If they just find some way to magically unfuck the memory bottleneck the whole Vega line will be saved. The Pascal family is really mature, we're already seeing all Nvidia can squeeze out of it with driver optimization. AMD however is can wring a lot more out of Vega.
The Vega64 might end up only 15% slower than the 1080ti which isn't terrible by any stretch.

Vega 64 on current drivers is massively front end bottlenecked by only having a polygon thoughput of ~4 triangles per clock. Once primitive shaders are enabled in drivers, Vega 64 will no longer be front end bottlenecked and its perf/watt will be competitive.

Vega 56 comparatively suffers less from the current 4 triangles per clock bottleneck than Vega 64 does, so it shows undervolted perf/watt that beats a 1070 FE already.

HEY GUYS LET'S ALL PRETEND TO BE SURPRISED AND ASTONISHED THAT AMD IS REPEATING THE SAME MANUFACTURING JAZZ THEY'VE BEEN DOING SINCE GCN1 LAUNCHED EVERY SINGLE YEAR WITHOUT FAIL


HOW CAN THIS BEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE!

I'm hopeful about vega 11 and low-cost hbm. with drivers ready, it can perform 1070 level. I hope q42017 or q1 2018 will be distant enough time to ready the drivers.

Why does the undervolted cards perform better than stock? That doesn't make any sense.

Concern trolls are noted, suddenly it's a good mining card when overvolting comes in

Oh my gosh who would have thought.

Lower voltage, thereby generating less heat, allows the card to sustain higher boost clocks as it isn't hitting thermal/power limits.

The era of fixed clock rates is over. We dynamic clock now. CPUs and GPUs will clock to what they can hit.

Because it's not throttling as much.

It only doesn't make sense since you're stupid.
It's called power throttling.

it doesn't get hot, so it doesn't throttle and keeps the clock steady.

Sorry, i don't have time to fuck around with overclocking and shit, I'm a busy man with a life, if AMD can't provide that experience out of the box then they're dead to me.

AMD made a card just for you fags that are too busy to pull a slider back but not to busy to shitpost on Sup Forums

There's no memory bottleneck.
The front-end is chocking, maggots shipped gaymen Vega with NGG path disabled.

Basically the yields are so poor they have to use higher stock voltages to ensure all the chips function properly. This means they generate more heat.

However, the better chips can run fine at a lower voltage, so you get less heat an better overclock/boost performance.

This is the silicone lottery.

this desu. Reviewers have gotten the better silicone as usual and undervolting so they look more attractive to miners.

A non-throttling Vega FE is already like 20% slower. They should seriously be making up that difference with drivers

You don't know the yields, AMD did this crap for tiny dies like Polaris 11 and Cape Verde, and pretty much any GCN GPU.
I can assure you those don't have bad yields or they wouldn't be sold.

AMD overvolts simply because it's cheaper to them and they get more units per wafer.

Performance increasing by significant amounts when memory clocks is increased = memory bottleneck.
There is no getting around this.

The theoretically deliverable bandwidth is far off from what measured bandwidth shows. Fiji had the same issue.

Should I undervolt my 390x?

Do you want to? Do you need lower power draw? Do you already have a OC on it?

It only increases for 56, which is literally segmented by memory BW.

Can't wait for all the retards who think this applies to all chips and cards start crashing unable to reach the voltage target at stock clock.

>Basically the yields are so poor they have to use higher stock voltages to ensure all the chips function properly. This means they generate more heat.

Or they just pumped the freqs to max to squeeze out all the possible performance out of the chip because otherwise it wouldn't be able to compete with Nvidia. This results in higher than anticipated power draw.

Polaris had the same issue. Heck, Zen has the same issue too (it was designed for lower clocks), except that Zen performs well enough anyway so it's not a problem other than being shit at overclocking.

>You don't know the yields
You can speculate the yields from the voltages they ship at.
While we don't know the actual yields if they had good yields they would literally toss the chips that required a higher voltage in the bin.
Seeing they use a higher than optimum voltage on the cards it seems likely they are shipping the less desirable chips as well, which they wouldn't do if they could supply enough good chips.

>Or they just pumped the freqs to max to squeeze out all the possible performance out of the chip because otherwise it wouldn't be able to compete with Nvidia. This results in higher than anticipated power draw.
Na, then you wouldn't be able to undervolt effectively except on like, the golden silicone.

You got a sample size of 2 for undervolting currently, we'll know the average soon enough.

Same motherfuckers who want everyone to buy a R5 1600 although most samples can't quite OC to 1600x levels.

How many FPS is that 100MHz?

AYYMD HOUSEFIRES

How do I enable dynamic clock on my cards?

And your cards are?

If its new enough to have the feature its already enabled.

FUCK FUCK FUCK FUCK FUCK FUCK FUCK FUCK FUCK FUCK FUCK FUCK FUCK

MINERS ARE GONNA EAT THEM ALL

No.
580 or 1070 are still better at mining.

there will not be any more vega user there are only 3
56 64 and that nano just like fiji

People are getting 45MH/s on a undervolted FE with alpha blockchain drivers, with no client support, Polaris and Pascal are dead.

Are you retarded?
Vega is full lineup, including smaller (Vega11), the smallest (Vega12), and big dick FP64 (Vega20).
Fiji was double Tonga.
Tonga existed before Fiji.

in gcn undervolting means higher clocks with less heat and less power draw

as i said
THERE WONT BE ANY MORE VEGA on this lineup

the full lineup is expected to come on the vega refresh NOT NOW because amd is building up stock binned chips for them

45mh/s at what 200w?
What are people getting with Polaris cards? 30mh/s for 150-180w~?

Vega will undoubtedly grow into the mining monster it was rumored to be. They should be able to hit 60mh/s given the amount of CUs and its clocks combined with an actual instruction set for it.

It means that everywhere if the clock/TBP/voltage is skewed too much in one field (which would be voltage for AMD)

gibbo is a fucking liar dont fucking repost his shit about 100mh/s

Vega refresh is who knows when and I doubt they'll refresh Vega10.

60 to 70 should be entirely within the realm of possibility.
36CU at 1200mhz~ is pulling 30mh/s.
Vega is 64CU at 1500mhz~

>Vega 56 coming close to 1070 power draw with near 1080 performance.
Guess those miners are gonna be really happy

around late q4 17 with q1 18 cause navi needs to be on track at q4 18 to q2 19

people dont realise that amd has to feed apple ms sony their new opening on the rendering farms their workstation cards
all of them needs high binned chips consumers are stucked with low bines hence the stupid overvoltage out of factory

ETH is bandwith and memory latency.
Not compute.

62 is the max i believe that it can achieve if you have a good hbm2 stack we will soon find out

Hashing is limited by bandwidth just like all compute. Each block of w/e is x MB so you can only work on as much as the effective bandwidth can fulfill. You can optimize for it but I doubt it'll go over 50 considering the out of box performance is junk

Nah it's stock oc

I'll try ocing it once I get my new build

1070

I use EVGA Zog, not sure if its dynamic its more like a log curve

EVGA use a log curve unfortunately

Until the front end bottleneck gets fixed with primitive shaders actually being enabled in drivers, it is impossible to accurately assess how memory bandwidth botttlenecked Vega is or isn't. Wait until after prim shaders are enabled to try and assess how much of a memory bottleneck there is.

>14 months late
that would be a great argument if I wanted to buy a graphics card 14 months ago, but if I want to buy one now what benefit is there to buying a graphics card that's 14 months older than the other?

DSBR saves a LOT of bandwith though.
I doubt there's a BW bottleneck in Vega.
Vega is fairly balanced assuming everything works as intended.

It's proven. It's cheaper. It's better. No Waiting™ for usable drivers.

Yeah, I understand what you are saying, but you can't test to confirm that until prim shaders are working.

AMD FAGS ON FULL DAMAGE CONTROL MODE WHEN VEGA IS CONFIRMED SHIT

:^)

this desu

>It's proven.
ye olde graphics card isn't much of a selling point.

>It's cheaper.
really?

>It's better.
by what metric?

>No Waiting™ for usable drivers.
drivers are already usable and since they're early there's still potential in contrast to ye olde drivers.

>Vega 56 better in an Nvidia Goyworks game than the GTX 1080
Thats impressive.

Whatever, it's your money

PSA: It's spelt 'silicon'. Unless you want a breast enlargement.

>Vega beating nVidia cards in goyworks gayme
niiice

About tree fiddy

Isn't goyworks all retarded amounts of geometry?
Actually working Vega will chew through it nicely due to hilarious triangle throughtput.

>Vega barely about equal to a 1080 in a known game after 3 days of setting hacking on the GPU

So... Is Vega secretly amazing?
> Undervolting gives significantly better performance and power consumption compared to stock
> Lots of missing features in the drivers that will result in significant performance improvements over time as they get added in

Too bad it costs $200 more than a 1070.

No. Vega doesn't even have the geometry throughout boosting primitive shader enabled yet. Nvidia goyworks is mostly physics simulations

The 56 msrp is 400 USD, $50 more than the 1070 msrp but cheaper on practice due to gpu price increases

>secretly
You can't read the damn whitepaper?

You are in for a whole world of butthurt

what? no. :^)

And since high end volta is still 8-12 months away I have to buy the 1080ti as a stopgap. At least I will be able to resell the thing and recoup a few hundred bucks.

The thing I hate the most tho is that it means I will have to buy a high refresh Gsync monitor instead of a freesync one like I had hoped to do.

BUT
WHY?

Its definitely not as bad as shill are trying to make it seem. Even with a rocky launch Vega56 looked like a clear winner for a high end card. Really the only issue is that Vega 64 isn't going head to head with the 1080ti.

What difference does msrp make when the product is sold at major markups by vendors??

You cannot software patch a hardware issue.

They may fix it in navi, but not before then.

Literally Nvidia payed them to.

It can't go head to head with GP102 because front-end currently chokes.
It'll be fixed in a driver update or two.

>incompetent devs

wow

Vega 11 will be targeting the 1050 performance wise.

Bold claim. Care to back it up?

nope, it is to replace rx580/570.

Because Vega 56 will be available at msrp for 10 minutes when it releases on the 28th