Poll: The Vega Review Thread.

[LoadsaMoney]

 

[Brazo]

Polaris is murdering the 1060 cards. If the design scaled and/or GloFlo offered the clockspeeds of TSMC Polaris would have looked great.

I think murder is a strong word but that's your opinion of course without any facts to back up.

A simple Google would show you a 1060 vs 480 after driver updated showing identical (across the board) performance.

http://blog.logicalincrements.com/2017/01/gtx-1060-vs-rx-480-graphics-card-better-driver-update/

Changing the box label from 480 to 580 doesn't add much either!

 

[JediFragger]

Before it was attrocious! Now it's just fairly poor against the 1080 (which is just an FE!)

 

[JediFragger]

Changing the box label from 480 to 580 doesn't add much either!

Only a re-review after the driver guy has had 9-10 months to improve them

 

[Brazo]

Before it was attrocious! Now it's just fairly poor against the 1080 (which is just an FE!)

Vega 64 has a long way to go to catch a 1080 but I expect it will, much in the same way as the 480 caught the 1060.

 

[JediFragger]

Vega 64 has a long way to go to catch a 1080 but I expect it will, much in the same way as the 480 caught the 1060.

Yup, who needs performance today when you can have AMD FineWine technology instead to make it feel like you got a great long-term deal

 

[jigger]

Yup, who needs performance today when you can have AMD FineWine technology instead to make it feel like you got a great long-term deal

Better than stale bitter?

 

[Chuk_Chuk]

I'll admit I was only thinking of the card from a gaming point of view. I'd imagine this scenario means it's harder to fetch the data before it's needed as it's harder to predict what will be needed (I'd imagine the really immediate stuff will be in the onboard VRAM?)
From that point of view not sure what good being able to render a 50GB scene is unless it can do it in 1/60th of a second or faster.

You are correct, for 3D rendering its not possible. At best you can consider all the textures in the immediate view of the camera and keep them all in VRAM and leave the rest on RAM (or the HDD).
I have yet to see threadripper/EPYC comparisons to GPU rendering, so ignoring those, GPUs are just so much faster than CPUs (depending on scene complexity the speed difference can vary wildly but GPUs can be 2-3x times faster than CPU even more in some cases), the biggest problem is the VRAM (for the user). So yes there is a penalty involved in using HBCC but if the render on the GPU is still faster than on the CPU then its worth using the GPU.

I'm not sure how much of a penalty we get from accessing system RAM (assuming that a given usage is contained on system RAM only), It would be interesting to see a comparison between the FE and 64 with a 12GB scene, to see how much it slows down. However considering that PCIE 3, is really fast i doubt it will be as terrible as we see when using scratch disks.

Hope that makes sense

 

[shankly1985]

Before it was attrocious! Now it's just fairly poor against the 1080 (which is just an FE!)

Vega 64 has a long way to go to catch a 1080 but I expect it will, much in the same way as the 480 caught the 1060.

Tbh guys Amd has always been behind nvidia on unreal engine titles. After the driver update its much better than it was.

Fact with this title though nobody runs ultra settings lol I play with everything on very low apart from shadow medium, draw distance and texture ultra..

All about seeing people. I asked him the other week can he do a competitive settings benchmark.

 

[five8five]

^ Was hilarious that

Yeah, Gibbo had to step in before the insanity got too much (Titan Killer)

 

[Brazo]

Yup, who needs performance today when you can have AMD FineWine technology instead to make it feel like you got a great long-term deal

When AMD themselves were describing vega as 'around' 1080 I knew it was more smoke and mirrors.

 

[Gerard]

Was hugely over-hyped by a sizeable section of AMD fans based on early graphs showing a large performance per watt increase from the previous generation. Most of these people expected it to be between 980 and 980 Ti/1070. In the end it only managed to match the 970 with higher power consumption.

I remember some AMD die-hards having meltdowns in the review thread.

In retrospect the card was not bad at all. It was just a case of AMD fans over-hyping their team to oblivion as usual.

Polaris wasn't bad but if Kyle from hardocp is to be believed it didn't turn out even close to how amd envisioned it, and therefore had to market it as a midrange card instead of what it was intended to be.

 

[five8five]

Polaris wasn't bad but if Kyle from hardocp is to be believed it didn't turn out even close to how amd envisioned it, and therefore had to market it as a midrange card instead of what it was intended to be.

It sounds like there may be a common theme!

 

[jigger]

When AMD themselves were describing vega as 'around' 1080 I knew it was more smoke and mirrors.

But TBF around 1080 performance can be taken as sli Titan killer. The focus group seem to get very nervous nowadays.

 

[Gerard]

It sounds like there may be a common theme!

The common theme is global foundries. There's a lot of talk that tsmc could potentially have had both vega and polaris have less leakage and higher clocks. But amd used global foundries for whatever reason.

 

[jigger]

TSMC couldn't deliver and everyone come second to Apple. They also failed to move to from the 28nm designe and that probably didn't suit AMD to well and AMD probably didn't have the money to pay for some of the **** ups TSMC made.

 

[AlamoX]

The common theme is global foundries. There's a lot of talk that tsmc could potentially have had both vega and polaris have less leakage and higher clocks. But amd used global foundries for whatever reason.

global foundries used to belong to AMD,when they were short on cash, they sold it off, and they had an agreement in place forcing AMD to manufacture a specific amount of chips at global foundry, not sure how much maybe about 1bil $ worth of chips, and if they don't they have to pay a fine, like they did back in 2014 i think about 300mil.
but iv read somewhere that AMD & glofo got a modified the agreement, and now AMD can switch to another foundry if glofo doesn't deliver on what AMD needs without paying the fine, that's good news i guess, but limited to some products, and GPUs doesn't seem to be part of it, so Navi is going to be glofo.
there is an interesting article here about the new deal, source : Hardware.fr

 

[Gerard]

global foundries used to belong to AMD,when they were short on cash, they sold it off, and they had an agreement in place forcing AMD to manufacture a specific amount of chips at global foundry, not sure how much maybe about 1bil $ worth of chips, and if they don't they have to pay a fine, like they did back in 2014 i think about 300mil.
but iv read somewhere that AMD now can switch to another foundry if glofo doesn't deliver on what AMD eeds without paying the fine, that's good news i guess, because glofo's 14nm sucked, and not suited for GPUs.

Yeah i remember their promo vids for it when it was starting to get constructed a lot of years back. They're essentially being strong-armed into using a foundry with a more inefficient process than the altenatives.

 

[ubersonic]

Vega 64 has a long way to go to catch a 1080 but I expect it will

Asuuming you mean 1080ti (as it already matches the 1080 in games and rapes it in applications) I'm not so sure, it will prob settle between 1080/1080ti performance IMO. I'm not expecting it to jump multiple tiers like the 7970 did lol.

 

[jigger]

Well I'm not sure sucked is fair. Some very nice cards have been produced at Global and the CPU's getting produced are some of the best we have seen in the last 10 years.