• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Vega Rumors

Ocellaris said:
Yes, great of AMD for making a pretty blue rectangle around a horrendously incompetent cooling design.
Click to expand...

It's curious that didn't use a DWDI fan. With a hub that thick and deep, with that RPM you could be starving it of air.

That said, if it was runing at 60C, then the cooling solution isn't inadequate. I'm not sure water would help much. I wonder if we clocked down the HBM2 it would lead to significantly more GPU headroom. That would tell us if it's the GPU or HBM2 that's the limitation.

13 Phrase VRM's is interesting design consideration.
 
So what was the point of releasing this or the RX with HBM2? Seem it is pointlessly adding cost supply issues.
 
_mockingbird said:
AMD announces Capsaicin Siggraph with no word about RX Vega

https://videocardz.com/70655/amd-announces-capsaicin-siggraph-with-no-word-about-rx-vega
Click to expand...

They have rewrote it.

Original:
On July 30, learn about our newest additions to the Radeon™ Pro family featuring the cutting-edge Radeon™ “Vega” architecture. Hear from our partners on how AMD’s product lineup – Radeon Pro, Ryzen™ Threadripper™, EPYC™ and Radeon Instinct™ – is revolutionizing the creative pipeline, leading industries from entertainment to engineering into a new era of precision and performance.
Click to expand...

New:
On July 30, learn about our newest additions to the Radeon™family featuring the cutting-edge Radeon™ “Vega” architecture. Hear from our partners on how AMD’s product lineup is revolutionizing the creative pipeline, leading industries from entertainment to engineering into a new era of precision and performance.
Click to expand...
 
harmattan said:
GCN certainly had its time with Southern Islands the most successful iteration: HD7900s, Xbox 1, PS4, and a myriad of mobile card rebrands. That said, GCN should have been put in the ground after Hawaii and is so far past being long in the tooth at this point. The fact that AMD is using essentially the same base architecture and instruction set after 7 years boggles the mind.
Click to expand...

they locked themselves in with Mantle i said it back when they announced the api
 
Simplyfun said:
hindsight is pretty much 20/20
Click to expand...

Eh, most of us predicted Fury's HBM being unneeded, costly, and potential supply issues as soon as we heard about it (months and months before it launched). That was years ago now... but they stuck with HBM! At least it's not limited to 4GB anymore.

Point being is if the assholes here could predict that either AMD is incompetent or it was a very political decision.
 
Last edited:
  • Like
Reactions: N4CR
like this
Granted this thread has been entertaining at times but even as much as I wanted to stay out of this festering boil, seems there isn't much delving and a lot of baseless bashing.

First the FE was purposed and marketed as a semi/dual-pro card. Meaning it would be capable of both pro work and gaming. Albeit nice to excel at both, it was far more likely a jack-of-all-trades-master-of-none. Hell the PCper cover of this GPU proves it. Look at the following graphs:










Notice the Nvidia cards... The TitanXp and the Quadros. In each of the graphs, one is higher than the other, denoting the difference in strengths also showing that neither is great at all. Now Notice the VegaFE is pretty much ahead of one or the other but never behind both ( gotta give me some leeway here being the P5000 is a $2000 workstation card and the P4000 is a $1000 workstation card - as stated by others so prices may be off ). This shows the FE to do just what Raja said it would do. He also stated that the FE WAS NOT INTENDED AS A GAMING CARD, although if one could not wait a month for the RX then it would obviously suffice.

Now one thing, well one of many if given the time, that I haven't seen stated is the Compute performance difference for the FE against the RX. Usually the gaming GPU has cut compute performance ie: 1/8 FP32 or such. Is that gonna be the case here as well? That alone could account for some efficiency gains in the RX.

Regardless of cut compute, the afore mentioned Cache I have seen mentioned apparently would only help with constrained bandwidth so not sure with HBM2 how often that might come into play, although the cache portion for rendered sets being kept to reduce the need for the GPU to do again might still be helpful. However the DSBR would greatly enhance efficiency to a degree enough that we could greatly see reduced Wattage.

I think for now it is too early for judgement. And being the FE is NOT solely a GAMING card, coupled with apparent lack of gaming improvement going from pro to gaming mode, now is definitely not indicative of RX, at least not conclusively.
 
do you know the major difference between pro drivers and game drivers is? Certification to work with certain applications, this costs a lot of money cause there are many applications that the drivers need to get certified for added to this all quality control for the boards and chips are done by the IHV, no AIB partner to cut costs. Vega FE doesn't have this certification process, nor does Titan XP, for nV they disable certain hardware accelerated features on non Quadro cards such as double sided polygon hardware acceleration, gen lock, etc. AMD on the other hand doesn't disable these features (ATi never did either). nV in the past actually did many of these features through bios locking, but some features since the G80 actually laser cut parts of their GPU's to stop certain features entirely like Gen-lock. That is what you are seeing in these graphs. Nothing else.

If you look at P6000 vs Titan (old) Pascal in games, P6000 actually wins out in some games in Titan, why? cause the driver base is the same.

https://hothardware.com/reviews/nvidia-quadro-p6000-and-p5000-workstation-gpu-reviews?page=6

hitman.png


time-spy.png




All applications still use DX or Ogl, so any optimizations done for the API's in games WILL translate over. Application specific optimizations on the other hand is what we will see RX Vega have over Vega FE.

And 1/8 FP 32 performance, are you talking about FP 64, cause FP 32 has never been cut lol, FP 32 is full precision and is a requirement of DX specifications so....... when either AMD or nV talk about Flops in generation its always FP32 and that is what is the main concern for games.
 
Last edited:
Dayaks said:
Eh, most of us predicted Fury's HBM being unneeded, costly, and potential supply issues as soon as we heard about it (months and months before it launched). That was years ago now... but they stuck with HBM! At least it's not limited to 4GB anymore.

Point being is if the assholes here could predict that either AMD is incompetent or it was a very political decision.
Click to expand...


I think they got locked in by exclusivity, so a little bit of politics but mostly a bad business decision by past upper management.
 
razor1 said:
do you know the major difference between pro drivers and game drivers is? Certification to work with certain applications, this costs a lot of money cause there are many applications that the drivers need to get certified for added to this all quality control for the boards and chips are done by the IHV, no AIB partner to cut costs. Vega FE doesn't have this certification process, nor does Titan XP, for nV they disable certain hardware accelerated features on non Quadro cards such as double sided polygon hardware acceleration, gen lock, etc. AMD on the other hand doesn't disable these features (ATi never did either). nV in the past actually did many of these features through bios locking, but some features since the G80 actually laser cut parts of their GPU's to stop certain features entirely like Gen-lock. That is what you are seeing in these graphs. Nothing else.

All applciations still use DX or Ogl, so any optimizations done for the API's in games WILL translate over. Application specific optimizations on the other hand is what we will see RX Vega have over Vega FE.

And 1/8 FP 32 performance, are you talking about FP 64, cause FP 32 has never been cut lol, FP 32 is full precision and is a requirement of DX specifications so....... when either AMD or nV talk about Flops in generation its always FP32 and that is what is the main concern for games.
Click to expand...
Vendor locks don't change the point at all though. Point is the FE is all around compared to the Nvidia cards in those graphs. Looking then at the FuryX you can see that the pro side is an improvement on the FE. Only the Gaming performance is off and again that should at least beg questions and the need to wait for the release of the RX line before condemning it entirely.
 
JustReason said:
Vendor locks don't change the point at all though. Point is the FE is all around compared to the Nvidia cards in those graphs. Looking then at the FuryX you can see that the pro side is an improvement on the FE. Only the Gaming performance is off and again that should at least beg questions and the need to wait for the release of the RX line before condemning it entirely.
Click to expand...


just edited my post to reflect that, might want to check p6000 benchmarks before you think pro drivers are crap in games, they are not. Pro drivers and Gaming drivers come from the same base. If you really want to get into it, every driver for GPU's in the same family (generation) of chips use the same base. GCN all of them use the same base.

And this is the problem, if we had a Titan P (new one in those graphs) it will be around the quadro 6000. Not greatly better. Titan P from the original Titan pascal of this gen, was around 10% difference right? And that is the score difference between Quadro 6000 and the old Titan.

RX Vega will end up right around FE granted they can get about 10% out of it more, that is a given for application specific optimizations, but can't really expect more than that. This is why they only showed Doom, BF, Star Wars Battlefront, those engines work better on AMD graphics hardware, and what they showed didn't impress back then cause they were around 1080 levels and that didn't change. 10% ain't going to do shit with these types of results.

Edit:

A defacto benchmark application like Futuremark, those are going to be out of the box optimized for right in the base driver, and that speaks volumes of what to expect in gaming versions of Vega.
 
Last edited:
Anarchist4000 said:
What garage? What I said looks pretty damn accurate! Just because the thread filled with hysterical Nvidia guys that haven't the faintest idea what they're looking at isn't my fault.


Not sarcastic. Vega is ahead of Pascal on the features. That is a fact if the top tier everything statement from AMD is accurate.

Register cache likely isn't enabled. In the case of Maxwell that saved 36% power with no energy loss. Vega with probably a larger cache may exceed that and even depend on that feature for throughput.

So let's knock 100W off the current consumption. Probably a bit conservative, but bare with me here. So about 1080 performance as you like to claim at 200W? Look to be ahead in perf/watt now while supporting a far superior feature set.

As Titan is the likely competitor they'll need a bit more. Tiled rasterization probably saves 20% or more bandwidth when enabled. FP16 should help with HDR, physics, and some energy efficiency. Primitive shaders save a bit more for devs who take advantage. Then of course HBCC to avoid Nvidia's slow ass software paging with upcoming games surely won't hurt. At least it looks like Maxwell and Pascal finally caught the HD7000 series and landed the bindless resources so DX12 development can finally continue.

So yeah it's a serious question if Pascal is only half-assing support for essential features 4 years after the competition. We know Vega got the features, Volta is still somewhat of a mystery. It does have the theoretical performance to go head to head, so should be interesting.


Just don't do anything. Force clocks high and don't run anything. Not too different from Nvidia when overclocking. High clocks but wasted cycles as nothing schedules to keep thermals in check.


What's the significance of that? Most cards are capable of 300W or significantly more. Enable that register cache and 225W at 1600MHz seems reasonable.
Click to expand...

I eagerly await the day I can tell you that you were right with regards to AMD being able to reduce power consumption by 33% while maintaining (or even improving) this performance
 
Anarchist4000 said:
What garage? What I said looks pretty damn accurate! Just because the thread filled with hysterical Nvidia guys that haven't the faintest idea what they're looking at isn't my fault.


Not sarcastic. Vega is ahead of Pascal on the features. That is a fact if the top tier everything statement from AMD is accurate.

Register cache likely isn't enabled. In the case of Maxwell that saved 36% power with no energy loss. Vega with probably a larger cache may exceed that and even depend on that feature for throughput.

So let's knock 100W off the current consumption. Probably a bit conservative, but bare with me here. So about 1080 performance as you like to claim at 200W? Look to be ahead in perf/watt now while supporting a far superior feature set.

As Titan is the likely competitor they'll need a bit more. Tiled rasterization probably saves 20% or more bandwidth when enabled. FP16 should help with HDR, physics, and some energy efficiency. Primitive shaders save a bit more for devs who take advantage. Then of course HBCC to avoid Nvidia's slow ass software paging with upcoming games surely won't hurt. At least it looks like Maxwell and Pascal finally caught the HD7000 series and landed the bindless resources so DX12 development can finally continue.

So yeah it's a serious question if Pascal is only half-assing support for essential features 4 years after the competition. We know Vega got the features, Volta is still somewhat of a mystery. It does have the theoretical performance to go head to head, so should be interesting.


Just don't do anything. Force clocks high and don't run anything. Not too different from Nvidia when overclocking. High clocks but wasted cycles as nothing schedules to keep thermals in check.


What's the significance of that? Most cards are capable of 300W or significantly more. Enable that register cache and 225W at 1600MHz seems reasonable.
Click to expand...
Are you for real man? Rys from RTG already confirmed the current driver is not gimped, AMD told Ryan the same thing. What kind of driver reduces power consumption by 50%? Are you high? Stop grasping at straws man. You are sounding less technical with every weird post you make. And your arguments become more ridiculous with every problem Vega faces.
 
Hameeeedo said:
Are you for real man? Rys from RTG already confirmed the current driver is not gimped, AMD told Ryan the same thing. What kind of driver reduces power consumption by 50%? Are you high? Stop grasping at straws man. You are sounding less technical with every weird post you make. And your arguments become more ridiculous with every problem Vega faces.
Click to expand...
How about a counter point? At least he gave numbers and references to what he THOUGHT could make a difference. He is right in part with some of the features and others are still questionable as far as impact. And this is that same, sorry but... , dumb shit inference of meaning when someone says gimped. None of the features for gaming listed in every slide AMD has released as far as power saving is working with the FE. Being it is a Prosumer card and shows great results in productivity benches however not so much in gaming shows that there might be something amiss with the Gaming Mode. Add to that the quote from Raja that says the RX (Gaming Vega) will perform better and brings new goodies with it (maybe those features in the slides) then it isn't out of the realm of possibility that AMD has yet released a driver that enables these aforementioned features.

See that is how one counterpoints.
 
I'll likely still buy one of the Vega gaming cards (price dependent of course - NOT at $1k).

I have three free sync monitors in eyefinity, powered by a couple Fury X now.

I just tried a 1080ti last night (bought three for mining) and had to cut my refresh rate down to 60hz (from 75hz), and with no freesync and lower refresh it just felt different for my gaming session - even though FPS was mostly capped at 60FPS. If you have freesync monitors you kind of get locked in to AMD. Just like you would for gysync with Nvidia. Lucky for AMD, freesync monitors are available 5:1 at newegg, significantly cheaper, and presumably more widely adopted.

Also helpful to my prospect for a gaming Vega is that CLC Fury X are much quieter than the reference type 1080ti under load and bonus, the video card case heat is all exhausted outside of the case. When I tried two of the 1080 ti for mining with the side panel off, the 1080ti cards throttled after just a brief while. (Top card quickly hit 85*) That never happens with the Fury X pair (they never go over 65* with the case side on!). I think if I switch back to green team for my primary card/s. I'll have to go with a seahawk or corsair hydro variant of the 1080ti. I'm spoiled by the Fury X water cooled cards for the last year. These 1080ti are for mining so I lowered power target to 65% and cut them loose. I'll see how that goes.
 
Last edited:
Archaea said:
I'll likely still buy one of the Vega gaming cards (price dependent of course - NOT at $1k).

I have three free sync monitors in eyefinity, powered by a couple Fury X now.

I just tried a 1080ti last night (bought three for mining) and had to cut my refresh rate down to 60hz (from 75hz), and with no freesync and lower refresh it just felt different for my gaming session - even though FPS was mostly capped at 60FPS. If you have freesync monitors you kind of get locked in to AMD. Just like you would for gysync with Nvidia. Lucky for AMD, freesync monitors are available 5:1 at newegg, significantly cheaper, and presumably more widely adopted.

Also helpful to my prospect for a gaming Vega is that CLC Fury X are much quieter than the reference type 1080ti under load and bonus, the video card case heat is all exhausted outside of the case. When I tried two of the 1080 ti for mining with the side panel off, the 1080ti cards throttled after just a brief while. (Top card quickly hit 85*) That never happens with the Fury X pair (they never go over 65* with the case side on!). I think if I switch back to green team for my primary card/s. I'll have to go with a seahawk or corsair hydro variant of the 1080ti. I'm spoiled by the Fury X water cooled cards for the last year. These 1080ti are for mining so I lowered power target to 65% and cut them loose. I'll see how that goes.
Click to expand...

I'll never ever do AMD liquid again due to the widespread issues they refused to even acknowledge for weeks until reviewers kept bashing them (and buying retail samples). You must have been one of the lucky ones or bought the Fury X later. I had both the bad fan and bad pump. Fan I just replaced, pump was out of warranty.
 
I want to see this miraculous driver which will bring power consumption down 50-100W and increase performance by 40% because AMD would have to be the dumbest company on the planet to be sitting on a driver like that while their Fail Edition racks up all the negative press for Vega. We might at best see a 10% gain in drivers and maybe a few more percent from AIB cooled products where the card can hold 1600 MHz. This should maybe enough to put it on par with 1080 but will still fall far short of 1080 Ti while ignoring the fact that Volta is coming.
 
I get the argument that people may still want to look at Vega for Freesync. I have a Freesync monitor and did see its value when I had my dual Furys. That said, after going back-forth between AMD and nV over the past year, I've decided I'd much rather have higher FPS with vsync or adaptive vsync rather than deal with lower performance and high heat/power -- AMD can't do that now, nor does it seem like they'll be able to with Vega.
 
5150Joker said:
I want to see this miraculous driver which will bring power consumption down 50-100W and increase performance by 40% because AMD would have to be the dumbest company on the planet to be sitting on a driver like that while their Fail Edition racks up all the negative press for Vega. We might at best see a 10% gain in drivers and maybe a few more percent from AIB cooled products where the card can hold 1600 MHz. This should maybe enough to put it on par with 1080 but will still fall far short of 1080 Ti while ignoring the fact that Volta is coming.
Click to expand...

Why do we have to counter this argument every single release it seems? I remember people crying that there would be world-saving driver improvements going back to 1800 XT. It has never happened.

We might see some general improvements from what's been shown, but on the order of 5-10% and that won't materialize in a single driver version but over time.
 
harmattan said:
Why do we have to counter this argument every single release it seems? I remember people crying that there would be world-saving driver improvements going back to 1800 XT. It has never happened.

We might see some general improvements from what's been shown, but on the order of 5-10% and that won't materialize in a single driver version but over time.
Click to expand...

I absolutely agree with you.

However, we still have not seen RX Vega, this card is interesting as hell, but it's not their gaming card. I don't even care how close the results are between this prosumer card and RX end up being.

I want to see their thrice damned RX Vega and it's price.
 
Dayaks said:
I'll never ever do AMD liquid again due to the widespread issues they refused to even acknowledge for weeks until reviewers kept bashing them (and buying retail samples). You must have been one of the lucky ones or bought the Fury X later. I had both the bad fan and bad pump. Fan I just replaced, pump was out of warranty.
Click to expand...

yeah I bought mine a year in or so (when the XFX Fury X dropped in price at newegg to like $350) and they have been perfect. I'd assume they've learned a lesson and probably the watercooled gaming Vega will have no pump/coil whine issues. Sorry your experience was poor. Mine has been stellar.
 
Simplyfun said:
I absolutely agree with you.

However, we still have not seen RX Vega, this card is interesting as hell, but it's not their gaming card. I don't even care how close the results are between this prosumer card and RX end up being.

I want to see their thrice damned RX Vega and it's price.
Click to expand...

Alright let's just deal with this before we all fall into a mass neurosis.

It makes no sense whatsoever for AMD to release a product that performs worse than it could by choice.

If performance is lackluster, then lackluster performance is the best they could do.

This doesn't seem like far fetched logic to me...

Has there been any precedent for *sweeping * performance improvements across all tested games/benchmarks?

Has there ever been a precedent for massive power reduction with a driver update without compromising performance?

If Vega FE is for developers to start developing with Vega's new tech in mind then what sense is there in not having it enabled?

Honestly the vast majority of justifications for the performance I've read are frankly incoherent

Many people are claiming tiled rasterization is not active, yet there are interesting discussions on B3D that consider the existence of a fallback rasterization mode kicking in under certain circumstances. Worth a read
 
Last edited:
JustReason said:
How about a counter point? At least he gave numbers and references to what he THOUGHT could make a difference. He is right in part with some of the features and others are still questionable as far as impact. And this is that same, sorry but... , dumb shit inference of meaning when someone says gimped. None of the features for gaming listed in every slide AMD has released as far as power saving is working with the FE. Being it is a Prosumer card and shows great results in productivity benches however not so much in gaming shows that there might be something amiss with the Gaming Mode. Add to that the quote from Raja that says the RX (Gaming Vega) will perform better and brings new goodies with it (maybe those features in the slides) then it isn't out of the realm of possibility that AMD has yet released a driver that enables these aforementioned features.

See that is how one counterpoints.
Click to expand...
Oh come on, the guy is hopelessly lost in his AMD bias that he is willing to twist technical facts in the face of AMD's massive failure to better suit his mental image of AMD.

AMD lied and continues to lie to cover up this mess, they lied when they said the card only needs two 8 pin connectors to give it better headroom, they lied when they said Vega FE delivers smooth 4K performance, they lied when they previewed Vega FE and claimed it's gaming performance is the same as TITANXP, they lied about not needing drivers to achieve the claimed rate of 11 triangles per clock. They lied about showing Vega in Computex, and they kept hiding their fps in multiple game demos. All of these actions constitute an ugly picture of AMD trying hard to sugarcoat a bad situation. It's obvious to anybody with a brain, we don't need ridiculous halfassed theories to justify it all.
 
JustReason said:
How about a counter point? At least he gave numbers and references to what he THOUGHT could make a difference. He is right in part with some of the features and others are still questionable as far as impact. And this is that same, sorry but... , dumb shit inference of meaning when someone says gimped. None of the features for gaming listed in every slide AMD has released as far as power saving is working with the FE. Being it is a Prosumer card and shows great results in productivity benches however not so much in gaming shows that there might be something amiss with the Gaming Mode. Add to that the quote from Raja that says the RX (Gaming Vega) will perform better and brings new goodies with it (maybe those features in the slides) then it isn't out of the realm of possibility that AMD has yet released a driver that enables these aforementioned features.

See that is how one counterpoints.
Click to expand...

dude no couter point to his BS he didn't know pascal was tier 3 that was his argument lol

Sorry vega is disappointing for folks but that is reality.

Doesn't everyone remeber FX excuses or the r600 excuses?
 
Last edited:
Hameeeedo said:
Oh come on, the guy is hopelessly lost in his AMD bias that he is willing to twist technical facts in the face of AMD's massive failure to better suit his mental image of AMD.

AMD lied and continues to lie to cover up this mess, they lied when they said the card only needs two 8 pin connectors to give it better headroom, they lied when they said Vega FE delivers smooth 4K performance, they lied when they previewed Vega FE and claimed it's gaming performance is the same as TITANXP, they lied about not needing drivers to achieve the claimed rate of 11 triangles per clock. They lied about showing Vega in Computex, and they kept hiding their fps in multiple game demos. All of these actions constitute an ugly picture of AMD trying hard to sugarcoat a bad situation. It's obvious to anybody with a brain, we don't need ridiculous halfassed theories to justify it all.
Click to expand...

He should go post in AnandTech forums.

Over there, they make every excuses for AMD.

If one post anything negative about AMD:

BOOM! Instant BAN!
 
razor1 said:
AMD has there teeth in many forums not just Anandtech. Its the mods and management of the forums

And it should be looked into! At least here that won't happen.
Click to expand...

I don't think they work for AMD, but they let their personal bias get ahead of the rules.

I am glad that hardforum (at least the moderators) is neutral.
 
Dayaks said:
I'll never ever do AMD liquid again due to the widespread issues they refused to even acknowledge for weeks until reviewers kept bashing them (and buying retail samples). You must have been one of the lucky ones or bought the Fury X later. I had both the bad fan and bad pump. Fan I just replaced, pump was out of warranty.
Click to expand...

Fair enough, but My 295x2 ran amazingly, one of the best cards I've ever had minus Crossfire issues.
 
Hameeeedo said:
Are you for real man? Rys from RTG already confirmed the current driver is not gimped, AMD told Ryan the same thing. What kind of driver reduces power consumption by 50%? Are you high? Stop grasping at straws man. You are sounding less technical with every weird post you make. And your arguments become more ridiculous with every problem Vega faces.
Click to expand...
I think you need to understand what gimped means. If the software wasn't ready, then current drivers aren't gimped. I provided numbers based on what occurred when Nvidia did something similar. I'm not sure why you're dismissive of something that occurred previously with Nvidia cards.

JustReason said:
Add to that the quote from Raja that says the RX (Gaming Vega) will perform better and brings new goodies with it (maybe those features in the slides) then it isn't out of the realm of possibility that AMD has yet released a driver that enables these aforementioned features.
Click to expand...
Goodies are probably game bundles. It took Nvidia a while to get their drivers going when they did something similar and remains to be seen how soon AMD will be ready.

Ieldra said:
Has there been any precedent for *sweeping * performance improvements across all tested games/benchmarks?
Click to expand...
Yes, Nvidia did it with Maxwell just a few years ago. Everyone seems to think their architecture is amazing because of it. CPUs are similar, and a register file cache is their basis for pipelining data to increase IPC.

http://dl.acm.org/citation.cfm?id=2155675

Search the title and you can find the paper. The citation there states 34-54% reduction from the registers. Registers will be a significant part ot power.

The more important part I feel is the 40-70% reduction in register file accesses that we're measured. That's a fairly universal stat based on a access patterns that won't change. Upon calculating a value it gets used in short order. That reduction also means 40-70% of the time the primary bottleneck in a processor is idle. So the design could be counting on that added bandwidth.

That one feature took Nvidia a while to get right. AMD is still hammering away from appearances. That alone could nearly normalize performance between AMD and Nvidia. Without it Vega would behave like Fiji. With it power goes down with no performance hit while opening the door to getting a lot more done as the registers are freed for other tasks.

Ieldra said:
Many people are claiming tiled rasterization is not active, yet there are interesting discussions on B3D that consider the existence of a fallback rasterization mode kicking in under certain circumstances. Worth a read
Click to expand...
There is always a fallback, but so far it looks like Fiji. The real problem is that in games there appears to be a no difference between Fiji and Vega beyond clocks. Hence the features likely aren't enabled.
 
Anarchist4000 said:
I think you need to understand what gimped means. If the software wasn't ready, then current drivers aren't gimped. I provided numbers based on what occurred when Nvidia did something similar. I'm not sure why you're dismissive of something that occurred previously with Nvidia cards.


Goodies are probably game bundles. It took Nvidia a while to get their drivers going when they did something similar and remains to be seen how soon AMD will be ready.


Yes, Nvidia did it with Maxwell just a few years ago. Everyone seems to think their architecture is amazing because of it. CPUs are similar, and a register file cache is their basis for pipelining data to increase IPC.

http://dl.acm.org/citation.cfm?id=2155675

Search the title and you can find the paper. The citation there states 34-54% reduction from the registers. Registers will be a significant part ot power.

The more important part I feel is the 40-70% reduction in register file accesses that we're measured. That's a fairly universal stat based on a access patterns that won't change. Upon calculating a value it gets used in short order. That reduction also means 40-70% of the time the primary bottleneck in a processor is idle. So the design could be counting on that added bandwidth.

That one feature took Nvidia a while to get right. AMD is still hammering away from appearances. That alone could nearly normalize performance between AMD and Nvidia. Without it Vega would behave like Fiji. With it power goes down with no performance hit while opening the door to getting a lot more done as the registers are freed for other tasks.


There is always a fallback, but so far it looks like Fiji. The real problem is that in games there appears to be a no difference between Fiji and Vega beyond clocks. Hence the features likely aren't enabled.
Click to expand...

I'm really hesitant on trusting something from AMD that requires "enabling". Track record sucks there.

Why would they release it like this and tarnish all of Vega?
 
tviceman said:
Yep. Two days ago I said Vega was a heaping pile of crap and got temp banned.
Click to expand...

It's the report squad in action. That place is crammed with Red Team Plus members (aka shills) disguised as regular forum users. I'm subject to constant threats from the staff, and personal attacks against me never get infracted. BTW there was nothing wrong with your post.
 
Last edited:
Dayaks said:
I'm really hesitant on trusting something from AMD that requires "enabling". Track record sucks there.

Why would they release it like this and tarnish all of Vega?
Click to expand...

Realistically, if they thought Vega FE dramatically underrepresented RX Vega in gaming, they would not have released it broadly to consumers like this.

I think it's time to face reality: RX Vega is expensive to build, hot, way late, and simply not fast enough.

Better luck next time, AMD.
 
-Sweeper_ said:
It's the report squad in action. That place is crammed with Red Team Plus members (aka shills) disguised as regular forum users. I'm subject to constant threats from the staff, and personal attacks against me never get infracted. BTW there was nothing wrong with your post.
Click to expand...

Some of those shills are on this forum under different names. Look through the AMD threads and the posting style and you'll see who they are.
 
You must log in or register to reply here.
Back
Top