[News] AMD Radeon Vega Frontier for 1,199 USD [Update: Benchmark]

[SinOfLiberty]

Some dum comments from "pro marketers" in here is absurd.

[chew*]

Some dum comments from "pro marketers" in here is absurd.

I am not a pro marketer......I just know that 1 vega will probably perform as good as my 2 290x ( 4g models so there is part of the math in my thinking ) and run a tad cooler.

I'm not trying to break WR with my daily and I know it will not perform @ the top level. I just prefer ATI/AMD graphics.

[SinOfLiberty]

Kingpin Ti was able to get 2400mhz, yes, power moded for now, but still a great result. Although, I dare say, BIOS was approved by NV to remove speed restrictions.

NV readies an answer and in such form, an unbeatable one.

bits n chips is full of Sh1t , dont believe them.

[zanzabar]

Kingpin Ti was able to get 2400mhz, yes, power moded for now, but still a great result. Although, I dare say, BIOS was approved by NV to remove speed restrictions.

NV readies an answer and in such form, an unbeatable one.

bits n chips is full of Sh1t , dont believe them.

when they interviewed vince there was something about using a non gpu controlled vrm controller so they could control the vrm outside of the bios.

[SinOfLiberty]

vrm is no biggie.

NV sets up a custom .rom that restricts the speed flow parameters. It is very complex.

They can unlock it when nda is up

TLDR: No more than 2100 mhz.

[kromosto]

I am not satisfied of amd because I want to buy an AMD

[zanzabar]

vrm is no biggie.

NV sets up a custom .rom that restricts the speed flow parameters. It is very complex.

They can unlock it when nda is up

TLDR: No more than 2100 mhz.

so they literally restrict clock speed and not a set of voltages? that seams like a really move and also shouldnt that make every 3d mark run using an unlocked card illegal since engineering samples and vendor supplied custom parts are not allowed since you could not mod to do that?

[SinOfLiberty]

Truth:

Early on, if u interested, OC`rs flashed 980 ti bioses on pascal and power moded `em.

Nowadays, U need to have a pro oc licence to get a proper BIOS. NV does now want anyone tinkering with pascals before the NDA. Imagine 2.4ghz 70/80/ti.. Would of killed the business of high end gpus.

No one would ve bought Ti if They could push such freq. out of their gpus, cuz the performance would skyrocket.

Now, that all Pascals are out, NV is ready to destroy AMD as they are dead set on Ti performer for ~<600usd.

The NDA is almost up and the long wait will be over!

[madmaxx]

I thought this thread was about Vega? At least people were whining about it's "performance", now we're talking about Nvidia sh*t?

Best I can tell is it's being limited right now for some reason, things don't make sense otherwise. A dangerous game you are playing AMD but if you pull it off, big balls.

[SinOfLiberty]

del.

[BeepBeep2]

It's not that much different from Fiji you know.

Also most of the "old drivers" "info" is bs and wishful thinking.

TBR isn't working properly, now that can be anything from not ready yet or AMD failing to get it to work properly.

Either way, with every feature-set "activated" and "optimized" drivers, and perhaps increased clock speeds, it can potentially get 20-15% close to the 1080 Ti on average.

It wouldn't surprise me if it is completely broken and requires a new stepping or even Navi before they get it fixed.

First Ryzen CPUs back from GloFo were DOA according to a story published by Forbes and they were scrambling to get it working for initial testing:
http://fortune.com/2017/06/28/amd-ai-chips-comeback/

Vega seems like a small change after reading GamersNexus' clock-per-clock review today, but the internal interconnect design with Infinity Fabric is also brand new, and the geometry engines are supposed to be able to process 11 polys / clk instead of 4 polys / clk. I would have to guess by the almost identical performance compared to Fiji that what ever is new in the geometry engines isn't functioning as planned either.

It was noted back in January that Vega was essentially running in "Fiji" mode, and nobody seemed to panic then, but it seems nothing has changed / been fixed since then either.

The latest Linux AMDGPU source code w/ Vega support doesn't even have the rasterization section written yet and Fiji and Polaris already use a relaxed out-of-order rasterization which is demonstrated by trianglebin and is different compared to GCN2 and older GPUs. Still similar to immediate mode, but with small tiles fully rasterizing chunks at a time before moving on across the screen. NVIDIA's TBR moves on to partially rasterizing new large tiles while AMD's Fiji and Polaris fully rasterize many tiny tiles.


So, stuff that seems non-functional / non-beneficial:
DSBR
New 2.2x Geometry throughput / clock
Primitive Shaders


These three things were literally the most important updates to Vega to move AMD forward in gaming.

dum comments from "pro marketers"

Taking time for analysis and speculation =/ dum comments.
"dum comments" = stuff like above quote and triple posts with no information apt for discussion

[vario]

I am not a pro marketer......I just know that 1 vega will probably perform as good as my 2 290x ( 4g models so there is part of the math in my thinking ) and run a tad cooler.

I'm not trying to break WR with my daily and I know it will not perform @ the top level. I just prefer ATI/AMD graphics.

3dmark 11 scores are out. Vega gets ~50-60% more than the single 290X

https://videocardz.com/70777/amd-rad...11-performance

[BenchZowner]

TBR won't add much in performance, on a case by case scenario it can add up to 10% or so.

Where did you hear about the updated 11tri/clock setup ? Everything I have seen is pointing at the very same setup Fury X has ( 4tri/clk )

[vario]

TBR won't add much in performance, on a case by case scenario it can add up to 10% or so.

Where did you hear about the updated 11tri/clock setup ? Everything I have seen is pointing at the very same setup Fury X has ( 4tri/clk )

I mean besides all the marketing slides and new fancy names for various technologies.
Vega is 484mm2 on 14nm
Fiji is 596mm2 on 28nm. It cant be just shrink of fiji, there has to be some big changes in there, otherwise it should be smaller.

[BenchZowner]

It is enhanced, but definitely not "ground up" as they claimed.
From 28nm to 14nm you don't get a 50% decrease in die size ( nor their 14nm is "precisely" 14nm either ).

[kromosto]

Truth:

Early on, if u interested, OC`rs flashed 980 ti bioses on pascal and power moded `em.

Nowadays, U need to have a pro oc licence to get a proper BIOS. NV does now want anyone tinkering with pascals before the NDA. Imagine 2.4ghz 70/80/ti.. Would of killed the business of high end gpus.

No one would ve bought Ti if They could push such freq. out of their gpus, cuz the performance would skyrocket.

Now, that all Pascals are out, NV is ready to destroy AMD as they are dead set on Ti performer for ~<600usd.

The NDA is almost up and the long wait will be over!

Msi lightning z have a peak TDP of 339 as Guru3d states. They oced it to 2050 and said that overall power consumption increased 100w. I cant imagine 2.4 .

[vario]

It is enhanced, but definitely not "ground up" as they claimed.
From 28nm to 14nm you don't get a 50% decrease in die size ( nor their 14nm is "precisely" 14nm either ).

Yes, well we dont have info on how much better it "packs" at 14nm Glofo vs 28nm TSMC.
But, we know how polaris packs. And polaris is close to fiji architecture wise.
234mm2x2 and we get 300W(2xrx480) card at 464mm2 with 72CU`s and 512bit GDDR5 memory interface. So a smaller chip with more CUs with same TDP (for the LOW variant of Vega).
So vega is still bigger with higher tdp and lower CU count. There must be differences even to polaris.
And a single RX480 gets 19K in Graphics score on 3dmark 11.So almost double it(its should scale almost linearly) it would be even faster ...
Something must have gone wrong.
Not to mention Rx480 runs at 1200-1300 and this thing at 1400-1600.

[chew*]

3dmark 11 scores are out. Vega gets ~50-60% more than the single 290X

https://videocardz.com/70777/amd-rad...11-performance

Ill wait for the launch day reviews and give my 2 290x a spin..more concerned with games vs benchmarks for the daily.
I do not even know what i score in 3d11 on the daily...plus...cpu influences outcome...game tests = interesting.

Total score is not.

I have the fury x here.

In titles that supports x fire its a no brainer.

Keep in mind i run stock but flashed powercolor LCS.

Basically just a full cover block water bios.

Keep in mind high quality is what interests me...my 4gb limit can ruin the "experience"

[BeepBeep2]

TBR won't add much in performance, on a case by case scenario it can add up to 10% or so.

Where did you hear about the updated 11tri/clock setup ? Everything I have seen is pointing at the very same setup Fury X has ( 4tri/clk )

I know, performance is so similar it seems unimaginable.

AMD Tech Summit Presentation on January 5th -

New Programmable Geometry Pipeline - Over 2X Peak Throughput Per Clock (See endnotes for details)

Geometry throughput slide: Data based on AMD Engineering design of Vega. Radeon R9 Fury X has 4 geometry engines and a peak of 4 polygons per clock. Vega is designed to handle up to 11 polygons per clock with 4 geometry engines. This represents an increase of 2.6x

I said 2.2 in the other post because I calculated 11/5 - had a long day.


The Tech Summit slides also give more away about the rasterization process, you have the Draw Stream Binning Rasterizer slide that everyone was focused on and then the slides that say "Render back-ends are now clients of the L2 cache" and "Helps improve performance with applications that use deferred shading" yet performance per clock in most games seems almost exactly the same as Fiji.

[SinOfLiberty]

Msi lightning z have a peak TDP of 339 as Guru3d states. They oced it to 2050 and said that overall power consumption increased 100w. I cant imagine 2.4 .

Expected such commend.

Read what I said prior.

[StAndrew]

https://www.youtube.com/watch?v=0l6Uc5kBrxY&t=768s

So basically a FuryX with transistor shrink + finfet?

[chew*]

That almost makes me want to see both cards on a full cover block.

Can not help but think its throttling back.

We already know fury x can and will throttle and that is on...meh water...no clue if vegas on meh water or air.

[kromosto]

Expected such commend.

Read what I said prior.

I can imagine now

[Revv23]

I've been not paying attention for a while now, can anyone explain to me why the frontier edition would be launched before the high end gaming part? Like I said I've been out of the game, but it's even more confusing with what seems like a complete lack of communication from ati on what the plan is. A $1200 card that is slower than a 1080 while consuming more power is pretty boring...

Might have to go Nvidia this time around.

[Tim]

I've been not paying attention for a while now, can anyone explain to me why the frontier edition would be launched before the high end gaming part? Like I said I've been out of the game, but it's even more confusing with what seems like a complete lack of communication from ati on what the plan is. A $1200 card that is slower than a 1080 while consuming more power is pretty boring...

Might have to go Nvidia this time around.

I'm not even sure AMD can explain this!