AMD Vega Discussion Thread

[uuuaaaaaa]

Even if GDDR6 was out, it only solves the bandwidth issues, not the latency ones. HBM on the other hand solves both.

It's similar to normal DDR for system RAM. We get higher frequencies with each release, but timings just keep on climbing. Anyone remembers RAM when it had 2-3-3-6 timings? These days, 10-10-10-20-ish is considered extreme top of the line RAM that costs both kidneys if you want it at high enough speeds. VRAM is not really much different. HBM on the other hand, physical construction allows that.

2 2 2 5 on samsung TCCD's

 

[FordGT90Concept]

HBM2 lower bandwidth from my R9 Fury X and those 1TB/sec Pro Duo.
Vega GPU alone may 15% better than Fiji GPU. But the HBM2 Bandwidth may 15-20% lower bandwidth than R9 Fury X HBM1.

It has half the stacks of memory which is effectively half the bandwidth (1024-bit instead of 2048-bit). Additionally, the Compubench and 3DMark leaks have it clocked at 700 MHz.

While HBC was first introduced in their FirePro card with SSD, it's very much feasible for for consumer gaming graphics.

There's a reason why professional cards have 32 GB VRAM. Games require substantially less so HBC could theoretically add more time per frame than it removes. Think FB-DIMM versus DIMM.

 

[RejZoR]

The problem with textures is that we haven't optimized them for decades. Main problem being, they are now MASSIVE. We have created texture compression as well as normal maps compression, but they are all lossy (and really shite), but once texture enters the rendering pipeline, it's converted to RAW images and those are even bigger. I think way to go would be capability for high level, high compression, high quality lossy or lossless compression that is both, in-memory as well as in-pipeline. So that processing of textures only takes long in the game studio, but on end computers, we save VRAM capacity textures occupy as well as give GPU more bandwidth because it would be able to process textures in its native optimized format without exploding them into massive unoptimized (raw) format. Plus, games would probably be at least 30-40% smaller install wise (which at 50GB game means 20GB less!). Now, I'm not an engineer, but knowing things it could be done if anyone actually cared. Because as things stand now, it's easier to just slam more VRAM on the graphic card and call it a day.

 

[FilipM]

2 2 2 5 on samsung TCCD's

Kingmax Hardcore DDR500 running 2.5-3-3-5-1T at 585MHz on an Athlon X2 4200+

Those were the days, bandwith was better than DDR2 800 and sightly worse than 1066...Latency? There wasnt any haha


Sorry for off topic

 

[R4E3960FURYX]

Latest benchmark of RX Vega to discuss buddy.

 

[uuuaaaaaa]

View attachment 87751

Latest benchmark of RX Vega to discuss buddy.

That entry is from April 12th.

 

[Kanan]

It has half the stacks of memory which is effectively half the bandwidth (1024-bit instead of 2048-bit). Additionally, the Compubench and 3DMark leaks have it clocked at 700 MHz.

*2048 bit for Vega and 4096 bit for Fury X or 4 stacks of HBM(1/2). They are trying to save money by going with two stacks instead of four this time, probably countering the lower bandwidth (yes, probably lower than Fiji) with its HBC.

 

[R4E3960FURYX]

Look like the remain reason people buy AMD GPU was image quality.

*2048 bit for Vega and 4096 bit for Fury X or 4 stacks of HBM(1/2). They are trying to save money by going with two stacks instead of four this time, probably countering the lower bandwidth (yes, probably lower than Fiji) with its HBC.

Absolutely Yes! This bring RX Vega GPUs were not able to compete with R9 Fury X in term of bandwidth.
And look like it may overkilled by Volta GV 100 1.4GHz 5120 Cuda Core 320 TMU 32MB SM Memory 16GB HBM2 @ 900MHz already now.

 

[Kanan]

Look like the remain reason people buy AMD GPU was image quality.


Absolutely Yes! This bring RX Vega GPUs were not able to compete with R9 Fury X in term of bandwidth.
And look like it may overkilled by Volta GV 100 1.4GHz 5120 Cuda Core 320 TMU 32MB SM Memory 16GB HBM2 @ 900MHz already now.

View attachment 87792

That Volta you linked is a professional GPU for servers, costing way over 100K probably, like the last "deep learning" thing Nvidia released, based on P100. News editor made a bad job by not saying this clearly enough and now all people go around and think we get 800+ mm² sized 12nm GPUs. Dream on, we're not even getting 600mm² 16/14nm atm. I already said it, I expect a Volta based GPU for consumer to be less than 500mm² again, like Titan Xp and have about 4200-4500 shaders (4500 being the perfect GPU without deactivated cores). It could be all the same again, imperfect Volta Titan X and then 1 year later the full activated one, named "Titan Xv".

 

[r9]

AMD Radeon Vega Graphics Card Featuring 16 GB HBM2 and 1600 MHz Clock Speed on 4096 Cores Leaks Out
http://wccftech.com/amd-radeon-vega-specs-leaked-4096-cores-16-gb-hbm2-1600-mhz/
That clock would put it between gtx 1080 and gtx 1080ti.
Carrying 16GB VRAM looks like that gonna be the top tier Vega.
On one hand its not bad and on the other I'm not sure its where AMD were aiming for.

 

[uuuaaaaaa]

AMD Radeon Vega Graphics Card Featuring 16 GB HBM2 and 1600 MHz Clock Speed on 4096 Cores Leaks Out
http://wccftech.com/amd-radeon-vega-specs-leaked-4096-cores-16-gb-hbm2-1600-mhz/
That clock would put it between gtx 1080 and gtx 1080ti.
Carrying 16GB VRAM looks like that gonna be the top tier Vega.
On one hand its not bad and on the other I'm not sure its where AMD were aiming for.

Hype train is accelerating, I heard a sonic boom already!

 

[dorsetknob]

AMD Radeon Vega Graphics Card Featuring 16 GB HBM2 and 1600 MHz Clock Speed on 4096 Cores Leaks Out
http://wccftech.com/amd-radeon-vega-specs-leaked-4096-cores-16-gb-hbm2-1600-mhz/
That clock would put it between gtx 1080 and gtx 1080ti.
Carrying 16GB VRAM looks like that gonna be the top tier Vega.
On one hand its not bad and on the other I'm not sure its where AMD were aiming for.

Why go their when its here on TPU

https://www.techpowerup.com/forums/...portedly-leaked-available-on-june-5th.233325/

 

[FordGT90Concept]

https://compubench.com/compare.jsp?...type1=dGPU&hwname1=AMD+6864:00&D2=AMD+687F:C1

The only thing they benchmarked on 6864 was the spot where 687F was particularly weak. I think this might be the Radeon Instinct implementation of Vega. The clockspeed matches the 12.5+ TFLOP suggested and 16 GiB VRAM makes sense in the compute environment. It also stands to reason that the Instinct chip was in the works before the Radeon chip which is why this sports a lower ID number.

 

[R4E3960FURYX]

Who want to challenge first? 1400MHz base clock available 2 July 2017.

 

[HTC]

View attachment 88058

Who want to challenge first? 1400MHz base clock available 2 July 2017.

Dude: that's supposed to be the MI 25 deep learning card. It's not the "gamers" version of Vega, sadly

 

[RejZoR]

Interesting naming scheme of different Vega versions. It's really unusual and pretty cool at the same time.

 

[xkm1948]

AMA session with Raja in Reddit is going on right now. So the VEGA FE can be used as a gaming card just fine.

 

[uuuaaaaaa]

AMA session with Raja in Reddit is going on right now. So the VEGA FE can be used as a gaming card just fine.

The mighty Raja!

http://i.imgur.com/thCSaDS.gifv

 

[R4E3960FURYX]

Coming Soon Radeon RX Vega

:: Unofficial spec :Reference Card : 1.7 GHz Core Clock 1.9GHz Boost Clock 108.8 Gpixel/sec 435.2 Gtexel/sec (@1.7GHz) 64 ROP 256 TMU 4096 Shader 16GB HBM2 @ 480GB/sec (*512GB/sec overclock capable*) 6 color scheme. Red with white LED blink , silver with red LED blink , grey with red LED blink , white with red LED blink for
fan solution. Blue with yellow LED blink and Yellow with blue LED Blink for Liquid cooling solution.

599$ price tag for FAN model , 649$ price tag for Liquid cooling model available June 5 , 2017 not sure street price may be higher than recommended price like R9 Fury X.

Another color see also ::
http://wccftech.com/amd-lower-priced-gaming-optimized-radeon-rx-vega-coming/
https://videocardz.com/69636/just-some-concepts

See more detail upcoming RX VEGA here :: https://videocardz.com/69679/official-amd-radeon-rx-vega-showcase-at-computex

 

[xkm1948]

OMFG this is SEXY as hell

 

[kruk]

AMA session with Raja in Reddit is going on right now. So the VEGA FE can be used as a gaming card just fine.

AMA recap: https://www.reddit.com/r/Amd/comments/6bzu05/raja_kaduri_ama_recap/

A even shorter recap of things i think are most interesting:

• The demo during Financial Analyst day was an air-cooled Frontier Edition, not an RX Vega card. There will be water cooled versions of Vega that will run slightly faster.
• RX Vega will be shown at Computex. It will not be available the same week but nothing else has been ruled out.
• Frontier runs comfortably using 1x6-pin and 1x8-pin but RTG put 2x8-pin on the production card for more headroom.
• Frontier was designed for an array of workload usages. RX Vega is for gaming and will be faster than Frontier.
• New geometry pipeline in Vega improves throughput per clock cycle and will require no extra work on dev's part to utilize.
• RX will have different drivers than Frontier that are optimized for gaming as well as additional goodies.
• Vega is the first GPU architecture to use Infinity Fabric and is in no way a re-hash of Polaris

 

[ratirt]

OMFG this is SEXY as hell

I agree it is . honestly I'm thinking about buying the water cooled Vega if the price is right. I only hope AMD does a good job with the Vega card. I'm still waiting for it and it's killing me. heh.

 

[Caring1]

The LC version must have a full block on it, as there is no visible fan on the card itself, only on the radiator, either that or it runs extremely cool not to require one over the VRM section.
I think it's funny they had to bring out an FE card also, it's almost like they are having a dig at Nv.

 

[Vya Domus]

Seems like it could very well rival the 1080ti , those clocks sure look extremely optimistic though.

 

[ratirt]

Seems like it could very well rival the 1080ti , those clocks sure look extremely optimistic though.

Well they do look optimistic. RX 580 with those clocks might give GTX 1080 a competition and this is Vega so who knows. Hopefully VEGA gives a 1080TI a decent competition.