• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Vega Discussion Thread

  • Thread starter Thread starter Deleted member 50521
  • Start date Start date
xkm1948 said:
Shhh, you are going to offend the AMD fanboys! They will call you a faggot and tear the mini skirts off your profile picture!
Click to expand...
strangest boner right now
 
xkm1948 said:
New rumors folks, rise and shine!

http://www.tweaktown.com/news/57418/amd-radeon-rx-vega-less-20-000-available-launch/index.html

According to tweaktown, the intial batch of Vega may be very limited due to HBM2 shortage.

This is really starting to look like FuryX release vol 2:

Super hyped up by fans: check
Delayed forever: check
Possible limited stock follow launch: check

Now we just need to find out the part regarding price and performance.
Click to expand...
obviously faster then a GTX 1337TiT
 
I think ~GTX 1070 is likely true.
 
HBM2 isnt real, its all an elaborate rambus troll
 
Grings said:
HBM2 isnt real, its all an elaborate rambus troll
Click to expand...
uhh no HMB is a product of AMD and SK hynix
and it is a JDEC standard
 
xkm1948 said:
New rumors folks, rise and shine!

http://www.tweaktown.com/news/57418/amd-radeon-rx-vega-less-20-000-available-launch/index.html

According to tweaktown, the intial batch of Vega may be very limited due to HBM2 shortage.

This is really starting to look like FuryX release vol 2:

Super hyped up by fans: check
Delayed forever: check
Possible limited stock follow launch: check

Now we just need to find out the part regarding price and performance.
Click to expand...
You are the one going on about vega daily mate ,its people like you chugging that hype train , i count myself as neutral but my policy based build says different and so would many others on here but,,,,,,,,


I am not on here chatting poop about a mythical unicorn and its daily bench run.

Im eager for Vega, im eager for the momentum in the marketplace and in terms of performance available but im not talking it up or down till that head crowns at least.
 
It's hilarious how everyone is bashing Vega for limited launch supplies. Should I remind everyone how unobtainable GTX 1080 was on launch? Or has everyone already forgotten about that one?
 
Nv have been smart and stuck with using HBM2 on their low yield high return GP100 parts, the other.... remains to be seen.
 
I think GDDR6 makes HBM moot again. HBM has advantages but in graphics, the supply problems outweigh them.
 
waiting till its out
 
FordGT90Concept said:
I think GDDR6 makes HBM moot again. HBM has advantages but in graphics, the supply problems outweigh them.
Click to expand...

And while compact format of HBM is cool, it's not really needed on graphic cards because the tiny ones are also cheapo ones which won't see HBM for the next 20 years and the big high end ones have to be big anyway because of the coolers. So, yeah, only thing left are the niche Mini/Nano type high end cards. However, HBM has an advantage with latency and a really huge one, so, for HBC on RX Vega, it's needed just because of that alone whether we like it or not. If it was a traditional design like GTX 1080Ti with shitload of VRAM, then it wouldn't matter. But because Vega will be swapping data in and out of cache quite a lot it needs it (latency almost more than bandwidth which can be achieved with GDDR5X already). I personally believe HBC is the future and NVIDIA will also adopt it in the future. Seeing the trends how VRAM capacity is moving forward quickly, I don't think it's feasible to stick so much of it on graphic card. It makes more sense to make a clever design that can fetch data directly from SSD or system RAM which can be used better for everything where graphics memory is just for graphics and nothing else.
 
Not even out yet...
 
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.
 
IMG_1453.JPG
GDDR6 may pair within NVidia GTX 2000 Series (Pascal Enhance) around Q4 2017 or Q1 2018.

Due its double bandwidth from GDDR5. We can see GTX 1000 Series pair with GDDR6 and rebranding as GTX 2000 Series on 2018.

This may cause AMD RX Vega base GPU pointless to competition. Performance gap between Red and Green way to high.

IMG_1454.JPG


Overclock headroom will be widely range when compare with HBM2 or HBM3. Many media expect Volta GPU may going with GDDR6.

If Nvidia follow GDDR6 route. AMD may suffer an atomic bomb on GPU market 2017-2020.
Redteam will death. Uncle raja may going to sell his peanut with salt to Apple after AMD sinking in the death sea.
 
Last edited by a moderator:
Yeah i read about it. It's just not ground braking news so as the ver 6 of GDDR itself.
 
RejZoR said:
And while compact format of HBM is cool, it's not really needed on graphic cards because the tiny ones are also cheapo ones which won't see HBM for the next 20 years and the big high end ones have to be big anyway because of the coolers. So, yeah, only thing left are the niche Mini/Nano type high end cards. However, HBM has an advantage with latency and a really huge one, so, for HBC on RX Vega, it's needed just because of that alone whether we like it or not. If it was a traditional design like GTX 1080Ti with shitload of VRAM, then it wouldn't matter. But because Vega will be swapping data in and out of cache quite a lot it needs it (latency almost more than bandwidth which can be achieved with GDDR5X already). I personally believe HBC is the future and NVIDIA will also adopt it in the future. Seeing the trends how VRAM capacity is moving forward quickly, I don't think it's feasible to stick so much of it on graphic card. It makes more sense to make a clever design that can fetch data directly from SSD or system RAM which can be used better for everything where graphics memory is just for graphics and nothing else.
Click to expand...
HBCC, most of a Vega for that matter, is aimed at compute customers (deep learning, OpenCL, DirectCompute, etc.).

ratirt said:
GDDR6? What are you talking about?
Click to expand...
http://www.tweaktown.com/news/56164/gddr6-launch-year-16gbps-bandwidth-offer/index.html
GDDR5X, HBM2, and GDDR6 all release in rapid succession.

ratirt said:
Yeah i read about it. It's just not ground braking news so as the ver 6 of GDDR itself.
Click to expand...
GDDR5 has been around a decade. That's about two lifespans in terms of memory tech.
 
While HBC was first introduced in their FirePro card with SSD, it's very much feasible for for consumer gaming graphics.
 
Remember HBM1 on my Fury X, the best version of HBM1 running on 1GHz.
Those HBM 1 GHz speed archive 1 Terabytes per second bandwidth.
It was big leap performance gain as you see below.

500MHz HBM1 (Stock clock)
IMG_1613.JPG


Overclocking HBM1 to 1GHz
IMG_1612.PNG


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.

+ 15% GPU performance paired with -15% HBM bandwidth.

R9 Fury X and RX Vega must be little different.
 
Last edited:
You must log in or register to reply here.
Back
Top