AMD: Speculation, Rumors, and Discussion (Archive)

Status
Not open for further replies.
Yeah the Nitro is a real beast of value. I was really surprise with its price. Still wanting one but still no availability... :(
 
I saw this Anandtech article on SK Hynix's Q3 lineup including 4GB HBM stacks at 1.6 Gbps & 2 Gbps. What's the consensus for HBM configurations on Vega? Is it generally assumed that both little and big Vega will both use HBM? I've been out of the rumor mill loop.

My gut says that a GP104-competing little Vega would use two stacks of 4GB HBM at roughly 1.6 Gbps. That would yield the same VRAM capacity as GP104 and roughly 25% more bandwidth. AMD GPUs tend to have slightly more bandwidth than equivalent competing Nvidia GPUs, so I think that ratio seems reasonable.

But what about big Vega? If it competes with GP102, then it oughta be roughly 1.5x of GP104-competing little Vega since GP102 is roughly 1.5x of GP104. If little Vega would use two stacks of 4GB HBM, then big Vega oughta use three stacks, right?

However, is a 3-stack configuration even possible? I noticed that an old SK Hynix slide only lists "4 / 2 / 1 Cube" for "Config. / System". Is that the possible number of stacks for a given system?

sk_hynix_hbm2_implementations.png


It would kinda suck if big Vega was forced to use a full-blown 4-stack configuration. In that case, I imagine it would be underclocked similar to the P100.

Also, what HBM parts is Samsung making? I recall an old article stating that they are in mass production, but I didn't see anything in their catalog. I'm assuming someone has to be producing the HBM for GP100-based stuff.
 
Big Vega will presumably be targeting the top of the professional market (to replace the more than 2 year old W9100, it will have to compete with both GP100 and GP102), which will pay a lot of money for as much memory as possible. Thus 4-stacks makes sense. As with GP100, we might see cut-down 3-stack SKUs.
 
xEx said:
Price were discus on Polaris launch. 150 / 100
Click to expand...

I think they mentioned that they want to target those price ranges but they didn't give any hard pricing did they?

$150 seems too low for a part which should offer between 85-90% of RX480's performance.
kalelovil said:
Big Vega will presumably be targeting the top of the professional market (to replace the more than 2 year old W9100, it will have to compete with both GP100 and GP102), which will pay a lot of money for as much memory as possible. Thus 4-stacks makes sense. As with GP100, we might see cut-down 3-stack SKUs.
Click to expand...

Yep agreed. I'm not so sure little Vega needs HBM though, GDDR5X at 11-12 Gbps (I'm assuming these will be available by Q4'16 or Q1'17) on a 256 bit bus ought to be sufficient (With 64 ROPs though)
 
GDDRX5 would mean that AMD would have to invest into their GDDR memory controller, which is still less effective than Nvidia's version. I think Vega will be HMB with the little one using 2x4GB stacks, the big one probably four. With the option of going to 32GB.
 
seahawk said:
GDDRX5 would mean that AMD would have to invest into their GDDR memory controller, which is still less effective than Nvidia's version. I think Vega will be HMB with the little one using 2x4GB stacks, the big one probably four. With the option of going to 32GB.
Click to expand...

"Less effective" in what way? They've already laid the groundwork with 8 Gbps on Polaris. The higher price of HBM for a consumer card is not really justified.
 
It's probably not representative of intended workloads, but overclocking the 8 gbps memory on my RX480 by 10% gives me exactly 10% more MH/s in Ehtereum Mining.
 
Erinyes said:
"Less effective" in what way? They've already laid the groundwork with 8 Gbps on Polaris. The higher price of HBM for a consumer card is not really justified.
Click to expand...
Less effective because a similar memory OC on a 1070 and a RX480 gives you similar performance gains, while the overall performance and processing power is not similar.
 
I personally hope for HBM on Desktop Vega, that would significantly improve browser, multi-monitor and video decode power consumption. Since AMD hasn't yet managed to built to a GDDR5 memory controller in the past eight years with adequate clocking support (one that doesn't switch to full-speed clocks if you move the cursor) HBM is my only hope on that front.
 
Triskaine said:
I personally hope for HBM on Desktop Vega, that would significantly improve browser, multi-monitor and video decode power consumption. Since AMD hasn't yet managed to built to a GDDR5 memory controller in the past eight years with adequate clocking support (one that doesn't switch to full-speed clocks if you move the cursor) HBM is my only hope on that front.
Click to expand...

I couldn't agree more, the way AMD handles memory clocks in these low power situations seems like it's from the dark ages.
 
Erinyes said:
I think they mentioned that they want to target those price ranges but they didn't give any hard pricing did they?

$150 seems too low for a part which should offer between 85-90% of RX480's performance.
Click to expand...

Well thats the official price, if partners are going to sell it more expensive IDK.

An update: The Sapphire are back in the listing in Amazon but still no stock. Btw will any website list the nitro in its official price of 210? cuz everything I've seem is 230.
 
Status
Not open for further replies.

Similar threads

hoom
AMD: Zen 4, Speculation, Rumours and Discussion
8 9 10
Replies
199
Views
35K
MfA
M
Rys
Reopening Architecture and Products + rules and guidelines changes
3 4 5
Replies
90
Views
13K
Shifty Geezer
Shifty Geezer
D
AMD: Zen 2 (Ryzen/Threadripper 3000?, Epyc 8000?) Speculation, Rumours and Discussion
16 17 18
Replies
340
Views
148K
PSman1700
P
Shortbread
Next-Generation NVMe SSD and I/O Technology [PC, PS5, XBSX|S]
105 106 107
Replies
2K
Views
192K
Karamazov
K
L
A rambling history of AMD and nVidia's rise to the top
2
Replies
20
Views
5K
Leovinus
L
Back
Top