AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

lanek said:
The roadmap was posted on VC 7months ago, so i dont know what was the projection for HBM at this time.
Click to expand...

He's taking about the change in SK Hynix's Q1 2017 catalog as compared to previous catalogs.

It suggests that either SK Hynix gave AMD all 2 GHz HBM2 capacity, so they didn't want to list it on their catalog, or they were unable to meet the 2 GHz mark.

The former means Vega 10 can debut with 512 GB/s of bandwidth. The latter means it only gets 410 GB/s, i.e. Not enough.

Check out their catalogs from the past couple quarters:

https://videocardz.com/65649/sk-hynix-updates-memory-product-catalog-hbm2-available-in-q1-2017

Q1 2017 No 2 Gbps, 1.6 Gbps Only

SK-Hynix-Q1-2017-HBM2.png


Q4 2016 2 Gbps Only

SK-Hynix-Q4-2016-HBM2.png


Q3 2016 1.6 & 2 Gbps

SK-Hynix-Q3-2016-HBM2.png
 
Or, you know, they've removed it from catalog because AMD eats everything they can produce?
 
CSI PC said:
Err how is he coming to the conclusion 16GB with 2-stacks.
No-one has an 8-hi product available, meaning 2-stack is 8GB.
The 8GB also matches two different 'leaked' tools used for results regarding Vega.

Cheers
Click to expand...


The prediction was seven months ago, at the time I think the time lines for the HBM 2 wasn't know. And after AMD's promotion of 8GB is enough, pretty sure Vega will only come out with 8GB. So 1 stack.
 
lanek said:
The roadmap was posted on VC 7months ago, so i dont know what was the projection for HBM at this time.
Click to expand...
Catalogues have it only as 4-Hi from both and 4GB a stack.
This ties in with the recent benchmark 'leaks'.
TBH both manufacturers (SK Hynix and Samsung) have missed their original spec-rating target for launch (both going to miss 1.6GHz effective it looks like going by the leaked benchmarks), but maybe Samsung could hit 1.6 to 2.0 Gbps/GHz effective clock this year on their 4-stack as they have 12 months production experience under them.
Cheers
 
Last edited:
Razor1 said:
The prediction was seven months ago, at the time I think the time lines for the HBM 2 wasn't know. And after AMD's promotion of 8GB is enough, pretty sure Vega will only come out with 8GB. So 1 stack.
Click to expand...
Yes but he is still using that prediction as part of new articles...
So your going to have lots of people who read that assume Vega is now 16GB, without knowing that the 'leaked' benchmark details for Vega say 8GB and also the catalogues only say 8GB.
Cheers
 
ToTTenTranz said:
Hard limit for stack height is 8, so it's probably there as a limit.
Click to expand...
8Hi stacks are probably the practical limit for now. There is actually no hard limit. The DRAM manufacturers are free to fit as many dies in the specified stack height (720 +- 25 µm) as they can (the HBM spec specifically mentions the possibility to distribute a 128bit channel over multiple dies, the restriction is that the access latencies need to be constant within a channel [different channels within a stack can operate with different latency settings and even different clocks]). Given that stacked dies in flash memory can be as thin as 30µm, one may expect 16Hi stacks in the slightly farther future.
hbm_channel_distri9cun0.png
 
Each day that goes by without any relevant Vega news or leaks makes me lean towards thinking AMD is going to launch Vega in Q2 but only do the keynote+party+capsaicin+cigar parts close to June 30th, with reviews coming up 3 weeks later and actual cards another week after that.

AMD seems to have a marketing deal with 20th Century Fox to cross-promote Alien: Covenant (assuming this wasn't just for the "Meet Walter" short). That movie comes out in 2 weeks. If there's still nothing when that movie releases, I think there will be no Vega in consumer hands in Q2.
 
Too good to be true and no link for it, you can search and get the other score but there's only a 1080Ti entry for that score.

Though the graphics and cpu scores are different in that and they also bothered to put in the driver string from the other benchmark.
 
How is the GPU cost the same between 438mm^2 Hawaii and 596mm^2 Fiji?

Regardless, if costs for PCB, HBM, Interposer, packaging and GDDR5 are correct in that pic, we're seeing a difference of $41 between using 4 stacks of HBM1 and 16*32bit GDDR5 chips.

Completely opposed to the "almighty prohibitive costs of HBM" idea that has been pulled ouf of the ass preached by some people here.
 
Fiji is 20 months newer, so price per waffer was lower and yields higher. As for the $41 difference - the table doesn't take into account, that HBM interface is significantly smaller then GDDR5 interface, so hypotetical Fiji with 512bit GDDR5 would be larger and thus more expensive. It's also not possible to compare these prices directly, because 512bit GDDR5 interface of R9 290X offered 320 GB/s, while 4096bit HBM interface offered 512 GB/s. Price of HBM solution was higher, but bandwidth also (+60 %).
 
At great volumes, even 1 extra $ can prove prohibitive if the product has no tangible benefits w/ hbm vs without. So you're simplifying way too much. Maybe you're doing so in order to justify your beliefs. I don't care, I'm simply advising you to be careful ;)

( edit : but i'm definitely implying that 40 $ is nothing to sneeze at . And I'm amazed also at the gpu cost of 80$. IIRC , some 7ish years ago an Intel mid-range CPU had a production cost of ~10$ . If that info was correct, how times have changed )
 
Last edited:
ToTTenTranz said:
How is the GPU cost the same between 438mm^2 Hawaii and 596mm^2 Fiji?
Click to expand...
Assuming the numbers are accurate(they look reasonable), the die costs could be arbitrary or set equal as the comparison was in regards to the memory.

entity279 said:
At great volumes, even 1 extra $ can prove prohibitive if the product has no tangible benefits w/ hbm vs without. So you're simplifying way too much. Maybe you're doing so in order to justify your beliefs. I don't care, I'm simply advising you to be careful ;)
Click to expand...
HBM does have tangible benefits though. Form factor at the very least and latency at smaller page sizes.
 
Anarchist4000 said:
HBM does have tangible benefits though. Form factor at the very least and latency at smaller page sizes.
Click to expand...

It has benefits. I was reacting a bit to ToTTenTranz's post which had a rather dismissive style.

The price diference is also tangible . As is probably the R&D cost of a new memory controller IP & drivers. I don't know if it was worthwhile . By Navi, we'll probably get an answer (direct or indirect) straight from AMD. A Navi w/ HBM most like means AMD considers this tech a success.

Untill then, and with the upcoming Vega datapoint, we will speculate here.
 
That comparison is not really fair to GDDR5 at the time. AMD went with a lowly clocked 512-bit design for Hawaii, arguably to lower the power consumption, but the 780 Ti released within the same month, used a 384-bit bus and actually had slightly higher memory bandwidth (336 GB/s vs 320 GB/s) at probably significantly reduced costs.
 
41$ extra is too much for mainstream cards. Add margins + profits (manufacturer + stores) and we could be talking about nearly 200$ -> 300$ retail price increase. This would limit HBM2 to 400$+ high end cards. Not cheap enough for high volume products yet.
 
You must log in or register to reply here.
Back
Top