Vega would have to have a GDDR memory controller to use GDDR.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Vega Rumors
- Thread starter grtitan
- Start date
Vega would have to have a GDDR memory controller to use GDDR.
No low cost HBM is not HBM3
Also Vega with 4 stacks, unless they halved their bus, that isn't going to happen.
Anything more than 8 gb is going to same stacks doubled up per stack, not 2 more stacks.
Last edited:
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
Why would using an identical bus be like asking to use a bigger bus? Wider bus provides bandwidth, deeper provides capacity. Bandwidth doesn't have to scale with number of stacks. Same concept as every other memory technology out there. The only power increase is from a simple doubling of memory. I wouldn't categorize 4GB and 8GB cards as hugely different power draws beyond reason.
Doesn't work like that with HBM. HBM doesn't have clam shell capabilities of GDDR. More stacks = more bandwidth that is the way HBM works. higher stacks means = bandwidth.
You wouldn't put a higher clocked 8 GB parts which show less perf/watt improvements over the first gen of 4 GB HBM, power draw won't be different? Increased bus size also increases die size granted HBM bus is a bit different than GDDR.
One thing I did really like about the HBM is it allowed for a really dense design. The R9 nano is still a pretty sweet card. I'll be interested to see if there will be a 16GB HBM2 part offered as was mentioned earlier, even if its at titan price levels.
Vega 20 in 2019 or so will have 4 stacks. You are not getting more than 2 stacks on Vega 10.
HBM3 isn't cheaper and its not around the corner. If lucky you see it around 2020. And I think you confuse it with the slower and cheaper HBM2 edition. Something that is not around the corner either.
Wow, you won't stop with your poor quotes. Directly from the original article that provided the photos above:
"Perhaps most importantly though, given the high cost of HBM1 and HBM2, HBM3 will be cheaper to produce." https://arstechnica.com/gadgets/2016/08/hbm3-details-price-bandwidth/
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
I am sure Raja lied to the crowd too. Just as the slides and all. Or maybe its just you....
Since I showed slides from 2 completely different presentations that you mixed. I think YOU should prove that low cost HBM and HBM3 is the same. Specially when low cost HBM is SLOWER than HBM2.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
Do what? 4 stacks? No it cant. Vega 20 can(2019 or so), but Vega 10 cant.
probably not,
and no its not benchmark related, look at comparable cards and it should be a fairly linear increase
This is a compute benchmark, this doesn't push graphics or anything like that, just the compute array.
True still never seen a card that gets 0% increase with 25% in clock speed. Never! I can bet every dollar in the bank that in gaming that wont happen lol. 1200mhz vega vs 1600mhz is not going to perform with 0% increase. Just won't happen, period!
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
Again, mixed 2 different presentations. That's Ars got it wrong as well is their fault.
But let me help you, HBM3 is expected to be 2x faster than HBM2. Low cost HBM is going to be around 20% slower than HBM2. Now put 2 and 2 together.
That post was removed though. So I don't know there might be some truth to it, cuz usually fanboys don't remove posts. lol
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
razor1 already showed you your lack of homework.
"In addition to HMB3, Samsung unveiled two other memory technologies at Hot Chips: GDDR6 and "Low Cost" HBM."
Not that its any surprise, since HBM3 is up to 512GB/sec and low cost HBM is up to 200GB/sec.
who? reddit? reddit doesn't give two shits about who posts what! Unless someone from AMDs side reported it to be taken off since is in AMD sub reddit.
I doubt the OP who originally said come back in three weeks to check my post suddenly thought he needed to add a source when he already said don't ask him for source.
Typical of you, change the subject and make up a new one, when you are losing the argument. The topic is HBM3, not low cost variants of it. Are you drunk?
There is a poster he
The mixed up is from a poster who thought HBM3 is Low Cost HBM when it isn't based on slides from ArsTech article. Frankly I am not even sure why HBM3 matters as Vega10 most likely wont use it.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
Regarding the SK Hynix Graphics memory Q2'17 catalogue info I posted.
Worth noting that again the 1.6 Gbps 4GB HBM2 slipped its product schedule date.
Previous quarter had the date Q1, now this quarter says Q2.
So it shows they are either experiencing production challenges (which Samsung did as well when they started last year) or if you are an optimist AMD is not just buying 2Gbps that was removed from the catalogue but also the 1.6Gbps - yeah they would not.
Personally I see it more to do with production challenges, 2Gbps is dead for now and also limited to 4-Hi (will be interesting if they launch Mi25 with 4-stack).
Cheers
Worth noting that again the 1.6 Gbps 4GB HBM2 slipped its product schedule date.
Previous quarter had the date Q1, now this quarter says Q2.
So it shows they are either experiencing production challenges (which Samsung did as well when they started last year) or if you are an optimist AMD is not just buying 2Gbps that was removed from the catalogue but also the 1.6Gbps - yeah they would not.
Personally I see it more to do with production challenges, 2Gbps is dead for now and also limited to 4-Hi (will be interesting if they launch Mi25 with 4-stack).
Cheers
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
14Ghz GDDR6 also comes this year and 16Ghz early next year as well. Looks fantastic for GDDR.
cageymaru
Fully [H]
- Joined
- Apr 10, 2003
- Messages
- 22,080
THat is one fucked up data book. It seems like its all over the place lol. I am sure though AMD is taking everything they can get their hands on right now. lol
I like Ars Technica generally but their computer technology coverage should be taken with a grain of salt. The only person somewhat qualified to make any comments about it is deficient in living up to his surname.
Yep data matches what I have and that catalogue is May release (this month).
I wonder if the catalogue is online, would be easier to link to it and is easy to read.
Cheers
SK Hynix also had that in their earlier catalogues for HBM2, along with the 2Gbps.
8GB went 1st, and then followed by 2Gbps.
Then 1.6Gbps keeps being delayed with every quarterly catalogue released - I do think it will be Q2 but I really think they have problems hitting 1.6Gbps and only reaching same level Samsung did when they launched and that was 1.4Gbps.
Only with the latest cycle has Samsung managed to hit above 1.6Gbps.
Cheers
cageymaru
Fully [H]
- Joined
- Apr 10, 2003
- Messages
- 22,080
More possible photoshop. 
https://chiphell.com/forum.php?mod=viewthread&tid=1735472&page=1#pid36401553
https://chiphell.com/forum.php?mod=viewthread&tid=1735472&page=1#pid36401553
cybereality
[H]F Junkie
- Joined
- Mar 22, 2008
- Messages
- 8,789
So a bunch of sites are reporting an unsubstantiated reddit post and not even quoting the source? Some fine journalism right there.
SighTurtle
[H]ard|Gawd
- Joined
- Jul 29, 2016
- Messages
- 1,410
Not sure why anyone is even caring, is Chiphell a good source?
TaintedSquirrel
[H]F Junkie
- Joined
- Aug 5, 2013
- Messages
- 12,689
Chiphell is the Asian equivalent of HardForum. It's not "a source" at all.
Magic Hate Ball
Limp Gawd
- Joined
- Jul 28, 2016
- Messages
- 366
If they can get SK Hynix HBM2 cheaper and faster Samsung HBM2 for more money, why not put them on their respective different tiers?
RX470 and RX480 don't use the same RAM chips, I don't see why the HBM2 for all tiers of Vega have to come from the same company.... it only makes sense to differentiate by your own product tier where it makes sense financially.
EDIT: And since the highest tier will sell the least (almost assuredly) it won't need as much total volume.
RX470 and RX480 don't use the same RAM chips, I don't see why the HBM2 for all tiers of Vega have to come from the same company.... it only makes sense to differentiate by your own product tier where it makes sense financially.
EDIT: And since the highest tier will sell the least (almost assuredly) it won't need as much total volume.
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
Well why did we just see a leak of a 64CU card with 16gb? Presuming it's legit (seems to be) then it's single GPU.
Either AMD has 8gb stacks or are using (like the Fury x many of you like to say it's a clone of) four stacks.... or have figured out a way to tack GDDR on there (unlikely).
Which one is it? I think 4 stacks is looking more likely at this point and it's been done before. Has AMD Specifically said Vega can't use 4 stacks?
edit: if it was vega20 memory pooling/etc 64CUs makes no sense
This is exactly why diversifying suppliers in your supply chain can help when scaling.They may have an MOQ/contract though to provide from one though. Who knows.
Last edited:
Is samsung in some sort of exclusive agreement with HBM2? I mean they have had this for a bit, what is stopping AMD from knocking on samsung's door to get more chips. If Hynix keeps fucking up and can't get their shit together to ramp up supply. I am sure there is no such thing as binding agreement that keeps screwing you over and delaying your product that hurts your position in the market. Unless ofcourse Samsung has some sort of exclusive agreement with Nvidia.
First, they said Zen won't be above 3Ghz. There's no way it's competitive against 5960X or 6900K for half the price...
Then the same bunch said there's no way Vega will clock high. They said it's a bigger chip so 1200mhz is it! It wouldn't be possible at 1500mhz, no chance in hell faster than that below 300W.. blah blah.
They claim there's no chance for Vega to reach Pascal perf/w, no way for it to match the 1080Ti, it's only around 1070 or 1080 level!
Now they say there's no way it's 16GB HBM2, because this and that, as if they know it all.
I wonder why these same people are such downers on everything AMD related. Really if you hate AMD so much, piss off out of these AMD related threads. Your negativity has been proven wrong many times, enough is enough?
Then the same bunch said there's no way Vega will clock high. They said it's a bigger chip so 1200mhz is it! It wouldn't be possible at 1500mhz, no chance in hell faster than that below 300W.. blah blah.
They claim there's no chance for Vega to reach Pascal perf/w, no way for it to match the 1080Ti, it's only around 1070 or 1080 level!
Now they say there's no way it's 16GB HBM2, because this and that, as if they know it all.
I wonder why these same people are such downers on everything AMD related. Really if you hate AMD so much, piss off out of these AMD related threads. Your negativity has been proven wrong many times, enough is enough?
now you will hear that it is AMDs fault they feel that way, lol. I have a 1080ti absolutely love it. But I sure as hell hope AMD brings Vega and it can clock at 1600mhz. Heck I would love that level of competition. Now people are saying oh it will be a dual chip card. Same people said to beat or meet 1080 it will have to clocked atleast 1500, now they need a dual chip card clocked at 1600?
You are right, there are way too many people that jump on anything positive AMD related and try to discredit it. It can't possibly beat their card in that rig. I welcome a competitive card from AMD that makes me consider another option to the card in my rig, its good to have that shit. I would love to have two options to consider from.