Has AMD Reduced RX Vega Supply & Prioritized Frontier Edition?
- Thread starter badsykes
- Start date
very smart
AMD might as well make the money or more money off the ****en miners as newegg
as miners are the only ones buying vega at this point anyway
AMD might as well make the money or more money off the ****en miners as newegg
as miners are the only ones buying vega at this point anyway
Frontiers are listed at almost 2000$ a piece Can, while they were are ~1200$ when i was debating the purchase of either the liquid version with 8GB onboard at the 1000$ mark, or going with the frontiers for the extra 200$ for each one.
Even though it's likely that the 16GB of HBM 2.0 adds cost to the overall card compared to the 8GB version, the current price tag more than offsets that and then some.
Even though it's likely that the 16GB of HBM 2.0 adds cost to the overall card compared to the 8GB version, the current price tag more than offsets that and then some.
Argoon1981
New member
Good for AMD, they were not selling them to gamers at that price and even if they were cheap the majority would go with Nvidia instead anyway, selling them to miners means more RD money for their GPU or CPU department, at lest that is my hope.
They posted a fourth quarter with healthy profits today, after years of losing money left and right right up until the middle of 2017, which was when Ryzen was introduced.
Now theres the Ryzen refresh in March and thread ripper 2.0 coming in the second half of 2018, regardless if there's a Vega refresh intended for gaming in 2018 ( or not ).
They're covered on the CPU side for 2018 at least.
AMD Ramping GPU Production, Confirms Memory Is Behind Shortage
https://wccftech.com/amd-ramping-gpu-production-confirms-memory-behind-shortage/
https://wccftech.com/amd-ramping-gpu-production-confirms-memory-behind-shortage/
I think this is more due to memory makers reorganizing production capacity to DDR4 memory, which as we all know has easily doubled in price over the past year, and is used everywhere from home desktops to servers, databases and supercomputers, and there the LPDDR4 side of the coin as well, which is used in cellphones too and the massive amount of those sold every year, worldwide.
By contrast, GDDR5 / 5x and HBM are only used in GPU's and nothing else, so the market is inherently much smaller and volatile.
LordHawkwind
Active member
And this is most likely why there isn't a 12nm RX Vega refresh on AMD's road map. Why bother? You can stay on 14nm and sell every chip you make regardless. Might as well keep working on Navi for the next 18 months and that'll be why they are looking at a 7nm Instinct card to test out the node. By the looks of things they might just jump from 14nm straight to 7nm and be done with it. I also think AMD, board partners and re-sellers are all now hooked on the crypto drug and are struggling to get off it. Those high prices are certainly addictive!
The potential real loser here is Nvidia because if the prices stay this high how are they going to sell the 1170/1180? I would imagine, Happa and a few others beside, most current Pascal owners will wait until the prices normalise again, if they ever do. TBH it could be a real mess for everyone as you can pick up an XBone X for £400 and play games at 4K or pay 3 x that to play at 4K on a PC. Doh..
My RX Vega 64 air now seems a real bargain at £449!!
And this is most likely why there isn't a 12nm RX Vega refresh on AMD's road map. Why bother? You can stay on 14nm and sell every chip you make regardless. Might as well keep working on Navi for the next 18 months and that'll be why they are looking at a 7nm Instinct card to test out the node. By the looks of things they might just jump from 14nm straight to 7nm and be done with it. I also think AMD, board partners and re-sellers are all now hooked on the crypto drug and are struggling to get off it. Those high prices are certainly addictive!
The potential real loser here is Nvidia because if the prices stay this high how are they going to sell the 1170/1180? I would imagine, Happa and a few others beside, most current Pascal owners will wait until the prices normalise again, if they ever do. TBH it could be a real mess for everyone as you can pick up an XBone X for £400 and play games at 4K or pay 3 x that to play at 4K on a PC. Doh..
My RX Vega 64 air now seems a real bargain at £449!!
Vega56 / 64 may become viable for collectors in the future...
At least now we have a better idea of what may come in the next 18 months, when it comes to Vega and Navi and an unnamed architecture beyond that, which Lisa Su is on record as saying they're committed to release by 2020.
https://www.anandtech.com/show/12233/amd-tech-day-at-ces-2018-roadmap-revealed-with-ryzen-apus-zen-on-12nm-vega-on-7nm
The last page in that article speaks about Vega and the plans for 2018, namely that AMD will release mobile versions as well as supply intel for some of their own APU's, and that they also mention the 7nm version but it's for deep learning workloads and not gaming, which also has the side effect of "cleaning the pipe " on the 7nm process for eventual gaming related products in 2019.
The most important thing however, is that she mentions there will be a separation between compute and more gaming focused architectures, as right now GPU's try to do everything but can't be as efficient, since gaming workloads don't deal with the problems that compute workloads have to, namely branching, speculative execution and more......It's all parallel workloads all the time when it comes to gaming.
https://www.anandtech.com/show/12233/amd-tech-day-at-ces-2018-roadmap-revealed-with-ryzen-apus-zen-on-12nm-vega-on-7nm
The last page in that article speaks about Vega and the plans for 2018, namely that AMD will release mobile versions as well as supply intel for some of their own APU's, and that they also mention the 7nm version but it's for deep learning workloads and not gaming, which also has the side effect of "cleaning the pipe " on the 7nm process for eventual gaming related products in 2019.
The most important thing however, is that she mentions there will be a separation between compute and more gaming focused architectures, as right now GPU's try to do everything but can't be as efficient, since gaming workloads don't deal with the problems that compute workloads have to, namely branching, speculative execution and more......It's all parallel workloads all the time when it comes to gaming.
Treeckcold57
New member
I'll buy both if they do that.
But otherwise, I seriously doubt that. Also selling the cards to either miners or gamers would be difficult.I'll buy both if they do that.
Samsung is designing a processor specifically for mining workloads, so there may be a lot of used GPU's on the market later this year...
More to the point, if AMD is separating product lines into compute focused and gaming focused using distinct architectures, the only reason for that is because reaching performance goals needed to be competitive for the time of release for both types of workloads ( 2019 ), while still using the same architecture, is becoming hard as hell.
NWR_Midnight
New member
The question is: will miners buy these Samsung mining hardware over GPU's? Because if crypto dies out or goes into a big slump, they won't be able to re-sell the Samsung mining hardware like they can GPUs.
The only way it would seem to entice them to buy such mining hardware is if it is able to do substantially out perform GPUs so it would make GPUs a waste even if they could resell them.
It's not that impossible if you think about it......We're already at 12.5 billion transistors on Vega at 14nm, now imagine a design exclusively meant for computing with that kind of transistor budget?
Fast forward to 2019 and if the 7nm process is only half the size of 14nm, where transistors are only half as long and half as high and use use much less power too, we're looking at designs packing 50 billion with the exact same die size as Vega, so the same production costs if yields are good.
A mining chip with that budget would bury any GPU that has to do both mining and compute.
Come to think about it, we're already seeing the split starting to happen right under our noses, since GPU's used to do FP64 double precision for compute workloads requiring that much precision, at half clock right up to the Hawaii seen in the R290 / R390 cards and ever since then, it's been set at 1 FP64 operation every 32 clocks.....it's pretty pathetic by comparison, and it also applies to Nvidia GPU's too.
Hawaii is outdated for it's graphics performance but can still kill anything more recent in FP64 performance, but that is the choice designers had to make to allow better graphics performance in later chips......one can only " fit " so much for a given fab process.
Another example is big Volta, with it's dedicated hardware for deep learning which does nothing for it's gaming performance and makes the chip a lot bigger, but it seems it's capable of 110 Teraflop under those workloads.......With it's 5000 shaders for FP32 single precision shader operations used in games, Volta has 15 Teraflop worth of math power.
Bottom line is that dedicated hardware is always faster and in Volta's case, nearly 8X faster than trying to make the standard shaders within the chip handle them.
Hawaii is outdated for it's graphics performance but can still kill anything more recent in FP64 performance, but that is the choice designers had to make to allow better graphics performance in later chips......one can only " fit " so much for a given fab process.
Another example is big Volta, with it's dedicated hardware for deep learning which does nothing for it's gaming performance and makes the chip a lot bigger, but it seems it's capable of 110 Teraflop under those workloads.......With it's 5000 shaders for FP32 single precision shader operations used in games, Volta has 15 Teraflop worth of math power.
Bottom line is that dedicated hardware is always faster and in Volta's case, nearly 8X faster than trying to make the standard shaders within the chip handle them.
Bigfoot
New member
FP64 performance isn't nearly as important for compute as people make it. Most scientists use it because they don't know or care whether or not they need it.
I have access to 4 p100s, 1 1080 Ti, 1 1080 and 1 980 Ti. FP64 is nice on the p100s because I can use mixed precision for even more performance (there are 50% more ALUs if you use both FP32 and FP64) but otherwise I run in single precision.
What I find strange is that earlier cards had such strong FP64 performance to begin with. The reason we see 1 per 32 these days is to maintain compliance with libraries like OpenCL and CUDA etc.
I have access to 4 p100s, 1 1080 Ti, 1 1080 and 1 980 Ti. FP64 is nice on the p100s because I can use mixed precision for even more performance (there are 50% more ALUs if you use both FP32 and FP64) but otherwise I run in single precision.
What I find strange is that earlier cards had such strong FP64 performance to begin with. The reason we see 1 per 32 these days is to maintain compliance with libraries like OpenCL and CUDA etc.
I don't know. I don't think the OEM's or AMD benefit from the current high prices. The re-sellers, retailers and distributors do by doing markup at the last moment.
The time-lines for production, sales, shipping etc etc etc operate in the region of many many month in advance. Cards arriving today, where probably ordered, built and purchased way before any of the pricing went insane.
Not to mention, the OEM's won't bite. The crypto market is too fast and um-predictable for them to adapt or bother trying to engage. At least for this generation.
The time-lines for production, sales, shipping etc etc etc operate in the region of many many month in advance. Cards arriving today, where probably ordered, built and purchased way before any of the pricing went insane.
Not to mention, the OEM's won't bite. The crypto market is too fast and um-predictable for them to adapt or bother trying to engage. At least for this generation.