DigiTimes is the bearer of bad news for fans of GPUs, as the supply challenges which have marked 2017 are now spreading to GDDR5(x). This month the price has spiked up just over 30% and that trend is going to continue into September and perhaps beyond. This will not have an immediate effect on the MSRPs of graphics cards, not that we would notice due to the price inflation from the current mining craze however it will reduce the margins that NVIDIA and AMD receive from sales. They do not specifically mention AMD in the article, nor HBM2, however the same companies fabricate both so there are likely to be repercussions felt by both technologies. On the positive side, flash storage prices are reported to have stabilized; so we have that going for us.
"August quotes for RAMs used in VGA graphics cards have risen to US$8.50, up by 30.8% from US$6.50 in July. Both RAM industry leaders Samsung Electronics and SK Hynix have allocated part of their VGA RAM production capacities to producing memories for servers and handsets, fueling the price rally."
Here is some more Tech News from around the web:
- Fresh Microsoft Office franken-exploit flops – and you should have patched by now anyway @ The Register
- Outage outed: Bing dinged, Microsoft portal mortal, DuckDuckGo becomes DuckDuckNo @ The Register
- How To: Taking Pictures of PCBs @ Techspot
Is GDDR5(x) made on the same
Is GDDR5(x) made on the same fabs as HBM2?
I would hope that this would
I would hope that this would affect HMB2 less due to major manufacturing differences in the process unless the production lines are interchangeable. Or maybe the prices will rise just because they all are and it can be gotten away with.
Can someone comment with more knowledge about the fab process differences between DDR and HBM?
Probably more the second
Probably more the second point. They use different processes and HBM2 doesn't have the demand of GDDR but it is still the same manufacturers and I would be surprised if the price increase does not effect HBM2 prices in any way.
The HBM2 bassed parts demand
The HBM2 bassed parts demand has already seen its share of price increases but AMD’s Vega 56/64/FE and Radeon Pro WX/Instinct demand is just ramping up with Nvidia’s Quadro HBM2 SKU demand increasing and Volta is incoming on the professional front while any consumer Nvidia SKU demand is going to be towards GDDR5/5X with GDDR6 implications, with AMD’s GDDR5/newer usage included on their maistream Polaris SKUs also.
HBM2 production is mostly SK Hynix and Samsung and Samsung has a whole lot of resources to throw at ramping up HBM2 production. Samsung’s working on a lower cost HBM variant and also HBM3 with its much faster clocks and 512GB per stack bandwidth and more than double the memory capacity(higher stacks).
The professional markets are going to get their HBM2 needs met come whatever compared to the consumer markets where there is still GDDR5/GDDR5X and GDDR6 production with Sk Hynix, Micron, and Samsung lookiing towards GDDR6 in 2018.
Samsung is really the one to benifit from HBM2 price increases with Samsung’s production ramp-up abilities and those price increases will most definitely allow Samsung to more quickly amortize It’s HBM2 R&D and fab line tooling exenses rather quickly. Samsung is supplying both AMD’s and Nvidia’s 8-Hi HBM2 production needs. But this Anandtech headline(1) is not good news for GPU HBM2/memory pricing and SK Hynix can also pay down/amortize its HBM2 R&D and other costs.
GPUs with all that mining price pressure influnce are becoming a Pay to Play at 4k market with 1440P and 1080P gaming in line to cost more also. The spoils will go to those memory producers who can ramp production up the fastest for the long run when pricing pressures began to relax. SK Hynix and Samsung look to be the ones with the product that everybody really wants for the future with HBM2/HBM3 eventually taking over from the older technology once that economy of scale pricing kicks in fully and the HBM2/newer fabs/Fab equipment is paid off fully and it’s mostly gravy from then on.
Look Out gamers, deep pockets are required and the only good news so far is that they have not figured out a way to make the gaming consoles effectively mine those coins.
(1)
“SK Hynix: Customers Willing to Pay 2.5 Times More for HBM2 Memory”
http://www.anandtech.com/show/11690/sk-hynix-customers-willing-to-pay-more-for-hbm2-memory
Hmh, some odd price numbers
Hmh, some odd price numbers for ram. Ain’t gddr unit prices depending more on memory speed and memory density. Or is that just some average price for memory chip.
??? I thought OEM build the
??? I thought OEM build the cards, not AMD.
ASUS for example buy a Polaris 10 from AMD for $70, buy the ram from samsung and others… and assembler their PCB and custom cooling.
AMD/nvidia should see ZERO impact on ram price. unless its HBM2, and in this case they will pass the price onto the OEM.
You are forgetting the new
You are forgetting the new business model though … both companies initially release Founders Editions directly, then let the third parties at it. The third parties walk in knowing what RAM they are allowed to use and what the price point is and will negotiate accordingly.
The there is that even better
The there is that even better market news and it’s called the professional markets where the great margins for GPUs will be gladly paid, with those professional market users able to deduct the price of the GPU/s from their taxes as a valid business expense! And the best news yet for any CPU/GPU business, that is in that CPU/GPU business, will be that they will no longer be hamstrung by any dependency on only the consumer/gaming markets that are fickle by nature. AMD’s Epyc CPUs and Vega micro-arch based Radeon Pro WX and Radeon Instinct GPU SKUs will be able to outperform the Pascal competition in many of the professional/compute/AI workloads.
Just go over and look at that Techgage article on RX Vega 64’s compute abilities, and that reviewer will maybe be getting his hands on a Radeon Vega FE with its 16GB of VRAM to put through some compute/cartographic benchmarking.
Should AMD’s marketing apologize for that Vega Pricing kerfuffle, sure and AMD should go on and raise its Vega pricing! Nvidia will, on seeing AMD raise it pricing, will raise theirs a C-note more than AMD raises its prices and price equilibrium will return to the GPU market place, at a higher average price point unfortunately.
Now AMD will have the funds to really engineer a gaming only GPU mciro-arch, and better compete with the competition!
Raja needs to get a great bonus for that compute/AI Vega GPU micro-arch design that can also game pretty well compared to the competition, even if it uses a little more power. Vega is a compute/AI GPU micro-arch first and formost. Great Job Raga that’s one number crunching Vega 64bit, 32bit, 16bit, and 8bit GPU micro-arch, and it games ok also(Just Look to DX12/Vulkan for the future)!
30% sounds ridiculous, but in
30% sounds ridiculous, but in the end it’s a fucking couple dollars. Big fucking deal, your graphics cards now cost a couple more bucks to make.
Fuck off with the misleading clickbait bullshit.