AMD: Navi Speculation, Rumours and Discussion [2019-2020]

Status
Not open for further replies.
HBM wins at "signal power" but it loses out at total power due to crap IO dies on the HBM itself. Imho AMD used an immature technology too early for the mass market.
That crap IO die is insignificant considering total power draw. AMD, Nvidia, etc all use HBM for high bandwidth designs since it uses less energy per unit of bandwidth. The simple physics of driving a signal over shorter distances with smaller drivers will always win against comparable designs. Less capacitance, less burn. I wouldn't say it was immature, just the entire ram industry was disrupted by mining and Vega seems to be selling for reasonable prices despite what everyone claims is an overly expensive memory system. Performance, size, cost; pick two.
 
Do we have any actual quotes on the HBM-price including packaging? I mean sure, it's more expensive than GDDR5, that's for sure, but apparently GDDR6 isn't that cheap either: https://www.3dcenter.org/news/hardware-und-nachrichten-links-des-jahresausklangs-2018
They're quoting 70% higher price for same density 14 Gbps GDDR6 compared to 8 Gbps GDDR5
HBM cost may have fluctuated, but mid 2017 the cost of HBM 2 8GB Memory Stack was about $160
https://segmentnext.com/2017/05/24/amd-rx-vega-hbm2-memory-cost/
 
Last edited:
Anyways, you currently get a Vega56 with 8 GB HBM2 for the same price as a GTX 1070 with 8 GB GDDR5 (both starting at roughly 349,- EUR in Germany). Right now HBM2 probably is equally expensive as GDDR6, minus the packaging cost.
 
Anyways, you currently get a Vega56 with 8 GB HBM2 for the same price as a GTX 1070 with 8 GB GDDR5 (both starting at roughly 349,- EUR in Germany). Right now HBM2 probably is equally expensive as GDDR6, minus the packaging cost.

Retail prices would be indicative of component costs if margins were constant across brands and SKUs, but they're not.
 
We are not talking companies, but technologies. Nobody cares about the margin they make... this isn't an investor's forum.
 
Profitability comes on how you leverage your said technology(ies).

You are assuming to know AMD's marketing around their new cards coming. And talking about profitability on a static part, not the whole Company's mindshare/marketing scheme, etc.

What if: AMD instead released a low margin, high selling (must have cost ratio) card..? A small profit of a killer card that grabs the market by surprise isn't judged solely by it's actual margins, but also by mindshare and other factors.
 
Profitability comes on how you leverage your said technology(ies).

You are assuming to know AMD's marketing around their new cards coming. And talking about profitability on a static part, not the whole Company's mindshare/marketing scheme, etc.

What if: AMD instead released a low margin, high selling (must have cost ratio) card..? A small profit of a killer card that grabs the market by surprise isn't judged solely by it's actual margins, but also by mindshare and other factors.
There are some things you should know. I haven’t been a product manager for a little while and never for GPUs but certain principles apply.

A) margin is everything. If you don’t have a lot of margin you have limited space to move and any fluctuation on the market causes severe damage to low margin products. So if you’re going to spoil up all your manufacturing for a low cost high volume product, you have to sell all of it because you’ve committed your manufacturing towards it. If you hit recession and no one buys your products, you’re dead. Which is why high volume low cost tends to coincide with iGPUs now. They work in multiple formats and devices.

B) the other way to do margin is to take as much advantage of binning as possible. Same product different performance levels. This is done today, and there are a great deal of many ways that GPUs pricing strategies can happen but leading with a low cost high volume product is pretty backwards. It’s much easier to optimize your inventory with higher products and bin downwards.

C) know your volumes and market size. You can’t sell more than what the market is willing to accept and you have to be realistic about your expected penetration into the market. Once again this is why low cost high volume doesn’t work so hot. Saturate the market with cheap stuff and you didnt profit much, that’s the best case scenario.

Profit = revenue - expenses. How you leverage technology is not in that equation, so you’ll need to explain that concept further.
 
Profitability comes on how you leverage your said technology(ies).

You are assuming to know AMD's marketing around their new cards coming. And talking about profitability on a static part, not the whole Company's mindshare/marketing scheme, etc.

What if: AMD instead released a low margin, high selling (must have cost ratio) card..? A small profit of a killer card that grabs the market by surprise isn't judged solely by it's actual margins, but also by mindshare and other factors.

Well, I'm sure he doesn't know EVERYTHING, but I guess Rys, working for RTG, has some insight...
 
Last edited:
What if: AMD instead released a low margin, high selling (must have cost ratio) card..? A small profit of a killer card that grabs the market by surprise isn't judged solely by it's actual margins, but also by mindshare and other factors.
This has already been executed and promptly abandoned - HD 4800 aka Sweet Spot.
 
Hahaha :LOL:

Yeah, maybe back down on this one w0lfram :p

Pffft, nah. Why would a corporation go after profits?

Regardless, for HBM better packaging options are definitely on the table in the near future. EMIB like bridges and etc. will make it less expensive to integrate, now if only memory manufacturers could get HBM prices down it might go back to being preferential to GDDR.

But that probably won't happen this year, I'd bet on Navi... whatever the cards will end up being. Low end/mobile, mass market consumer, and high end late in the year, all being on GDDR6. Beside you only need 8gb for mass market consumer, and 12-16gb for high end. Maybe AMD could dig up it's old small footprint 512bit bus designs.
 
Status
Not open for further replies.
Back
Top