Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]
- Status
Last edited by a moderator:
A possible reason is that the register file only has one read port thus allowing one access per cycle. Going to a multi port register file would have significant area and power implications.
Anarchist4000
Veteran
Little off topic here, but Microsoft indicated in the future they may be doing away with consoles. Question being, what are the practical differences between Scorpio and and a Zen APU for instance? Shared memory and likely similar GPU power, so with SM6.0 could console code be ported almost directly? The other question is how discrete cards with separate memory pools might coexist with that environment? Would treating VRAM as L3/4 cache alleviate this and remove a lot of the resource management or are devs going to have to code for each? That seems to be the sticking point with a lot of DX12/Vulkan ports so far.
It will be quite some time before a consumer Zen APU has similar GPU power to Scorpio.
Anarchist4000
Veteran
Around the middle of next year? Maybe not widely available, but I wouldn't be surprised if AMD started pushing down prices of those configurations for higher performance SFF systems by the end of next year. I'm still expecting a full Vega to get bolted onto a Zen. In the timeframe of a console refresh a system like that would seem a likely solution. Building a new console would seem a bit irrelevant, so the "doing away with consoles" move seems appropriate. Which goes back to the original question, how to handle the shared/discrete memory pools.
The reasons why this will not happen are economic as much as technical. AMD are not going to introduce a product that threatens their discrete GPU business, so the integrated GPUs will always have performance at or near the bottom of their current product stack in that category. The PS4 has been out for 3 years now and there still doesn't exist an APU with matching GPU specs (the latest APU will slightly exceed the GPU in the XBOne). There is, IMO, zero chance of a 6TF GPU in a consumer GPU next year.
Last edited:
D
Deleted member 11852
Guest
That's flawed logic. If somebody is going to cannibalise AMD's existing products it's better than it's a new AMD product rather than an Intel or Nvidia product. You can't afford to stand still and adopt a turtle strategy in the tech sector.
High preformance APUs for laptops seems a sensible move. I doubt people will build gaming rigs with APUs as those no upgradability. Enable seamless GPU sharing across APU and second GPU and you add a strong tie to the AMD ecosystem.
I would agree if Nvidia made APUs or if Intel had anything like this on the GPU side. They don't, though, and nothing on either of their product roadmaps indicate that they will have anything close any time soon.
Scorpio-class chip in a laptop, though? In the near future? I just don't see it.
Anarchist4000
Veteran
They've already announced HPC Zen, which is effectively a big APU for middle of next year. They also said leading graphics capabilities for APUs coming with Zen. For laptops and SFF (which is a fair approximation of a console) they might as well sell the product if they have it. An APU with HBM on it could probably make use of a 6TF GPU. I'm not sure they should be worried about cannibalizing discrete GPU sales with a CPU+GPU combo. That seems like a big win to me.
Intel does have Phi, but it doesn't really do graphics. Nvidia lacks a CPU (beyond PowerPC) so not an option there. Sounds like a good market.
Could be more or less, but does it need to equal Scorpio or XB1? Even in a SFF design cooling would be simplified as everything vents externally. Smack a giant 120mm fan atop a little board and port the sides.
Define near? Two years or with a drop to 10/7nm it'll likely be readily available. Heck someone got a pair of 1080s in a laptop, a scorpio class chip should be easy by comparison. Cost would be the only concern and that situation is only improving with time. It could be a Steambox, XB mockup, or any other platform out there as well.
Which gets back to the original question, what to do with the memory pool? Shared memory, or a fast connection to system memory, would seemingly address a lot of the resource management and timing issues of the low level APIs.
D
Deleted member 11852
Guest
Nvidia and Intel aren't standing still. I thought hell would freeze over before Intel really got into ARM but we living in strange times. AMD are not in a great position financially and can't rest on their laurels or assume that their competitors will.
That's not a consumer product and I made that distinction very clear. It is a product that only makes sense for the market it is being targeted at.
To have leading graphics capabilities for an APU, they just have to exceed what Intel offers.
The discrete GPU market is a lower risk for AMD because some of the burden is shared with the OEMs who actually make the products being sold to consumers. They buy the chips from AMD and if they don't sell, they take at least part of the hit (I'm sure AMD doesn't completely leave them holding the bag). By making a product like this not only does AMD risk alienating their OEM partners, but if AMD manufactures it in volume and it doesn't sell AMD is screwed.
A "good market" that Intel are choosing to ignore for....reasons.
Wait, what? That's what this whole discussion has been about.
That laptop OEM bought one CPU chip and two GPU chips for each unit they made. I'd imagine the companies who made the chips are quite happy with that. That OEM was also able to, from their purchased stock of those same exact components, make laptops with one CPU and one GPU and one CPU and no GPU to serve different market segments and if one of the products doesn't sell they can still make money off of the purchased stock of components by making more of the products that do. I'm sure they are quite happy with that. To paraphrase Jurassic park, I think you are spending so much time thinking about if they can you are not thinking enough about whether they should.
They aren't standing still. They are also definitively not moving in the direction of PC APUs with "performance"-class GPUs. Even Nvidia's Shield TV (which I think is awesome) development is stagnant and the product has been re-focused as a streaming box.
D
Deleted member 11852
Guest
You are staggeringly well informed about what Intel and Nvidia are doing.
It is staggeringly easy to be so informed. Intel and Nvidia have to share with investors what they have in the pipeline to address current and future markets and on top of this Intel roadmaps intended for their industry partners leak constantly and those go even further out.
- Status
