Next Generation Hardware Speculation with a Technical Spin [2018]

[Gubbi]

Vega 64 has 484GB/s for 12.5TF, Vega 56 has 410GB/s for 10.5TF.

With a 384 bit interface you have the choice of 12 or 24GB GDDR6, one is too little, the other too costly.

Zero chance of 18Gbit/s GDDR6. PS4 launched with 5.5GHz memory 2½ years after 7GHz devices were available. Cost and power.

Cheers

 

[AlNom]

With a 384 bit interface you have the choice of 12 or 24GB GDDR6, one is too little, the other too costly.

The JEDEC spec also includes 12Gbit chips.

 

[Gubbi]

The JEDEC spec also includes 12Gbit chips.

Didn't know that. Are anyone making those ?

Cheers

 

[anexanhume]

Check out these webpages before they get locked. Hopefully someone at reset era doesn't take credit if this is a unique discovery

I don’t understand the idea behind being protectionist of information publicly shared on a discussion forum with the expression purpose of stimulating conversation.

So you think an APU with an 8-core Zen 2 and a 12TF GPU part will have lower memory bandwidth available than a single Vega 64? I doubt that. Vega 64 has a memory bandwidth of 484GB/s, and Zen can peak at 50GB/s. Then you have to also consider that memory bandwidth will drop when both CPU and GPU access memory at the same time, which means there will be much less than your 448-480GB/s available. That's simply not good enough.

They could increase memory bandwidth to an acceptable 576 GB/s by using 18 Gb/s chips, but I don't think that's going happen. From what I have read, only Samsung had announced at one time that they have reached 18 Gbit/s for GDDR6 - but it's not even on their product page at this moment, so availability will probably be very limited.

It's a whole different picture if they choose a 384bit bus, because even with just 14Gb/s chips, they can reach a very impressive 672GB/s, which is a good fit for such a powerful APU.

Keep in mind we’re talking about consoles launching over a year from now. I would imagine GDDR6 offerings to be a little more readily available and varied. That being said, I don’t expect higher than 16Gb/s personally.

 

[Shifty Geezer]

Say if it's a 30-40% difference in power at launch I'll say it's gonna matter quite a bit, after all multiplatform titles are a good chunk of your console investment. 10-20% is a no issue though and I think that'll most likely be the difference if there's any if they launch in a similar time frame.

Xbox was way more powerful than PS2. Didn't make it sell better. A more powerful console may sell better, but you have presented it as a certainty, and it's not; power is just one of the variables. Wii was the least powerful console of the PS360 era yet sold the most, as an extreme example. I think PS4 is the first time the most powerful console of the generation sold the most.

This is an old line of reasoning and I'm surprised it's coming up again! The 'most power wins' theory has been blown apart time and time again, along with the 'first to market wins' theory.

 

[goonergaz]

True.

It is a value proposition, and the theoretical gigabollocks of the machine is just a part of that.

Cheers

Also worth noting if both have BC, the ability to take your library will be a big factor too.

 

[Shifty Geezer]

Looks like 3rd party publishers/developers have gotten or are getting somewhat finalized specs for the nextgen consoles in Fall 2018

Where are you reading that info? The bits you've quoted could just be working on PC hardware and appropriate DX12 techs without any knowledge of what the next-gen consoles will have.

 

[mrcorbo]

How? PS3 was a huge success, after all, and it launched more than 10 years ago. Yes, Sony struggled for quite a while, but the end result was good, and I think that a similar situation wouldn't be as bad in two years from now, being the new console a less exotic piece of tech, most likely, with more possibilities of a price drop in a near future, maybe.

Even though it was a tough move for Sony, many players didn't mind the price and payed for what they perceived as a superpowerful console. I don't see why people can't do the same again if they have a clear idea that there's a decent generational leap instead of a "slightly more powerful" console than the previous iteration.

PS3 sales were tragic until Sony took a hacksaw to the hardware to enable a price drop. In fact, with the exception of the Dual Shock 3 replacing the original Sixaxis and adding rumble motors, the PS3 shed hardware features with every revision so they could continue to reduce their cost. By comparison, the 360 actually gained hardware features with later revisions and still got price drops. The original PS3 design that made it necessary to charge $599 (and still lose money) was a problem they spent the entire generation resolving and the design of the PS4 was clearly driven by lessons learned after the mistakes they made with the PS3.

So, no, I would not call the PS3 (at it's $599 price point) a success.

That having been said, I don't see a problem with launching the next gen at $499. If they do, though, consumers had better see that $100 directly manifested in the performance of games on the system. The prior attempts to launch at that price point were such spectacular failures due to that price difference going into hardware features that didn't deliver better performance in games. As long as nobody does that, they can probably get away with a price bump at launch.

Edit:Changed the references to $499 for PS3 to $599. As I've said before the $499 PS3 model had limited availability, so for a lot of people it was $599 or nothing.

 

[AlNom]

Didn't know that. Are anyone making those ?

Cheers

Haven't seen any news about it, but I suppose the focus for now is probably on pumping out 8Gbit.

For the moment they've defined configurations for 8/12/16 with room for 24 & 32 (TBD).

---

I wondered what the thought process was for that - mobile and/or laptop?

The twin 16-bit channels per DRAM is also curious, but maybe that's more beneficial for CPU-type accesses (and APU) ?

 

[anexanhume]

Haven't seen any news about it, but I suppose the focus for now is probably on pumping out 8Gbit.

For the moment they've defined configurations for 8/12/16 with room for 24 & 32 (TBD).

---

I wondered what the thought process was for that - mobile and/or laptop?

The twin 16-bit channels per DRAM is also curious, but maybe that's more beneficial for CPU-type accesses (and APU) ?

I was hoping someone would have weighed in on that by now. I saw something that suggested it would help it look like a 32 bit GDDR5 interface at half the speed for legacy compatibility?

 

[iroboto]

whoever fixed Pixel's post, thank you. I was going to post about it being unreadable in the dark theme.

with X1X being 384-bit would there be issues for BC compatibility for enhanced titles moving to 256 bit for next gen?

 

[Shifty Geezer]

You can answer that question yourself when you consider PCs are able to emulate PS2 with its 2560 bit bus.

 

[iroboto]

You can answer that question yourself when you consider PCs are able to emulate PS2 with its 2560 bit bus.

haha oh yea

 

[mrcorbo]

whoever fixed Pixel's post, thank you. I was going to post about it being unreadable in the dark theme.

with X1X being 384-bit would there be issues for BC compatibility for enhanced titles moving to 256 bit for next gen?

As long as there is enough bandwidth available on the smaller bus I don't see it making much of a difference. If MS could reconcile the differences between the Xbox One and One X's memory configuration, whatever differences there are with what comes next should be relatively trivial.

 

[MrFox]

Didn't know that. Are anyone making those ?

Cheers

I've been betting on it for some time, it would be a nice 18GB at 14gbps.

No parts announced yet, but it needs a product requiring 12gb parts, and a console with a stable 200m parts per year, for many years, is a nice contract. Then memory manufacturers have a lot of the higher speed bin to sell at a high price for gpus.

I think this can be a repeat of the PS4 getting a good price on their 8GB. And from that point of view it might allow 24GB if memory prices crash back down to normal, or if they wait for 2021.

 

[BRiT]

Where are you reading that info? The bits you've quoted could just be working on PC hardware and appropriate DX12 techs without any knowledge of what the next-gen consoles will have.

I think he's keying in on the marketing sales pitch of selling the desire to want the job(s). Here's the only semi-relevant blurbs:

Infinity Ward, developer of numerous award winning games, is seeking a Senior Lighting Artist, preferably with experience in game development and an interest in working with next-generation technologies.

Activision Blizzard is seeking talented engineers to join its Central Technology division on an internship or co-op basis for Fall 2018. This is a rare West Coast opportunity to develop cutting edge games technology, learn the next-gen consoles inside and out, and to work with top developers like Sledgehammer, Treyarch, Infinity Ward, and Vicarious Visions.

 

[mrcorbo]

To expand a bit on memory bus-width vs. bus speed (as far as I understand it). You're basically looking at being able to service more memory requests at any given moment with a higher bus width, but with a faster bus you can finish each request faster. So the wider bus benefits when there is a higher number of requests but the faster bus is always going to be able to return at least part of the available work faster while, at full saturation, maybe having to queue some requests for a later cycle. Like with multiprocessing, the faster solution wins all else being equal because it's benefit is always at least somewhat realized, no matter the workload. The choice to go wider is generally made when physics/cost prevent getting the same performance by going faster.

 

[iroboto]

To expand a bit on memory bus-width vs. bus speed (as far as I understand it). You're basically looking at being able to service more memory requests at any given moment with a higher bus width, but with a faster bus you can finish each request faster. So the wider bus benefits when there is a higher number of requests but the faster bus is always going to be able to return at least part of the available work faster while, at full saturation, maybe having to queue some requests for a later cycle. Like with multiprocessing, the faster solution wins all else being equal because it's benefit is always at least somewhat realized, no matter the workload. The choice to go wider is generally made when physics/cost prevent getting the same performance by going faster.

Would a wider bus reduce disproportionate bandwidth contention however?
60 FPS doubles bandwidth use over 30fps, while physical bandwidth keeps as a constant.
That being said, the CPU is running the memory twice as hard gutting the bandwidth available to the GPU.
*correct me if wrong*
Say CPU normally takes up 20 gb/s at 30fps. The bandwidth available to GPU is now 120 say.
But at 60fps it's going to be 40 gb/s. The bandwidth now will be < 100.

If the goal is high framerate support for next gen, will bus width be a factor?

edit:
wrt this old document

 

[mrcorbo]

GB/s is GB/s. If they enable the same number of GB/s with the faster bus they don't need a wider bus. It's possible at any given instant the narrower memory controller could have more requests in flight than there are lanes to service them, but there will surely be plenty of cycles (when the cycles are measured in nanoseconds) within that second where there are fewer requests in flight and the faster bus can catch up.

 

[MrFox]

I wondered what the thought process was for that - mobile and/or laptop?

The twin 16-bit channels per DRAM is also curious, but maybe that's more beneficial for CPU-type accesses (and APU) ?

Yeah, gddr5x had to double the prefetch size, I wonder at which point data granularity becomes a problem on the CPU side, versus the GPU side.

I think the idea is that with the latency remaining the same, the only way to scale seems to be either more independent channels, or coarser data granularity.

In the case of gddr6, it's practically like having two gddr5 chips in one package, double rate, half width and keeping the same prefetch size. All commands and timings are supposed to be similar from the controller point of view, it just needs twice the controllers on the SoC side (and a new phy that can do 16gbps)