Next Generation Hardware Speculation with a Technical Spin [2018]

Status
Not open for further replies.
Because they lack exclusive reasons to own other than for those who must own the latest and best.

Doesn't that suggest exclusives are the real draw, not as power as you can cram in?

Totally unfair comparison, most people already bought their base consoles so the incentive to upgrade to a marginal power boost is not nearly as strong.

Based on Sony's own PS4 sales trajectory, this isn't true from the point when PS4 Pro launched. There will likely be as many, if not more, PS4 sytems sold after October 2016 than the period between PS4's 2013 launch and Pro's launch. And One X's power boost was over the original Xbox One was anything but marginal.
 
Imagine what 4 TFs of compute not used for drawing graphics could achieve though? Fluid simulations and all sorts. If used in the right places, I don't think the impact would be as diminishing as some believe.
Developers can always work with more resources, but people complaining about 10 TF have yet to even see what it can do, not to mention a better CPU.

Of course consumers will ask for more power...until someone tells them to get a second job.
 
Doesn't that suggest exclusives are the real draw, not as power as you can cram in?



Based on Sony's own PS4 sales trajectory, this isn't true from the point when PS4 Pro launched. There will likely be as many, if not more, PS4 sytems sold after October 2016 than the period between PS4's 2013 launch and Pro's launch. And One X's power boost was over the original Xbox One was anything but marginal.

I think you're taking everything said literally.

I am in team 'exclusives' camp, always have been (even when laughed at for the suggestion)... however, that doesn't mean that power and price are not important factors. I recall when PS2 came out it was being touted as a powerful beast of a console.

Likewise the wording of ultragpu is 'correct'. Most people who would be interested in pro and X had already bought the base machine.

I agree the X upgrade is much more significant (as is the price) which is why I'm interested in seeing figures.
 
AMD just announced a semi-custom SoC for the chinese console manufacturer Zhongshan Subor with the following specs:

- Single CCX Zen 4-core / 8-thread at 3GHz
- 8GB GDDR5
- 24 CU Vega GPU at 1.3GHz for 4 TFLOPs

It seems they mentioned Rapid Packed Math on the GPU, so it should be an actual Vega and not a Polaris 1.5 like Kaby Lake G.
There's apparently no indication on the memory using 128 or 256bit. The press release states the GDDR5 is "on package", so if it's on the same substrate (like the later RSX revisions) I'd bet it's 4x GDDR5 32bit chips. That would mean 128bit for ~128 GB/s total bandwidth if they're using 8 GT/s memory. It seems a bit anemic if they're going for 4K, but sufficient if the target is 1080p.

Looking at the specs, I'm guessing they're going for PS4 Pro / XBone X IQ on PC ports of MOBA / MMO / Battle Royale games, at 1080p resolution.

H8dN8P1.jpg




So the cat's out of the bag: there's already a console out there using Zen cores. It should be pretty safe to assume the Jaguar architecture isn't coming back.


EDIT: It seems the company will be launching a "Gaming PC" and a console carrying this same hardware. The only difference is the console will have a custom OS pre-installed, whereas the gaming PC is probably coming with Windows. Is this the first time there's a consumer Windows device using GDDR as main system memory?

I wonder if they'll be able to sell the SoC for other markets. NUC form factors using this SoC could become quite popular. It could take away a good chunk of Kaby Lake G's (already nichey) market.
 
Last edited by a moderator:
It seems they mentioned Rapid Packed Math on the GPU, so it should be an actual Vega and not a Polaris 1.5 like Kaby Lake G.

According to the article I read, they didn't say the GPU has RPM. It was more of a general statement about AMD's commitments. The wording seems to intentionally muddle the waters, not clarifying what the SoC actually supports, other than FreeSync.

https://community.amd.com/community...yzen-cpu-and-amd-vega-gpu-for-gamers-in-china

We are also committed to working across the developer community and with leading publishers such as Ubisoft to drive adoption of next-generation technologies like Rapid Packed Math that can deliver more realistic and immersive experiences. Not to mention our work as we equip the world’s most competitive eSports teams like Fnatic with powerful AMD Ryzen and AMD Vega gaming processors.
 
According to the article I read, they didn't say the GPU has RPM. It was more of a general statement about AMD's commitments.
That statement comes directly from AMD's press release. It would be terribly misleading if they mentioned RPM on bold when announcing a new semi-custom SoC that doesn't support it.

I don't know if they contacted AMD directly, but Anandtech is claiming the new SoC does have RPM.
 
That statement comes directly from AMD's press release. It would be terribly misleading if they mentioned RPM on bold when announcing a new semi-custom SoC that doesn't support it.

I don't know if they contacted AMD directly, but Anandtech is claiming the new SoC does have RPM.

It is not bold, it is a link to a page about the initial Vega release. So AMD would be misleading now to say that, but was not misleading before when it allowed Intel to say the GPU on Kaby Lake G is Vega, when it is Polaris? They opened the precedent for misleading already.
 
AMD are using that press release to chuck in some generic trumpeting; the opening paragraph is about the new SOC, whereas the second paragraph is about what AMD are doing for gaming in the wider scope. The purpose is to use the press release to advertise what else they might offer to potential clients - "we've made this custom SOC, and we invented Freesync, and we have this software, and we help developers including really big ones, and we support eSports teams." Only the explicit opening statement in that second paragraph links directly to the SOC announcement. Everything else is just general propaganda.
 
So AMD would be misleading now to say that, but was not misleading before when it allowed Intel to say the GPU on Kaby Lake G is Vega, when it is Polaris? They opened the precedent for misleading already.
IMO they were misleading with naming "Vega" To KBL-G's GPU.
Using the RPM expression within the same paragraph of the newly announced GPU would add yet another layer of deception.

AMD are using that press release to chuck in some generic trumpeting; the opening paragraph is about the new SOC, whereas the second paragraph is about what AMD are doing for gaming in the wider scope.
The second paragraph (the one that mentions RPM) starts with the following sentence:
The new gaming SOC is the latest example of how only AMD can combine high-performance CPU and GPU technologies to give gamers the most immersive experiences possible.

I'm not saying the new iGPU definitely has RPM. I'm saying the press release is misleading if it doesn't have.








But what I'd really want to discuss is that bus width.
128bit too narrow or not?
 
The second paragraph (the one that mentions RPM) starts with the following sentence:

The same paragraph also ends with

Not to mention our work as we equip the world’s most competitive eSports teams like Fnatic with powerful AMD Ryzen and AMD Vega gaming processors.

Unless Fnatic are using this SoC, the fact that RPM is mentioned in this paragraph tells us nothing about this SoC having RPM or not.
 
It does say that, but the paragraph isn't about the SOC, but about AMD in general. The SOC is mentioned as an item in a list of things AMD does for gamers.

1) We create custom SOCs, this is just the latest - we've made lots.
2) We create cool stuff like Freesync.
3) We make awesome software.
4) We work with developers to help them use our hardware features.
5) We equip eSports teams.

There's nothing there to tie Ubisoft to the new console, or Fnatic, nor anything else in that paragraph.
 
To be honest, even if this SoC does not have RPM, I do not consider it misleading. Right on the first time I read the press release, I understood that they were not saying it has. It does not need to mean AMD is trying to mislead, the PR team might have as well just copied that text from another press release without the intention to mislead. It might be unfortunate that it might lead people to think otherwise, but it may an honest "mistake" (again, if it does not have RPM, which I'm not saying it has or not).
 
Interesting development. I guess Anandtech's Ian Cutress is also confused about AMDs wording in what this new SoC is/includes with his final paragraph?:

The new SoC, name unknown, will support FreeSync, the Adrenalin software, and Rapid Packed Math, confirming that this is a true Vega design (unlike the chip used by Intel in its combination product).

The main questions I have about this are how will the 8GB GDDR5 be implemented on the package and how big will the SoC be? 16/14nm or 12nm?
 
The press release states the GDDR5 is "on package", so if it's on the same substrate (like the later RSX revisions) I'd bet it's 4x GDDR5 32bit chips. That would mean 128bit for ~128 GB/s total bandwidth if they're using 8 GT/s memory.
Unless I'm mistaken 16Gb gddr5 chips don't exist today. Hence for 8GB you need 8 chips anyway (regardless if with a native 8x32bit configuration or using clamshell mode for 128bit), unless you come up with some die-stacking methods. Although I'm not entirely convinced what "on package" here really means...
 
Unless I'm mistaken 16Gb gddr5 chips don't exist today. Hence for 8GB you need 8 chips anyway (regardless if with a native 8x32bit configuration or using clamshell mode for 128bit), unless you come up with some die-stacking methods...

You can also use a clamshell arrangement, but that would cause huge issues for a package intended to be mounted on a mainboard!

Looking at the PS4Pro, I'd think 1 x 8GB on a 256 bus, possibly avoiding the top rated speeds (8 and 9 gbpp). Maybe 7?
 
Unless I'm mistaken 16Gb gddr5 chips don't exist today. Hence for 8GB you need 8 chips anyway (regardless if with a native 8x32bit configuration or using clamshell mode for 128bit), unless you come up with some die-stacking methods. Although I'm not entirely convinced what "on package" here really means...
Samsung advertises 16Gb density as shipping for GDDR6, so it’s not out of the realm of possibility that GDDR5 could exist too.
 
Putting chips "on package" in this case would be just on a laminate, which is inexpensive and have no real size limitations. The memory works on a PCB impedance anyway so there's no need for expensive interposer materials or assembly.

I don't see why they couldn't make a clamshell design on the laminate, with board cutout or standoff layers... A laptop motherboard usually have crazier design stuff.
 
Status
Not open for further replies.
Back
Top