Baseless Next Generation Rumors with no Technical Merits [pre E3 2019] *spawn*

Status
Not open for further replies.

mm... I wouldn't be terribly surprised. If Lockhart didn't exist, then $399 would be more likely to happen, I think. As I mentioned in another thread, getting stuck in between pricepoints of the competitor might be a little awkward in terms of positioning for a "true next gen" experience, so... maybe they should just compete at the higher pricepoint and boost the HW accordingly.

There can still be some value to keeping around the current mid-gen consoles (if they do a fab revision); early cross-gen ports can be differentiated by framerate thanks to the CPUs while the gaining ubiquity of dynamic res handles the rest. On the other hand, the Xbox One Family probably has less reason to continue unless there's some uptick in HW sales. With PS4 encroaching 100M territory, a revised 4Pro could be positioned as a lower priced entry-point (lower than Lockhart) although a fair bit will depend on the relative uptake of next gen sales circa 2020-2022, which will need to be observed/analyzed before publishers start toying with the idea of going next-gen only because the PS4 does command that 100M-ish userbase. Meanwhile, the architectural similarities make porting less of a headache.

Bolt on Ray Tracing like any early cross-gen gimmick until devs are more experienced in shifting the engines. The segment of the PC gaming market that is served by cross-platform might hamper things for a little while, but that sort of lag is typical (see DX9 & DX11 engines. We're still not yet at a majority for DX12/Vulkan.)
 
Last edited:
AMD_E3_2019_678x452.jpg


The image provided in the anandtech article, I suspect this means all their next cards are 7nm. not that we didn't already know this, but I suppose this is the confirm. And on their roadmap 7nm is Navi. So this is really a presentation about Navi. More or less confirmed.

Like Jay, i have a hard time believing that the base console will be weaker than X1X. It will mean that the new console can't run old games at X1X settings. Just seems downright wrong and I would expect a lot of customer complaints for this.

At the very least Lockhart must match the performance of X1X.

I also think we're at the majority of engines being DX12 now and somewhat matured.
 
Like Jay, i have a hard time believing that the base console will be weaker than X1X. It will mean that the new console can't run old games at X1X settings. Just seems downright wrong and I would expect a lot of customer complaints for this.

As if the customer "knows" the settings of games on the X1X and can really put them in relation to the 1080p product running on the low-end next generation. The faster cpu will also skew FPS comparison most of these sites might deliver or which is relevant for these customers. Most X1X customers will upgrade to the top model anyway so I'm really not getting your point here.
 
As if the customer "knows" the settings of games on the X1X and can really put them in relation to the 1080p product running on the low-end next generation. The faster cpu will also skew FPS comparison most of these sites might deliver or which is relevant for these customers. Most X1X customers will upgrade to the top model anyway so I'm really not getting your point here.
From my perspective there's a couple reasons why you want Lockhart to match 1X.
Xbox one games will either run at 1S or 1X settings, so that means usually 900p on 1S not 1080p.
OG Xbox and x360 games wouldn't get the 1X treatment on a next gen console if ran in 1S mode.
I'd be surprised if it wasn't in 1X ballpark power wise, therefore may as well hit it to run as 1X not 1S.

I doubt that uarch efficiency will be enough to brute force 1X emu at 4TF, but I'm not saying has to be >=6

For me it's not about the TF figure that customers wouldn't know or care about. It's about legacy support.
 
As if the customer "knows" the settings of games on the X1X and can really put them in relation to the 1080p product running on the low-end next generation. The faster cpu will also skew FPS comparison most of these sites might deliver or which is relevant for these customers. Most X1X customers will upgrade to the top model anyway so I'm really not getting your point here.
You're right in that sense, except that most of our customers are super informed now. The 0.5TF difference converted to something massive against xbox at the beginning of this gen. Coming in weaker than 4Pro is going to be a tough pill to sell. Not even 3x the power of XBO... i dunno They do their research a lot of things are passed around really quick. Messaging would get out quickly that it's weaker than the last gen X1X and 4Pro. It's a really weird place to be marketing wise if the console is meant to be the base system for next gen. Not to mention super confusing.

The thing is, i'm still not sure what this dual SKU strategy is supposed to be. If Lockhart is the base entry model, it's a damn shame it can't run the last 3 years of 4K enhanced titles. So they would absolutely need to sell it as the 1080p machine. But if we following along that line of thinking, 4TF is going to hamstring any ability of ray tracing, thus, it won't be present. We can see that 6TF is probably around the absolutely minimum we can have for ray tracing. And for it to become a feature for next gen, minimum thresholds need to be somewhat upheld. legacy should be maintained. They've kept a tight ship on this < 30 days now and we still know very little.

I don't know. Honestly, i've read a lot of different viewpoints on the matter and there are patterns that arise often. On one perspective, people seeing Lockhart as this ultra low-powered device and the PS5 and Anaconda is this head to head competition. The other major perspective looks like people believe that the Lockhart and PS5 will be head to head and the Anaconda is this separate beast entirely.

The first perspective has less draw for me as it would appear that there's not a lot of room available for flexibility in price points. $299 doesn't seem realistic. $399 still seems low to me given the specs we've read ($450 is probably the highest the base model can go), but PS5 can't exceed 499 at the maximum because they know what happens when they launch a new gen above $499 and it's going to be bad.

The other perspective has a lot more flexibility. Because the Anaconda SKU that is expected to be purchased less has a lot of wiggle room on the price point so sky is the limit there; they could come in at $599 and no one would flinch at the thought. Provided they have a good baseline SKU between $399-$450.

This is all of course if Lockhart is actually meant to be a baseline console and not some streaming SKU that everyone got wrong.
 
This would make the Anaconda crazy powerful if I attribute a TF to Lockhart.
And I expect Lockhart to be around the 5.5 - 6 TF. Not for marketing reasons, just so it can play 1X games in BC mode.

So Anaconda would have 2 * 32 cu gpu chiplets not 3. So over twice as powerful not 3. Otherwise talking min 15TF. That would be nice but possibly to far.
May as well keep refining this until get some details.
So to make it run at a reasonable tdp.
Lockhart
46 cu @ 950Mhz = 5.6TF 4 disabled 1 chiplet

Anaconda
48 cu @ 1.15Ghz = 7.07 TF 2 disabled * 2 chiplets = 14.14 TF

Cloud
7.07*3=21 TF azure use

Not sure if need to disable 2cu for Anaconda as already got 4 disabled from dies for Lockhart which may be enough for overall yields.

The overall package would be big, but would run pretty cool.
 
May as well keep refining this until get some details.
So to make it run at a reasonable tdp.
Lockhart
46 cu @ 950Mhz = 5.6TF 4 disabled 1 chiplet

Anaconda
48 cu @ 1.15Ghz = 7.07 TF 2 disabled * 2 chiplets = 14.14 TF

Cloud
7.07*3=21 TF azure use

Not sure if need to disable 2cu for Anaconda as already got 4 disabled from dies for Lockhart which may be enough for overall yields.

The overall package would be big, but would run pretty cool.
You won't ;)

I think all the MS leaks are weak. We've been here before. We heard very little about Xbox Scoprio until it was announced E3 2016. I barely recall much in terms of leaks before Scoprio was announced. They had a presentation with limited spec discussion, talking heads but immediately after announcement at E3 dev kits went out and that's when we started getting a lot more news about scorpio specs.

We'll see the same thing here.

recap how our threads went back for mid-gen
https://forum.beyond3d.com/threads/middle-generation-console-upgrade-discussion-scorpio-4pro.57982/

for reference we never got hit with a scorpio spec leak proper until may 25th, just before E3.

though looking back @3dilettante is beast to see this shit from 2016, i mean who writes blade and crosses it out in 2016.
#15
 
Last edited:
You think MS will announce specs this early?

I think we'll get something similar to the PS5 pre reveal announcement, but with a fancy stage and some slightly stilted corporate PR approved mouth sounds.

"Ray tracing", "Super fast storage", "True next-gen", "Unprecedented BC", "Bigly", "So powerful it's in our super computer servers".

Only thing I'm really expecting is if there's hardware ray tracing acceleration and they talk about ray tracing, we'll here the word 'hardware'.
 
May as well keep refining this until get some details.
So to make it run at a reasonable tdp.
Lockhart
46 cu @ 950Mhz = 5.6TF 4 disabled 1 chiplet

Anaconda
48 cu @ 1.15Ghz = 7.07 TF 2 disabled * 2 chiplets = 14.14 TF

Cloud
7.07*3=21 TF azure use

Not sure if need to disable 2cu for Anaconda as already got 4 disabled from dies for Lockhart which may be enough for overall yields.
Side note - You’ll need to disable an equal amount of CUs per shader engine grouping.


———-

Something that’s still nagging me is the rearrangement of shared instruction & scalar data caches that was done for Polaris, where up to 3 CUs shared the I$ and K$ instead of up to 4 in all other iterations, which was done for shortening the pathing/wiring, and I’m not clear if that was similarly done for Vega. The implication for Polaris 10 was that the full chip was more so a multiple of 12 (including the quad SE config) rather than some multiple of 4 per se (pun not intended).

The reason I bring it up (again) is the continued assumption of “64” for a bigly console chip. Perhaps it’s a moot distinction if they are taking the next step with twinning for even shorter paths, more cache per ALU etc. Something something super-SIMD.

zoidblerg
 
Last edited:
I think we'll get something similar to the PS5 pre reveal announcement, but with a fancy stage and some slightly stilted corporate PR approved mouth sounds.

"Ray tracing", "Super fast storage", "True next-gen", "Unprecedented BC", "Bigly", "So powerful it's in our super computer servers".

Only thing I'm really expecting is if there's hardware ray tracing acceleration and they talk about ray tracing, we'll here the word 'hardware'.
Well, the Director of HW at Xbox is tweeting about pre video production makeup as part of his E3 prep, so yeah, I'm betting we'll see something similar to the Scorpio reveal at E3 2016:

 
May as well keep refining this until get some details.
So to make it run at a reasonable tdp.
Lockhart
46 cu @ 950Mhz = 5.6TF 4 disabled 1 chiplet

Anaconda
48 cu @ 1.15Ghz = 7.07 TF 2 disabled * 2 chiplets = 14.14 TF

Cloud
7.07*3=21 TF azure use

Not sure if need to disable 2cu for Anaconda as already got 4 disabled from dies for Lockhart which may be enough for overall yields.

The overall package would be big, but would run pretty cool.

2 CU disabled on 48... That's a 4.1666 % safety margin on Anaconda chiplets with a brand new GPU and lithography...
Microsoft must like taking chances.
And another question... 48 CUs means 3 Shader Engines... That means you must disable at least 3 CUs...
 
Well, the Director of HW at Xbox is tweeting about pre video production makeup as part of his E3 prep, so yeah, I'm betting we'll see something similar to the Scorpio reveal at E3 2016:

It works for them. Keep the leaks tight. Wait for E3 with a half announce and then real devkits go out.

I don’t know what they’ll give us in terms of specs. But definitely want to see the messaging around the 2 consoles and how they are being positioned. Definitely the most important aspect imo
 
2 CU disabled on 48... That's a 4.1666 % safety margin on Anaconda chiplets with a brand new GPU and lithography...
Microsoft must like taking chances.
And another question... 48 CUs means 3 Shader Engines... That means you must disable at least 3 CUs...
You don’t need redundancy if you’re binning your chips. Something to take note of. Not that I think this is he path, but binning solves a big part of the redundancy problem.
 
You don’t need redundancy if you’re binning your chips. Something to take note of. Not that I think this is he path, but binning solves a big part of the redundancy problem.
That was my line of thinking.
The fact that it is a common chiplet, that the Lockhart would get the ones with most disabled cu's to help yields.
Balance between total cu's, disabled, size and yields would need others input. Just throwing out a theory for discussion.

Year ago I considered and ruled mcm and chiplet out for few reasons, but now that we can see that it's the direction for amd in general and is already out there for CPU's. A lot of the reasoning for CPU chiplets hold true for gpu also.
But amd said the issue comes from how it's presented to developers, which isn't an issue for servers from what they've said. I would argue it's also less of an issue for consoles compared to the pc space.

Not against monotheistic as I think it would work well for ps5, but Scarlett has different considerations.

Apart from development considerations what are the other issues?
 
That was my line of thinking.
...
Apart from development considerations what are the other issues?

Power consumption and cost!
You would have two GPUs. Double memory controllers, doubles buses, double infinity fabric, etc. This would increase costs to more than a single GPU with the same CUs.
Besides, all these components draw power, so you would also have a bigger power consumption than the one with a single GPU with the same CUs.
 
Status
Not open for further replies.
Back
Top