That confirms nothing .rumor says memory is cable of handling a console gpu of upto 17 TF .not that it is 17tf.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Next-gen PS5 and next Xbox speculation launch thread - MY ANACONDA DON'T WANT NONE
- Thread starter Phoenix Splash
- Start date
- Status
Yeah, the conversation around what isn't possible for RAM setup and price of the PS5 is basically pointless because we have no way of accounting for those things. What seems too expensive RAM wise may not be as bad for Sony, and loss may not be something they shy away from considering how much better they are doing financially.
Even the PS4 had loss of ~$60 and that was when Sony was flirting with bankruptcy.
I have to agree, so far all of the things that leak have predicted have come true. How lucky do you have to be to get literally everything correct about there reveal we got? Not only the info in the reveal, but also the timing and style of the reveal?
We'll probably get a bit more insight into their goals for this new fiscal year. That may provide something to add to discussion.
The timing is Q2 2019 that could be interpreted in a wide way.
The correct things are also the most vague one:
-8k upscaling (not native)
-8 core Zen 2
-Navi
-Backward Compatibility
The leak talk about flash memory which could've been guessed.
Do you think these potential rumors of 8 and 14 TF, and also raytracing and not having raytracing could be a potential indicator of 2 SKU's from Sony similar to what Microsoft is apparently planning?
A lot people in here are posting Lockhart spec and price both of which are way too high IMO. There has to be a noticeable and strategical difference between Lockhart and Anaconda if you're gonna launch both at the same time (which for the record I don't even think they should do). To me it's a bad strategy to have 2 SKU's that heavily resemble a mid-gen hardware refresh spec at launch. But if MS really is insisting on 2 SKUs Lockhart needs to be clearly distinguishable by the consumer from Anaconda, having them separated by $200 and ~3-4 teraflops simply doesn't do that.
In order to truly go after separate consumers Lockhart should be viewed as a an entry level streaming box perhaps with some basic 'traditional console' capabilities to play games at 1080p..and it shouldn't be a penny over $199. Anaconda would of course be the opposite..most likely a $499 console potentially sold at loss which would be quite the beast depending how much of a hit MS is willing to take per console.
Maybe these are my personal thoughts on what should happen versus what actually will who knows.
In order to truly go after separate consumers Lockhart should be viewed as a an entry level streaming box perhaps with some basic 'traditional console' capabilities to play games at 1080p..and it shouldn't be a penny over $199. Anaconda would of course be the opposite..most likely a $499 console potentially sold at loss which would be quite the beast depending how much of a hit MS is willing to take per console.
Maybe these are my personal thoughts on what should happen versus what actually will who knows.
Software sells. Its proven that exclusives (or the perception of exclusivity) matters. If Lockhart is $399 and PS5 $499 it doesn't automatically mean that the nXbox will sell better by default, at all.
Spider-Man sold something like 9 million PS4 bundles by itself in the first 3 and a half months IIRC.
Spider-Man sold 9M copies (physical and digital) and bundles were included in that number.
Summary: he believes the "LITE" in Navi 10 LITE is not about reduced CU count, but reduced architectural changes.
Upscaled - Using 'chiplets' to build next-gen CPUs
Neat video explaining Chiplets that's easy to understand.
-
This part is pretty cool! https://youtu.be/f-4hxNKvEY0?t=244
"AMD and Intel have both proposed solutions, their approaches are different, but they all revolve around breaking up the CPU into discrete 'chiplets' and connecting them together to form a sort of, 'Voltron CPU'.
Even if the 2020 consoles don't use them, I totally expect mid-gen upgraded refresh consoles to do so, because by 2023/2024, we may only still be on 5nm and not have reached 3nm for mass production, especially for hardware that would go into development in 2020.
-
This part is pretty cool! https://youtu.be/f-4hxNKvEY0?t=244
"AMD and Intel have both proposed solutions, their approaches are different, but they all revolve around breaking up the CPU into discrete 'chiplets' and connecting them together to form a sort of, 'Voltron CPU'.
Even if the 2020 consoles don't use them, I totally expect mid-gen upgraded refresh consoles to do so, because by 2023/2024, we may only still be on 5nm and not have reached 3nm for mass production, especially for hardware that would go into development in 2020.
Last edited:
The highest end performance speculation almost never happens, so until we get some really compelling information that says different, 10-12 teraflops feels realistic. There are plenty of other things Sony can spend finite resources on past just teraflops.
On an unrelated note, that Peter Rubin interview is worth a listen. Some pretty safe things to extrapolation from it:
- Juicy dev-kit leaks are coming very soon and that's part of why the Wired article exists.
- PS5 really is aiming for a higher price point, likely not cripplingly high, but more than $399.
- Sony is holding off telling as much as possible, and are opting to show quite a bit instead, they are just waiting until there is enough to show.
On an unrelated note, that Peter Rubin interview is worth a listen. Some pretty safe things to extrapolation from it:
- Juicy dev-kit leaks are coming very soon and that's part of why the Wired article exists.
- PS5 really is aiming for a higher price point, likely not cripplingly high, but more than $399.
- Sony is holding off telling as much as possible, and are opting to show quite a bit instead, they are just waiting until there is enough to show.
Last edited:
Google translate:
"Navi 10 and Navi 10 LITE are identified, but as long as GFX ID is different, they should be completely different GPUs. I think that the meaning of LITE is not that the number of CU is small, but that the element of Navi is thin in the architectural sense."
Hmm...what do you make of this Anex?
I got all hyped about the pastbin
Because of this video at 9min mark
But after reading the exact quote about the reveal
Pastebin leaker:
"PS5 official info from Sony somewhere around next E3(Sony will not be participating on E3),i'd say Q2 2019 small reveal"
I think I got a bit ahead of myself,
I mean it could be true Sony did do a small reveal in q2 2019, but it was not really a reveal, it was a strange internet article/interview.
Yeah, I got ahead of myself, sorry.
You're missing the point entirely. Until around 18 months into next-gen, there will be NO games made by third parties that are next-gen exclusive, and thus not playable on PS4 and XB1S.
Cross-gen games at the start of a generational transition are a reality of the economics of games development today. You're refusing to acknowledge reality by denying next-gen games will not all initially run on legacy consoles.
It is a hard limit for Vega/Polaris's Front End design - it's not really a 64CU limit, it's a 4x Shader Engine limit. It's not a hard limit for GCN.
The diminishing returns beyond 56CUs of Vega has more to do with memory bandwidth limitations, however, as evidenced perfectly by Radeon 7's performance.
The point of diminishing returns is not a static fixed quantity independent of the specific GPU and memory subsystem design. It's entirely defined by the bottlenecks in the GPU pipeline together with the hard limitations of the GPU architectural design.
So while, yes, increasing the number of SPs per CU for Vega would just shift the problem elsewhere, Navi is a new architecture, and so one would have to assume that the RTG designers in designing the GPU arch for a higher SP/CU ratio, they would adjust other aspects of the GPU microarchitecture to accommodate that change. There are a number of ways AMD can effectively improve GPU throughput and utilisation of the execution resources.
I agree, it should be as low as possible. The whole point of it is to undercut all the advanced stuff going on with PS5 and Snek. If they are around the $499 mark, Lock will benefit by being $200-300 lower than that.
I don't know how you got all that from what I said.
But yes I agree with you.
Thanks folks! I haven't followed the Quest too closely so I honestly wasn't sure. I agree that the pricing seems off here then. If our rumored specs were basically an Oculus Quest without included controllers or built-in processing I'd think $250 sounds reasonable. Basically doubling certain features seems like a $350-$400 headset even if it does depend on PS5.
No FP64 would be one guess.
Anything that is focused towards NN training as well. They may still want to do inference stuff, and they definitely want AI assisted denoising if they're serious about RT.
https://blog.inten.to/hardware-for-deep-learning-part-3-gpu-8906c1644664
So with Gonzalo, can CPUs be told to run at any clock speed between their minimum and maximum? IE, if the 1.6ghz is for PS4 compatibility an the 3.2ghz speed is for PS5 games, can it be told to run at 2.1ghz for PS4 Pro compatibility?
That's an interesting question. Games with Pro support will run as on the Pro or the backward is just for PS4 assets at higher res?
Regarding the 4TF Lockhart rumor, do RT cores impact that number at all, or is it separate? I know we don't really have AMD RT hardware yet, but what about NVidia card specs?
I'm not sure, but maybe it's in this one. http://n4g.com/news/2257993/playsta...ubin-45-mins-podcast-interview-gamertag-radio
Just started listening.
Regarding streaming services like xCloud, wouldn't it make sense to have both Anaconda and Lockhart hardware in their servers? Anaconda can stream to users with 4K enabled devices and Lockhart could stream to users with 1080p and lower devices. It doesn't make much sense to dedicate a 4K high end session to a device that can't exceed 1080p.
That reasoning alone is what had me pondering if Sony could be going for 2 SKUs as well.
That reasoning alone is what had me pondering if Sony could be going for 2 SKUs as well.
It has to do with the GPU front-end. Basically there are things called shader engines. These feed tasks to the CUs. Basically in the GCN there are 4 of them, each only able to effectively support 16 CUs. Adding More will require a ar bigger redesign of the entire architecture.
It seems that rather than redesign said front-end in this iteration f GCN, coming from vega to navi what AMD may have focused on is achieving higher efficiency. Allowing them to clock a lot higher while reducing power draw. In a manner of speaking, 1.8Ghz could very well be to navi what 800mhz was to Southern islands (code name of GPU range the OG PS4 belonged to).
This is how much power you chip can safely use before the heatusing that power kills it.
Practically, its indicative of the capacity of heat dissipation your cooling system must account for. The higher the number the better the cooling system you will need, larger the PSU you will need, bigger the console you will need and higher the overall system draw of the console.
I agree, and I've said as much. 3 SKU's make no sense.
Make the Lockhart the Streambox, and price it cheap.
Hopefully, Sony can do a better implementation of BC in PS5 than they did on Pro making this a moot point. I think there are even patents indicating such. It's much better if you handle BC more in software, with some hardware support, than having it be primarily hardware based. It allows you more flexibility.
Just finished reading the whole thread, I saw a lot on TFs but I don't think anything was answering this:
Can anyone explain the red RAM counts to me? I get that there are potentially split allocations & types of RAM but I am unclear on the roles for the allocations.
I've seen 20GB of the 24 being usable by developers for games and 4 for OS before but here I am seeing;
So what is going on with the HBM2 8GB? How does that get utilised theoretically vs the DDR4? And what does that get mean for devs vs how is it exposed to them?
Also can anyone give a refresher on the green line - what was "wrong" with the memory solution in PS4? I thought the shared mem pool fixed a lot of the problems the PS3 had? Is it a bus/latency thing?
Sorry if already covered.
Can anyone explain the red RAM counts to me? I get that there are potentially split allocations & types of RAM but I am unclear on the roles for the allocations.
I've seen 20GB of the 24 being usable by developers for games and 4 for OS before but here I am seeing;
24GB == "8GB in form of 2x 4-Hi stacks of HMB2" + "additional 16GB in form of DDR4"
"additional 16GB in form of DDR4" == "4GB reserved for OS" + "12GB usable by games"
"additional 16GB in form of DDR4" == "4GB reserved for OS" + "12GB usable by games"
So what is going on with the HBM2 8GB? How does that get utilised theoretically vs the DDR4? And what does that get mean for devs vs how is it exposed to them?
Also can anyone give a refresher on the green line - what was "wrong" with the memory solution in PS4? I thought the shared mem pool fixed a lot of the problems the PS3 had? Is it a bus/latency thing?
Sorry if already covered.
Last edited:
Yeah. The Xbox One S is always cheaper than the PS4 and Sony is outselling MS by more than 2:1.
I Microsoft has the cheapest and more powerful consoles on the market at this time and things are only getting worse, could even be 3:1 worldwide this early 2019 months
If I m not mistaken ddr4 will be split between cpu and gpu but hbm2 will be all given to gpu.gpu need much faster ram compared to cpu.ddr4 is good enough for cpu.
He never predicted specs correct ? Just that ps5 is a 4k 60 fps monster in his leak
HBM2 would be used almost exclusively by the GPU and only data that was going to require the high bandwidth of that memory would need to be stored there. All other data would exist in the slower pool and data would be moved from the larger, slower, pool to the smaller, faster, one as needed.
There's nothing wrong with a unified memory setup, and in fact it has some advantages. It may just be at this time, with the current state of technology, having a split pool offers more advantages.
Person who leaked many info about ps5 and said Sony is skipping e3 before Sony confirmed they r skipping e 3
I wonder if we might get a little of Cerny talk at Computex as one of Lisa Su's special guests at the keynote?
https://www.computextaipei.com.tw/en_US/news/info.html?id=6994382A4DFCD609
If there is any truth in AMD making Navi for Sony it might make some sense for Mark to be there?
The HBM is used solely for games. Of the remaining 16GB DDR4, 4GB is reserved for the OS.
HBCC (high bandwidth cache controller) is the intelligence that manages what's in the HBM versus DDR4 for games. The leak says devs have the option to override this.
The problem with the PS4 memory solution was that as the CPU demanded more bandwidth, it decreased the bandwidth to the GPU by more than what the CPU was taking. Basically, as the CPU reads more data, the chance it's reading a bank the GPU wants to access goes up, meaning someone has to take priority.
With this solution, the chance of that happening goes way down. The DDR4 alone would have the same number of banks as the GDDR5 in PS4, and the HBM would have 512 banks for very granular access.
Small .that's around 15% .Abit of resolution /fps at Max with next gen engines
So just a couple of months ago Shawn Layden did a keynote at the D.I.C.E summit where he noted that Sony learnt the painful lessons with PS3 and its many failures with price being the worst thing.
So they now plan a $500-$600 system? Makes sense!
"There is no concrete date for the first unveiling of PlayStation 5 has been outlined yet. But as per the source, Sony will possibly go for April 2019 to showcase the console for the first time ever."
Seems like he was spot-on. This was written in November
If Lockhart exists to sell to the $299 consumer, that consumer may not be the one with the internet service to support a streambox. They may want physical media they can trade in for instance.
I agree. A 2 TF difference wouldn't be that much given a 10-12 TF baseline.
Yet that would likely not be the case between Anaconda and Lockhart, if the difference between them is something like Lockhart having 1/2 or 1/3 the GPU performance, i.e. 12 vs 6 TF or 12 vs 4 TF or 14 TF vs 7 TF. Lockhart might not be able to run next gen engines with the same feature set as Anaconda. It's not just a difference between 1080p and 4K, unless they hold back Anaconda's potential in fidelity (graphics complexity).
- Status