Do you remember that leak from pastebin? It has almost the same amount of memory as Reddit leak - 24 gb gddr for games and 4 gb ddr for system. Maybe Sony got that hbm deal very recently so they change that.
-
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.
Next-gen PS5 and next Xbox speculation launch thread - MY ANACONDA DON'T WANT NONE
- Thread starter Phoenix Splash
- Start date
- Status
Do you remember that leak from pastebin? It has almost the same amount of memory as Reddit leak - 24 gb gddr for games and 4 gb ddr for system. Maybe Sony got that hbm deal very recently so they change that.
You are making a mistake in thinking Sony not cutting the price mean they don't care about cost cutting on there end.
We know that PS4 was made with that in mind but it selling so well that Sony just riding the profit train .
Them keeping PS4 price so high more about sales and profit but cost cutting would be very on the PS5 list of things.
Imagine if we get another '8GB GDDR5'... wouldn't it be possible that the HBM2 is just a 'placeholder' and the final system will have dual stack HMB3?
Twice the capacity, much higher bandwidth, more energy efficient, and comes in at a lower cost than HBM2.
One of the reddit rumors said that Sony wants 32GB of RAM in the system but would settle for 24GB if the price is too high. Going double stack HBM3 would raise the capacity to 16GB + 16GB DDR4.
I guess the biggest problem with this would be securing supply for the consoles, but it's something to consider. Of course this all depends if the HBM rumor is even true...
Edit- also, this could be a reason why Cerny call the current dev kits they're sending out "low speed" versions. Switching from HBM2 to 3 for the final version would increase the memory bandwidth significantly.
Last edited:
It is just possibility not a mistake. I've previously posted about how Sony love to iterate and cost reduce. Just that with them being able to 'ride the profit train' this gen it might change slightly how they go about things for next-gen?
As far as multiplataform games goes we ll always have one lower common denolinator so I dont see PS5 getting hold back either.
Or at least not sigfinicantly.
Specially because the difference between Lockehart and PS5 will be smaller than the difference between Lockehart and Anaconda.
But when it comes to 1P games ? Now thats where games shine in visuals almost everytime and with that in mind having your own anchor may not be a good thing on the long run...
Multiplataform games and Xbox 1P games will have on Lockeheart as their lower common denominator.
Sony 1P games wont.
Well if they get another gen like this they will want the cost cutting even more.
So the can ride the profit train again .
If things don't go as well they still want the option to be able to cut PS5 if they need to stay competitive .
They first year that don't be a problem but later on is when you want that so it works out best for them either way .
It would be impossible to tell if that's not just developer talent.
In the ps3/360 gen a lot of people cited ps3 exclusives as proof of its technical superiority to the 360, but it's a mute argument because we will never know what those devs could do with 360 hardware.
So for those who do not understand much. What would be the great benefit of using HBM2 + DDR4 instead of just using DDR6? Can HBM be allocated only for GPU or something like that?
And another question, why DDR4 instead of DDR5? HBM2 + DDR5 simply does not work or the price would be out of the reality?
And another question, why DDR4 instead of DDR5? HBM2 + DDR5 simply does not work or the price would be out of the reality?
PS4 success (vs XB1 failure) was mainly possible because of the idea to use a fast (but not too fast) unified memory pool (from the dev perspective) so it allowed a beefier GPU, etc.. In the end PS4 was actually the most balanced console, specs, easy for devs, even cheaper to make than XB1, etc. I think unified memory pool is still probably the most important spec asked by devs today (well before super fast loadings obviously :P).
So now suddenly PS5 could have a memory architecture very similar to XB1 ? (Yes 8GB HBM + 16 GB very slow DDR4, just a bit faster than XB1 DDR3, is exactly that: XB1 2.0). After all those years of threads discussing and mostly agreeing upon this ? I know that the simple mention of HBM make it look sexy and all, because it could be the memory of the next gen (PS6 and XB?).
But those specs are exactly that: XB1 2.0. The biggest problem of the XB1 was not the esram, which was fine, it's the DDR3 which was way too slow for 2013 games: 68 GB / s particulary shared with the CPU with contention.
What are developers are going to do with 16 GB of 100 GB/s ram shared with hungry Zen 2 (so not much faster than XB1 DD3) in 2020 and after ? Store the audio ? And 400 GB / s for the main bandwidth is too slow. They can get more than 600 GB / s by using GDDR6, while having 3 times more ram (24 GB).
So now suddenly PS5 could have a memory architecture very similar to XB1 ? (Yes 8GB HBM + 16 GB very slow DDR4, just a bit faster than XB1 DDR3, is exactly that: XB1 2.0). After all those years of threads discussing and mostly agreeing upon this ? I know that the simple mention of HBM make it look sexy and all, because it could be the memory of the next gen (PS6 and XB?).
But those specs are exactly that: XB1 2.0. The biggest problem of the XB1 was not the esram, which was fine, it's the DDR3 which was way too slow for 2013 games: 68 GB / s particulary shared with the CPU with contention.
What are developers are going to do with 16 GB of 100 GB/s ram shared with hungry Zen 2 (so not much faster than XB1 DD3) in 2020 and after ? Store the audio ? And 400 GB / s for the main bandwidth is too slow. They can get more than 600 GB / s by using GDDR6, while having 3 times more ram (24 GB).
The last two. The SoC will burn less power talking to memory, and the advanced packaging should help thermals on top of that.
_____________________
AMD have posted many of their GDC talk PDFs here:
https://gpuopen.com/gdc-2019-presentations/
What's stopping them from just switching to HBM mid-way through cycle? Maybe cost / benefit isn't quite there right now but in 3 years time ala X1 to X1X?
GDDR6, there is no DDR6. GDDR6, is a bandwidth-optimized derivative of DDR4. There is no production DDR5, and the standard isn't quite finished yet, but should be this year. DDR4 will be far cheaper than DDR5 and not much slower though.
HBM2, meanwhile is practically an order of magnitude faster than GDDR6 and AMD has this fancy tiered memory technology called HBCC that can tier together different speeds of RAM and fast SSD storage into a single pool of active memory for a GPU, and without much performance loss over just having an amount of the fastest kind of RAM as big as your pool
So, with HBCC, 8GB HBM2 and 16GB DDR4 with, say, 256GB of high-speed SSD cache wouldn't actually perform all that much slower than 24GB of HBM2, and would give you 280GB of total effective GPU memory, unless you're doing some specific task that needs truly random access at full bandwidth (which is almost nothing). There are definite limits to this (you're not keeping 280GB of on-screen textures, sorry), but it may enable some crazy open world game designs among other things. This is essentially how the Radeon Pro SSG worked.
Last edited:
Thanks! And in terms of performance, what would be the big difference between combining HBM2 with DDR4 and using GDDR6 only?
If the buffer between the power difference and the resolution difference is big enough, developers should never reach this point. Will 900p games happen? I'm guessing yes. Will it cause another resolutiongate? I don't think so. One S games still hit 900p and 720p all the time but no one cares anymore because the X is kicking everyone's butt. All the people that care about "gates" will look at is the difference between the PS5 and Anaconda, if Microsoft can make the PS5 look bad compared to the Anaconda they are good in terms of "gates". a 200$ cheaper console that is aimed for 1080p drops to 900p? Doesn't seem like a big deal. Sony's flagship doing 1800p while Anconda does 4k? The gods of the internet will implode.
When you say 35%-40% higher costs, do you mean vs GDDR6?
The question is, if the PCB allows it, if It's 8GB HBM2 + 16GB DDR4 vs 16GB GDDR6 or if it's 8GB HBM2 + 16GB DDR4 vs 8GB GDDR6 + 16Gb DDR4. If the PCB does allow this setup, according to the leak, 8GB GDDR6 + 16GB DDR4 will not only be cheaper, it will be faster too (448GB/s). The only question is how many GDDR6 and DDR4 chips can fit on the PCB and if they can achieve a split setup with GDDR6.
Nice, I was reading about HBCC the other day and it looks like it would solve the "problem" that the Xbox One had in 2013, as Andromeda pointed out in a post above.
That sounds optimistic ;) Separation memory pool for gpu and cpu is possible as Sony said "CPU bandwidth reduces GPU bandwidth disproportionately"
I found the Forza 7 + Horizon argument a compelling one for your stance. Using ports farmed out to external studios, in several cases months after the launch of a game's initial launch just wasn't really convincing to me in terms of clear evidence.
What gives me pause to think that a minimum base line of GPU performance can have an effect on the overall design of a graphics engine is the case for the launch of the current-gen systems. UE4 used an entirely real-time lighting engine, including SVOGI, and advertised game development that involved no baking what so ever before the specifications of the current-gen systems were fully made known. Once the baseline performance of current-gen systems was revealed, UE4 underwent massive changes, with many of its biggest advertised features not showing up until years later, if at all.
Regardless, my thoughts aren't based on a full knowledge of what really went on behind the scenes, so the Forza examples work for me.
It would have helped, but I doubt it could have totally covered the gulf between DDR3 and GDDR5. SRAM of any kind is the fastest kind of memory there is, bar none, but only 32MB of it just isn't enough for HBCC to really help.
And that may well be what they decide to do, it's just that Mark Cerny's emphasis on the SSD in game performance sounded an awful lot like an HBCC configuration. We probably won't know until, at least, the PS5's full reveal.
The idea is that nothing is bandwidth starved. The problem on PS4 was that cpu would sometimes suck bandwidth from GPU or vice versa choking the other one off. When you don't have sufficient bandwidth to feed something you are not getting its full potential. It's like having an exotic sports car and putting in standard gas instead of the required premium. Sure it'll run, but it's performance (horsepower) will be impacted. Same thing with a bandwidth starved cpu or GPU. Now interestingly....personal computerscomputers (all sorts, not just gaming) use dedicated video ram (hbm, gddr5, gddr6) for graphics cards and ddr4 to fuel the cpu. So in a way it's closer to a PC. But some differences are that both the ddr4 and hbm2 can be used to the fuel the four of needed. And with HBCC to a developer it looks just like one unified pool of RAM even if in reality it's not. It's a quite a clever solution
Yes, vs. GDDR6.
The alternative you're proposing is essentially a 512-bit memory interface.
Ahh, so in other words 32 GB HBM3 confirmed for PS5? :)
Kidding, kidding... it indeed does sound like a good solution.
What is the difference otherwise? You said "in Sonys case"
Cerny directly spoke on that in his presentation "The Road to PS4" and I suspect they won't split it like that. It's not going to have HBM imo. What everyone keeps suggesting with all of these rumors and ideas goes against the entire concept that helped build the PS4.
Simple architecture with straight forward development.
- 8/16 core/thread Zen 2 1.6Ghz base 3.2 GHz boost
- 10-12TF Navi with little or no hardware specific raytracing capabilities with built in hardware audio capabilities
- 24 GB shared GDDR6
- 4+ GB/s Reads (maybe lower writes) PCIe 4.0 NVMe 1TB M.2 drive (user upgradable)
- Same unified OS as PS4 with some PS5 specific options enabled for use with that hardware
Possibly some internal VR stuff that's irrelevant to overall power/ development
This is a classic chicken and egg conundrum. For prices of HBM to ever drop significantly they need OEMs placing orders in volume. OEMs won't do that unless prices drop. If sony/ms start making orders for their next consoles, they literally will need millions of those stacks each year. Thats probably in the range of 20 times whatever order of HBM are right now. More companies will have to step up to tap into that kinda demand, which in turn forces prices down.
Or they end up using HBM3. Wilbe 4GB to 8GB GDDR5 all over again. Funny enough HBM3 seems more likely, as its due to be read in 2020 and its primarily supposed to be cheaper to make. An even if they use the cheapest version of it, you are looking at 717GB/s of bandwidth. The cheapest HBM3 is almost twice as fast as GDDR6. And thats assuming they use 16GB of GDDR6. But if they use 8GB of GDDR6 then ts even worse because they max out at a 128Bit bus. In that case even the most expensive chips of GDDR6 will not them only 320GB/s of bandwidth.
You couldn't be more wrong if you tried.
TSV is a hard manufacturing problem that's why HBM has so few producers. Those top-end enterprise GPUs like the Tesla V100 or Google's TPU are incredibly lucrative and if it was worth their while, I have no doubt they would produce as much HBM as the enterprise players want. Can Sony or MS make it happen? I'm not sure.
And as far as you know, it's hbm 2 and not hbm 3? I'm just asking because some people have brought that up as a possibility and I feel it's best to head off wild speculation before it goes too far. Because the logic for hbm3 does seem to be there
No doubt HBM will be key in datacenters like Stadia? Azure and AWS will likely follow suit... I am sure there is a decent demand already without consoles.
I really don't get what makes anyone think sony doesn't push boundaries when it omes to hardware design.
If sony were being simple they wouldn't have used GDDR5 for the PS4. Oh.... PCIe 4 NVMe drives will give you significantly more than 4GB/s speeds. If its a poor drive it should still give at least 6GB.
And the whole HBM thing, I don't know if its real or not, but what I know is that if they do it it would be because that was the most efficient way for them to build their console and it would be either heaper upfront or in the long term. I guess if all that happens that will be the new keeping it simple.
If that were the case then HBM would be cheap by now. Just know that with all tech, adoption force prices down.
The most recent Anaconda rumour indicates that it'll be 'clearly' more powerful than PS5. If that was true which I don't necessarily believe for now then maybe lockhart isn't that catastrophic in terms of raw power?
Besides that I agree with you overall.
The rumor clearly stated HBM2 and the speeds quoted match HBM2.
what was the quoted speed? do you have a link to the original post? I somehow missed it.
Thanks for the clarification. I wonder what is meant by "slower devkits" then. Cerny is very deliberate and exacting with his language,
Probably means PCIe 3.0 vs 4.0 when that becomes available later in the year?
https://reddit.com/r/PS5/comments/bdfarm/_/ekxsjlv/?context=1
1.7Gbps per pin for a total of 435GB/s
I think people are reading too much into this. It's just a non-final spec.
Will HBM3 even be ready for 2020? Anyway, so the benefit of have 8 GB of HBM2 is that there will never be a case where the CPU can access that memory, thus no bandwidth contention? Is HBCC granular, efficient, and/or fast enough to manage (and gate off) memory accesses between the CPU and GPU? Will the CPU only see 14GB of ram, while the GPU will see 20GB?
Exactly what anexanhume said, clockspeeds are definitely not final, and possibly major parts of the hardware design. Last gen was just like this too.
That would make sense yes.was also wondering about those two amd Gonzalo samplings. Maybe they're sending out kits with specs closer to the final stage Gonzalo than the one with GPU supposedly clocked at 1gig
The rumor said "try to keep CPU access to DDR4" which sounds like adjustments to how HBCC manages the memory, not some impenetrable wall between the pools.
The ship on HBM3 has sailed. Different interface that would require a die spin.
It's the fact that it's low that makes it believable. If they had quoted the latest 2.4Gbps speed it would rightly draw criticism for needing only the best performing parts.
Frankly the 3200MT/s DDR4 should give some pause there.
I just realized I should have asked if the PS4 super slim is using GDDR6. The quoted die size dropped 52%, which makes sense for a 16nm to 7nm transition.
This part of the article where Peter Rubin says this:
Makes me think it is more than a PC-based devkit. Even if not final hardware it should be getting there?
so they are making a worse pc than? the one advantage a console has that lets it punch above a pc, is its share ram pool. not having to sync things between vram and ram is a big plus, if normaly just pc brut forcing power making it iralavent.
This is one contiguous memory space. There is no synchronization or replication of data such as you'd see in a cache hierarchy. HBCC manages just one address space.
From a technical point of view I would prefer 8 Gb of HBM3 and 16 GB of DDR4 but from a business point of view of the rumors is true this is good.
The 4GB/s number is being very conservative in even the earliest tests done from earlier this year. They tested at 4.5 GB/s reads. I'd imagine it'll be faster obviously especially if users go with a more advanced Samsung EVO or equivalent later down the line.
I'm not saying they don't push boundaries, all I'm saying is that they do what's best for developers and themselves. That system spec I mentioned earlier effectively abolishes any learning curve for developers moving up from PS4. Its just vastly more power.
Having some exotic HBM3/GDDR combo or internal NAND with a slow spindle drives can create obstacles. Unless we are seriously suggesting they are going to have an high performance storage I/O option AND 16GB of HBM3 for GPU use and GDDR for OS use. If that's the case then sure 599.99 or higher it is.
Their loss is going to come from the SSD implementation imo. 16GB HBM3 + the SSD is asking way more than 100 USD at 499.99 assuming 100 dollar loss.
If it's 16GB HBM3 + SSD + Zen2 + Navi 10 I'm going to be all over it for 499.99 but I don't see it.
One can dream I guess. If that's what they announce then I can't really imagine MS doing anything better tbh.
- Status