also the DDR3 can came from AMD
http://www.tomshardware.com/news/AMD-Radeon-Memory-Gamer-2133mhz,22499.html
http://www.tomshardware.com/news/AMD-Radeon-Memory-Gamer-2133mhz,22499.html
- Status
GrimThorne
Newcomer
So the general consensus here is that Microsoft shoved esram and move engines into their new console so that they could achieve nothing?
Yeah that makes alot of sense
No, the general consensus on this board is that those features were added, in part, to compensate for a lower than ideal bandwidth from main RAM, and also to reduce latency so that the GPU stalls less on complex calculations. (Note, I'm just stating the consensus as I've seen it crystallize on the board. I'm not making any pronouncements)So the general consensus here is that Microsoft shoved esram and move engines into their new console so that they could achieve nothing?
Yeah that makes alot of sense
I think the concensus was that MS added an elaborate DMA engine and ESRAM to compensate for the lack of bandwidth of DDR3. Otherwise the 12CU would have been starving. OTOH, if they'd have enough bandwidth from GDDR5, the ESRAM is questionable expense.So the general consensus here is that Microsoft shoved esram and move engines into their new console so that they could achieve nothing?
Yeah that makes alot of sense
I wouldn't characterize the consensus as saying an SRAM pool would have no effect.
I would say that if we assumed all of the rumored specs were true, except for the SRAM, there would probably be a very sizeable performance deficit.
What is being debated, until we know more, is if the two different directions taken for the consoles are different routes to generally equivalent end points.
I would say that if we assumed all of the rumored specs were true, except for the SRAM, there would probably be a very sizeable performance deficit.
What is being debated, until we know more, is if the two different directions taken for the consoles are different routes to generally equivalent end points.
So the general consensus here is that Microsoft shoved esram and move engines into their new console so that they could achieve nothing?
Yeah that makes alot of sense
I agree.
As R&D from microsoft, started in 2005, after 8 years of intense development, give us useless, costly pieces of silicon that achieve nothing if not try to catch a standard solution desktop solution as PS4's GPU is, a simple 18 cu gpu poorly customized gpu with gddr5 (shock) from AMD
very believable.
So the general consensus here is that Microsoft shoved esram and move engines into their new console so that they could achieve nothing?
Yeah that makes alot of sense
So clearly you haven't read any of the threads discussing this. There is a fairly well held view that ESRAM is there to make up the bandwidth defficiency of the DDR3 which itself was used because of the large amount of memory Microsoft wanted to put in the console without going to the expense of using GDDR5. The move engines are then required to help data move around between the memory pools efficiently (amoungst other things).
There is also a view that the low latency of esram *might* have some appreciable effect on graphics performance but I don't think anyone (with genuine knowledge of the subject) is really expecting it to be huge or to make up anywhere near the raw throughput difference between the two consoles. If it does have any difference at all, I expect it's more of a beneficial side effect than a key driver for the design choice. I've little doubt that if Microsoft could have picked up fast GDDR5 for the same price they are paying for the DDR3 then Durango would have had a similar memory setup to PS4.
Technically feasible with 24x4Gbit chips or 12x8Gbit chips.
Economical feasibility is another thing altogether (i.e. die size constraints then yields, cost per chip, mobo wiring, memory chip binning).
upnorthsox
Veteran
Custom MC too as I don't believe anyone has an IP for a 384bit DDR3 solution.
so Microsoft choose
8/16 GB DDR3 30$-60$
eSRAM 30+ $
Move engine blocks don't know how much they cost in R&D and to add in silicon, around 20$?
for a range from 80 to 120+ $
adding to this complex development to internal software tools and to developer's engines and code
trying to catch 100-105 $ of simple 8 GB GDDR5, nothing less, nothing more?
it's no cheaper, where's the sense of it?
nobody has the doubt that maybe the things are different?
8/16 GB DDR3 30$-60$
eSRAM 30+ $
Move engine blocks don't know how much they cost in R&D and to add in silicon, around 20$?
for a range from 80 to 120+ $
adding to this complex development to internal software tools and to developer's engines and code
trying to catch 100-105 $ of simple 8 GB GDDR5, nothing less, nothing more?
it's no cheaper, where's the sense of it?
nobody has the doubt that maybe the things are different?
Last edited by a moderator:
No, the suggestion is that Microsofts design is a cheaper way of achieving a console with 8GB of memory that is not bandwidth restricted than Sony's way. But it comes at the expense of computational resources. This might have been a good sacrafice if Sony had gone with 4GB of memory as expected. That's why the 8GB PS4 announcment was such a big deal.
Are you including R&D costs or making up numbers?
I'm not certain that the GDDR5 chips are going to go that high in price, or for long if they are that high.
For Durango, I don't know how you can affix such a large dollar amount specifically to die area of a monolithic chip, or how a few mm2 of move engines of all things would cost anything near that.
eSRAM adds to the cost now, but after a shrink or two it will cost a lot less. 8GB GDDR5 will in all likelihood be always more expensive than the Durango DDR3+ESRAM setup over the lifetime of the consoles.
the ram and esram was taken from an analysis on component of durango, I've read it on those boards, but for move engine blocks I've included R&D cost because they are not separated in production from the rest
the important thing is that this is not a cheaper solution than simple gddr5 (I've read somewhere that the cost is no 105 but sub 100$, but 10$ don't change the whole idea)
how can the fantastic R&D from microsoft, with a lot of engineers from IBM and AMD, in 8 years give the born to a such stupid thing that try goofy to reduce the speed differences from a simpler solution that costs the same?
this can be true but it's not an argument so solid to take the decision to complicate the life fo developers and power of the console for no advantages
unless GDDR5 and 18 CU on ps4 means a price higher than 150-200$ per console, and this is not the case in my opinion
Last edited by a moderator:
That reasoning is flawed, proven by simple example. Sony invested a gazillion bucks in an esoteric hardware that performed no better than MS's much cheaper, simpler solution. Likewise, NASA invented a million dollar space-pen to achieve what the Russians used a 20 cent pencil to do.
There has been ample discussion on the custom hardware explaining how they very efficiently add improvements to performance in a small way. Durango's focus has apparently been on price and perhaps low-power, which would be a choice fitting a business model that isn't focussed on More Power. Your reasoning is too reliant on sweeping generalisations and made up numbers to prove anything. It's like Wii U, with all that RnD, and all Nintendo's engineers came up with was something about as powerful as PS360 - but it did so at a much lower price and power draw. With no knowledge of MS's business strategy, it's pure guesswork what their targets are for Durango and how well they've met them. And that's another discussion, an evaluation of the hardware once we know it.
Do you know the R&D expended solely on the move engines and eSRAM?
It's a convoluted relationship between up-front engineering costs and the expected lifetime cost of design decision.
For a device that sells tens of millions of units, a lot of money can be spent up front to save dollars per unit.
For the sake of argument, if a console sold 70 million units, you could spend up to $70 million up-front if it saved a little over a dollar per unit. The claimed differences in the cost of RAM are larger than that.
The move engines are not a high-science concept, and on-die memory isn't a new concept.
Says who?
So we have other speculations.
If my numbers (they are not mine, except the cost of move engines) means nothing, care to prove that durango is focused on price?
and for the low-power, it is not a portable console, so who matter if it is a 90-120W machine and what is the useless advantage over a 150-180W machine as could be a far more powerful console?
if they reveal 12-16 GB of DDR3, the whole ram+eSRAM will cost more tha 8 GB gddr3, don't you agree?
More or less this must be it ... they wanted a lot of ram without having to pay for expensive GDDR5 so cheap DDR3 was fine and with esram they could lessen the impact from low bandwidth . It could work pretty good if PS4 stayed at 4GB.
The 8GB GDDR5 announcement must have been a slap in the face for them ..
Your made-up math alone leaves a 20-90 dollar per-unit gap in RAM costs.
Are you sure there's not some wiggle room to fit R&D in the production savings across 70 million units?
I originally brought that theory up toying around of ways to improve leaked specs.
-the bw came out to 102GB/s, equal to the leaked ESRAM spec.
-losing 2-3 GB of memory for OS function was not as impacting.
-keeps memory advantage
-more information per frame, 204GB/s total system BW.
But Im not an engineer just an armchair forum poster. I have no idea what it takes to swap out memory controllers or the resulting board complexity for mfg.
I think I also said ESRAM was to small (40-64), and need more CUs.
- Status
Similar threads
- Sticky