I read what he wrote, but don't agree with it. There are no generations anymore for MS. Just bring out a great machine in 2026 for $599 and sell the X for $399 and the S for $199. Then in 2030 (two years after Sony) bring out another machine for $599 and drop the 2026 machine to $399 and the X to $199 and drop the S or keep it alive as the M. Always support 3 (maybe 4) machines. 30 studios putting out 6 AAA games every year. Day One on GP and 2 years later on Nin/PS.
I think MS can shed generations and just have rolling hardware
Zen 5 is later this year and I believe zen 6 would be fall 2026/27. It could actually be ready for Ms earlier depending on a bunch of factors one of which is money lol.
warning here for you since I've seen this behaviour a lot. Please do not post unsubstantiated content here. At least wait until it hits some major journalistic sites before posting here. Tweets are low effort to make and never provide any proof of where they get their info from.
Interesting obersvation. I think the general consensus is you get better performance in terms of latency with unified high bandwidth memory on fixed hw platforms as well as easier development since devs dont have to allocate certain objects or functions to faster memory. This was one of the biggest lessons of the 360 PS3 gen IIRC. Seems the split memory system is a better trade off for PC open systems where you can upgrade the system memory and Video RAM with a new GPU.
I honestly dont see a 2-5x improvement over gaming using an RTX 4090 over a PS5 or Xbox Series X for example. Besides improved frame rate the core visual experience of the fidelity is largely the same. Maybe if you factor in accelerators not present in the current gen systems like dedicated RT blocks you can get better Raytracing. But its largely the same. I dont think a large number of games are fully utilizing discrete GPUs.
Interesting observations, I'm not sure if churning out more hw systems would reconcile with low hw sales and dev complaints about the current hw. Simpler tends to be better. But I would be keen to see how this plays out. Maybe churning out multiple fixed hw configs would come with its own advantages. Maybe some sort of OEM partnership as well.
Yeah and I think that's why they screwed up. I feel sorry for all the work devs have put in without being properly supported financially. I dont know what MS was thinking. But something was off from the way they were describing Gamepass's financial success from the beginning. They weren't very transparent and the wording was very careful. It's a shame because I don't know how Ninja Theory will recuperate their costs.
We are already in the realm of diminishing returns with 10 GB of video memory for the GPU and 2.5 GB for the CPU. Even having something like a 192 bit bus with 20 GB of GDDR7 and 4GB DDR5 for the OS would be plenty.
Gamepass is a long term solution. It effectively locks players into a monthly subscription with a deep library, and destroys the secondary market. People can't sell/trade games if they don't own them! And that means there aren't people buying used copies of your games in 6 months for pennies on the dollar. Remember when people were freaking out when Xbox one launched and it was meant to have single use codes with the discs? This is a more front facing consumer friendly solution to the same problem Microsoft was trying to solve.
Actually I went to visit my young siblings and asked them if they still wanted to use the Gamepass subscription and to my surprise they told me they prefer buying the titles and owning them. Told me they wouldnt mind at all if I stopped paying for their Gamepass subscription. Its just one example but if it turns out to be typifying example of larger sentiment then I think the future of the industry still requires ownership as a major component of gaming. Subscription services are definitely part of that future but likely outright ownership is going to continue being a major part as well. Seems the way we consume games may be a bit more different than how we consume say music or movies as other have suggested here.
I think you could be onto something but will do some research as well. I didn't think memory has reached diminishing returns. Especially considering how much devs have been complaining about Series S memory as well more memory has been a general request by developers.
do you have data to back up that claim? I wonder that 'cos while I think that might be true, MS stopped giving gamepass numbers a long while ago.
I fear for gampeass. Because after what they have done after purchasing of Activision, they could want to buy it to close Xbox business and that there will never be a trace of that name ever again on the face of the earth.
You generally want lower latency for CPU workloads which is why DDR memory is better since it has lower latency than GDDR. But there's another latency that comes from preparing data and sending it between the system and video memory. You'll have lower latency with unified memory since all the memory is in the same physical memory so CPU calls are faster as you dont have to transfer certain data from video to system memory first for example. Its a much more efficient memory subsystem for consoles.
It's a 100% misleading to equate the interleaved unified memory of the Series X to the split memory subsystem you find on most PCs which is separated into system and video memory. On the Series X its 100% unified high bandwidth GDDR6 memory at 14Gb/s per pin so the CPU and GPU have access direct access to the same physical memory unlike what you describe. On the Series X although its unified its interleaved such that 6GB runs at 336GB/s for OS related tasks while 10GB has access to the full bandwidth of 560GB/s for gaming related tasks. But certain gaming related workloads which dont require high bandwidth memory like say audio could still be in the slower memory interleave. But still I have heard there have been issues with the Series X subsystem where certain data meant to run on the higher memory interleave ended up on the slower memory interleave. So Xbox Software APIs to ensure data ended up in the right DRAM module was wonky. Its simply best to have all memory module have access to the full memory bandwidth. I dont think Xbox will be doing this again because it can bottleneck your expensive SoC. They could have have just gotten ten 2GB modules which would be dirt cheap right now instead of six 2GB modules and four 1GB modules as it is today.
I completely agree about the CPU which is why I always take it with a pinch of salt when I hear CPU/GPU/memory increase has reached diminishing returns. We heard the same at the start of the gen how the Zen 2 8 Core CPUs of the new gen wouldnt have any issues for the whole lifetime of the gen. Turns out this wasnt entirely accurate. New advanced tasks requiring faster compute sprung up. And yes you're right about RDNA 5 it could very well be a huge leap.
Yes you're partly right DDR5 has just enough bandwidth as well is low latency by its very nature. I remember Sony as well on the PS4 pro added 1GB of low bandwidth memory for the OS. But they havent done this again I think its for power reasons or something else so they choose what they consider most optimal. A lot of people(including myself) expected them to do the same for the PS5 pro, maybe add 2GB low bandwidth memory for the OS but from all we're hearing it only has 16GB of unified memory not even an addition like last time. I think if it was left to me or you we would propose the split memory but we're not experienced in building these systems. Also leaks suggest due to power budget constraints they were facing a challenge of matching the base PS5's GPU max clock on the Pro. So I think these are things they have to consider which we on the forum dont. So when we bring it back to the Series X or future systems you wont be seeing . The engineers at Sony and MS are not simply using GDDR memory because they dont know these things but because they consider a lot of factors which we miss out on in the forums.
But if you have two busses one for the ddr directly through the cpu and one from the gpu to the gddr ram wouldn't you simply have the best of both worlds. Also wouldn't you have less contention for bandwidth if they aren't sharing the same bus
Yup second generation in a row that the cpu is a huge bottleneck
I am sure its all a numbers game.
No this is not correct there isnt a distinct pool of ram for CPU and GPU they share the same physical memory. I've come to conclusion most people dont know what unified memory is and its benefits. I have heard Moore's law is dead claim Series X doesnt have unified memory as well which is shocking because he's a tech Youtuber. Yes PC memory subsystem works fine but there are benefits of using unified memory especially with closed systems like consoles.
Again there is no distinction between CPU and GPU memory with unified memory architecture. Only difference(A difference MS will likely not repeat) exists on the Series X is there is lower bandwidth interleaved memory on the Series X with 6GB of the GDDR modules having access to only 336GB/s of memory bandwidth but this 6GB of physical memory along with the 10GB of interleaved high memory bandwidth can both be accessed by both the CPU and GPU. You dont have this contrivance on the PS5 because they went with 8 2GB modules on a 256 bit bus at 14Gb/s per pin for 448GB/s so no interleaving to achieve different memory bandwidth for different ram modules. I strongly recommend reading up on the advantages and disadvantages of a unified memory architecture. You get some latency pros and cons as well with going with unified memory. But the overwhelming empirical data supports this approach for consoles.
