Rumoured Switch 2 specs - via @Zuby_Tech

[Plextorage]

100GB/s full duplex means you can send 100GB/s and receive 100GB/s at the same time, so the total combined bandwidth is 200GB/s.
GDDR only either sends or receives.

Bullshit. Xbone's ESRAM 200GB/s is every 8th cycle. It would be 200GB/s IF it was EVERY CYCLE, but it was every 8th. So, theoretical Xbone's ESRAM speed is 102GB/s and NO MORE than that. Btw. PS4 RAM speed theoretical is 176 GB/s and it also has read/write cycles ( hUMA, Onion+Garlic bus ), but it is not every cycle. If you are applying the same for Xbone's ESRAM, same can be said for PS4 GDDR5 RAM.

Lol WTF am I reading here? No one said anything about the Series X. The XBOne had to resort to embedded RAM to compensate for using DDR3 instead of GDDR5. It's a strategy that failed because Sony ended up managing to put 8GB GDDR5 in there and use a much wider GPU by not spending die area with ESRAM.

You are reading it very well. It is deja vu again! Same shit in 2013 and same shit in 2024. Stop with a spin.
C'mon, will you say how cloud exponentially increases Xbone power by 12x or how DX12 will increase Xbone GPU performance by 20%? Shit, gap will be closed. Finally!!! :/

 

[Alabtrosmyster]

That is the reason 99% of the games were lower than 1080p in resolution… in fact most games where 720p compared to 1080p on PS4.

Switch Pro doesn’t need to go over 720p due DLSS..: it can even go lower like 540p and be fine with DLSS upscaling it to 720p/900p.

After all I don’t think the Swith Pro screen will be 1080p… most likely it will be around 720p.

Probably, it depends how the chip works and how much cache it has.

But I will not argue about target resolutions just yet... However going from less than half the resolution (1\4th the pixels) to 1080p is not good at all, even with DLSS... 540p will still look like a blurry mess.

720p to 1080p I can see this kind of work.

I don't understand the Switch success, neither will I get it if the SW2 was to be a hit.

 

[Alabtrosmyster]

Plus 32MB EDRAM providing it 100GB/s duplex (200GB/s effective).

The Xbox One's EDRAM is the precursor to Infinity Cache.

The ESRAM was too small to fully compensate for the slow RAM speed of the xbone, it was better than if it had nothing at all, but it did not make things as good as they could be.

This is like saying your old spinning drive that had a few MB of RAM based cache was as fast as an SSD because the SSD has RAM cache as well... Cache is great, but it gets saturated very fast.

From memory, a game like Killzone Shadow Fall used 145MB just for its frame buffer, no way a few MB of cache allowed anything close to that to happen near 1080p on the xbone.

All the spin has been debunked back in 2013... no need to reash that.

Those numbers assume "best case" scenarios, most of the time the cache is busy doing something else or the CPU\GPUs are working on tasks where this theorhetical speed is not relevant.

 

[Alabtrosmyster]

You are reading it very well. It is deja vu again! Same shit in 2013 and same shit in 2024. Stop with a spin.
C'mon, will you say how cloud exponentially increases Xbone power by 12x or how DX12 will increase Xbone GPU performance by 20%? Shit, gap will be closed. Finally!!! :/

If you stream a Black Myth Wukong from a PC that has a 4090 in it you can get an xbone to run games with ray tracing, DLSS, frame generation and what not!! at INSANE resolutions!

 

[ToTTenTranz]

Bullshit. Xbone's ESRAM 200GB/s is every 8th cycle. It would be 200GB/s IF it was EVERY CYCLE, but it was every 8th. So, theoretical Xbone's ESRAM speed is 102GB/s and NO MORE than that. Btw. PS4 RAM speed theoretical is 176 GB/s and it also has read/write cycles ( hUMA, Onion+Garlic bus ), but it is not every cycle. If you are applying the same for Xbone's ESRAM, same can be said for PS4 GDDR5 RAM.

Dude, will you stop pretending you know something that goes completely over your head?

The ESRAM bandwidth is full duplex (read+write) during 7 out of 8 cycles, at the 8th cycle it can only do a read. It can do 15 out of 16 operations and that is already taken into account in the 204GB/s number presented in the documentation. Otherwise it would be 218GB/s.

You are reading it very well. It is deja vu again! Same shit in 2013 and same shit in 2024. Stop with a spin.
C'mon, will you say how cloud exponentially increases Xbone power by 12x or how DX12 will increase Xbone GPU performance by 20%? Shit, gap will be closed. Finally!!! :/

What spin?! The cache in the XBone is there to compensate for the lower bandwidth on DDR3 vs. GDDR5. Sony ended up with the faster GPU because without that cache they could use more die area on more execution units. That's just facts.

No one said anything about cloud or DX12. Are you delirious? What the hell is this persecution complex?
I'm not even saying the XBone is better than the PS4, I'm saying why the PS4 is faster than the XBone.

Those numbers assume "best case" scenarios, most of the time the cache is busy doing something else or the CPU\GPUs are working on tasks where this theorhetical speed is not relevant.

No, the impact of Infinity Cache is very effective. The hitrates in those graphics are probably very good estimates made by AMD.

RX 6900XT: 256bit GDDR6 16GT/s for 512GB/s VRAM bandwidth (+ 128MB IC)
RTX 3080: 320bit GDDR6X 19GT/s for 760GB/s VRAM bandwidth
RTX3080: 384bit GDDR6X 19.5GT/s for 936GB/s VRAM bandwidth

4K results at release date:





FYI Nvidia did a similar approach for the RTX40 Ada GPUs, but they did massive increases on the L2 instead of creating a 3rd level cache.

 

[Plextorage]

100GB/s full duplex means you can send 100GB/s and receive 100GB/s at the same time, so the total combined bandwidth is 200GB/s.
GDDR only either sends or receives.

From anandtech:







The shock,

Dude, will you stop pretending you know something that goes completely over your head?

The ESRAM bandwidth is full duplex (read+write) during 7 out of 8 cycles, at the 8th cycle it can only do a read. It can do 15 out of 16 operations and that is already taken into account in the 204GB/s number presented in the documentation. Otherwise it would be 218GB/s.

View attachment 5592
View attachment 5591






What spin?! The cache in the XBone is there to compensate for the lower bandwidth on DDR3 vs. GDDR5. Sony ended up with the faster GPU because without that cache they could use more die area on more execution units. That's just facts.

No one said anything about cloud or DX12. Are you delirious? What the hell is this persecution complex?
I'm not even saying the XBone is better than the PS4, I'm saying why the PS4 is faster than the XBone.




No, the impact of Infinity Cache is very effective. The hitrates in those graphics are probably very good estimates made by AMD.

RX 6900XT: 256bit GDDR6 16GT/s for 512GB/s VRAM bandwidth (+ 128MB IC)
RTX 3080: 320bit GDDR6X 19GT/s for 760GB/s VRAM bandwidth
RTX3080: 384bit GDDR6X 19.5GT/s for 936GB/s VRAM bandwidth

4K results at release date:

View attachment 5593



FYI Nvidia did a similar approach for the RTX40 Ada GPUs, but they did massive increases on the L2 instead of creating a 3rd level cache.

No, it is not a full duplex and it is not 200 GB/s

Using a beyond3d forum? Really?

How about this quote by Baker and Goosen back in 2013 :

"Theoretical peak performance is one thing, but in real-life scenarios, it's believed that 133GB/s throughput has been achieved with certain ESRAM operations.

The same discussion with ESRAM as well - the 204GB/s number that was presented at Hot Chips is taking known limitations of the logic around the ESRAM into account. You can't sustain writes for absolutely every single cycle. The writes is known to insert a bubble [a dead cycle] occasionally... one out of every eight cycles is a bubble so that's how you get the combined 204GB/s as the raw peak that we can really achieve over the ESRAM. And then if you say what can you achieve out of an application - we've measured about 140-150GB/s for ESRAM.

"That's real code running. That's not some diagnostic or some simulation case or something like that. That is real code that is running at that bandwidth. You can add that to the external memory and say that that probably achieves in similar conditions 50-55GB/s and add those two together you're getting in the order of 200GB/s across the main memory and internally."

So 140GB-150GB is a realistic target and DDR3 bandwidth can really be added on top?

"Yes. That's been measured."

You see. Every 8th cycle. And it is not even a 200GB/s when it hits, it is less.

No one said anything about cloud or DX12. Are you delirious? What the hell is this persecution complex?
I'm not even saying the XBone is better than the PS4, I'm saying why the PS4 is faster than the XBone.

I've said it because we are back to square one with Xbox spin. You've mentioned esram with 200GB/s effective ( which is false), that's why I've added bullshit about DX12 and cloud.

 

[ToTTenTranz]

You see. Every 8th cycle. And it is not even a 200GB/s when it hits, it is less.

"A bubble every 8 cycles" is exactly what you just quoted. The other 7 work with read+write simultaneously.


You can't even comprehend what you just read, and then complain about the explanations from actual engine architects and GPU hardware engineers from B3D.

 

[Plextorage]

"A bubble every 8 cycles" is exactly what you just quoted. The other 7 work with read+write simultaneously.


You can't even comprehend what you just read, and then complain about the explanations from actual engine architects and GPU hardware engineers from B3D.

one out of every eight cycles is a bubble so that's how you get the combined 204GB/s as the raw peak that we can really achieve over the ESRAM

ONE of every 8th cycle is how you get 200GB/s.

 

[Alabtrosmyster]

Dude, will you stop pretending you know something that goes completely over your head?

The ESRAM bandwidth is full duplex (read+write) during 7 out of 8 cycles, at the 8th cycle it can only do a read. It can do 15 out of 16 operations and that is already taken into account in the 204GB/s number presented in the documentation. Otherwise it would be 218GB/s.

View attachment 5592
View attachment 5591






What spin?! The cache in the XBone is there to compensate for the lower bandwidth on DDR3 vs. GDDR5. Sony ended up with the faster GPU because without that cache they could use more die area on more execution units. That's just facts.

No one said anything about cloud or DX12. Are you delirious? What the hell is this persecution complex?
I'm not even saying the XBone is better than the PS4, I'm saying why the PS4 is faster than the XBone.




No, the impact of Infinity Cache is very effective. The hitrates in those graphics are probably very good estimates made by AMD.

RX 6900XT: 256bit GDDR6 16GT/s for 512GB/s VRAM bandwidth (+ 128MB IC)
RTX 3080: 320bit GDDR6X 19GT/s for 760GB/s VRAM bandwidth
RTX3080: 384bit GDDR6X 19.5GT/s for 936GB/s VRAM bandwidth

4K results at release date:

View attachment 5593



FYI Nvidia did a similar approach for the RTX40 Ada GPUs, but they did massive increases on the L2 instead of creating a 3rd level cache.

You just don't get it... sure cache is nice and it's effective, but it only gets you so far.

You kind of made my point when you mentioned that they INCREASED cache sizes drastically, if it's too small it gets saturated too fast to have its full effect.

32MB was not enough, even back then, the number that went around was 128MB (which it what Intel used for some of their high end laptops iGPU, that did not last).

Intel Adds Crystal Well-based Skylake-R Processors: 65W with 128MB eDRAM

www.anandtech.com

I'm not saying it made no difference, but in the end the 200GB\sec bandwidth number is marketing gugely gook buzz word salad for those with a big imagination to think that will make the bone outclass the PS4 in some situations.

 

[ToTTenTranz]

You just don't get it... sure cache is nice and it's effective, but it only gets you so far.

You kind of made my point when you mentioned that they INCREASED cache sizes drastically, if it's too small it gets saturated too fast to have its full effect.

32MB was not enough, even back then, the number that went around was 128MB (which it what Intel used for some of their high end laptops iGPU, that did not last).

Intel Adds Crystal Well-based Skylake-R Processors: 65W with 128MB eDRAM

www.anandtech.com

I'm not saying it made no difference, but in the end the 200GB\sec bandwidth number is marketing gugely gook buzz word salad for those with a big imagination to think that will make the bone outclass the PS4 in some situations.

The 32MB made a difference but its usage wasn't automated like it is for Infinity Cache. The devs had to manually write in those addresses and choose what data should get in there, which took a long time (and therefore money).

Things got worse as the gen progressed and Xbox software sales collapsed, so devs stopped optimizing for the cache altogether. That's why some late XBone ports had piss poor performance.


It's not about 32MB being too little. The RX 7600 XT uses 32MB IC to achieve PS5 performance despite 288GB/s VRAM bandwidth, and so will the Strix Halo premium APU with 40 CUs RDNA3.5.

 

[Alabtrosmyster]

Things got worse as the gen progressed and Xbox software sales collapsed, so devs stopped optimizing for the cache altogether. That's why some late XBone ports had piss poor performance.

Nah, this is because these games got more and more complex and all the holes in the xbone architecture became more apparent.