Xbox One (Durango) Technical hardware investigation

Status
Not open for further replies.
What I find very sad is that for many is it easier to cling to hopes and believe in the existence of a "super secret sauce" for Durango that nobody ever saw, rather then accept that Durango is not a bad machine at all.

Exactly...

Some people are to use to see MS make powerful machines,that they don't want to believe MS will not go for power this time,and will play it safe.

Either way most multiplatform games would look the same,and lack of power doesn't mean a product will fail.
 
For the rumored 68GB/s, they have to be using 2133. If they overclocked the GPU to the same frequency, it would become a 1.6TF part. But it would screw royally with any thermal dissipation designs they've made. I doubt they'll do it.

2133 or half of 2133?

I think the following is the only possible stuff that they can do in the time frame.

Free up some of the rumored 3GB reserved for OS. 500mb or more would be a good start.
Overclock the cpu and gpu to 1.8ghz and 900 mhz respectively (I think the CPU and GPU should be able to handle that). You'll have a ~15% performance increase across the board for little to no extra cost.
 
Last edited by a moderator:
I wonder how this will work out for them.. hopefully its not called XBOX RT

There will probably be a subset of apps (think PS minis) that will have requirement to run across all their metro-ish devices (Windows rt tablets, windows phone) including this low cost set-top box thing. Its only a matter of time before apple opens up ATV for apps and then the war is really on.
 
For the rumored 68GB/s, they have to be using 2133. If they overclocked the GPU to the same frequency, it would become a 1.6TF part. But it would screw royally with any thermal dissipation designs they've made. I doubt they'll do it.

2133 or half of 2133?

I think the following is the only possible stuff that they can do in the time frame.

Free up some of the rumored 3GB reserved for OS. 500mb or more would be a good start.
Overclock the cpu and gpu to 1.8ghz and 900 mhz respectively (I think the CPU and GPU should be able to handle that). You'll have a ~15% performance increase across the board for little to no extra cost.

Wouldn't it be cheaper and "safer" long term to take the hit on yields and activate redundant CUs instead?
 
my bet is that the 2nd "Xbox" is the Surface gaming tablet that was leaked by TheVerge, just like Sony is using Vita for remote play gaming

While incredibly unlikely, it would be amusing if Microsoft released a tablet that was basically an Xbox in a tablet. :D Unfortunately, it would be difficult to use Kinect 2.0 with that. :D

Regards,
SB
 
2133 or half of 2133?

I think the following is the only possible stuff that they can do in the time frame.

Free up some of the rumored 3GB reserved for OS. 500mb or more would be a good start.
Overclock the cpu and gpu to 1.8ghz and 900 mhz respectively (I think the CPU and GPU should be able to handle that). You'll have a ~15% performance increase across the board for little to no extra cost.
Frequency for 2133 DDR3 is 1066 MHz. There are GCN cards running at that frequency today, or at least within spitting distance. The 7970 gigahertz edition is at 1050MHz.
 
why couldn't this system basically run the newest games for the next 2-3 years on par with PS4, build on the social, kinnect and media features and then in a couple years add full game streaming, effectively negating the need for OMG! hardware?
 
You can't just change the amount of eDRAM (aka 1T SRAM) without a costly respin (new masks, 3-4 months fab throughput for 28nm etc), new package design, new PCB design etc. Unless of course it's a seperate die like Nintendo has done as a SiP or MCM. Still on-chip would be nice ;-)

I don't think they will use DDR3 above 933 MHz (1866) (official specs are even 800 MHz). AMD APUs mobo's can handle e.g. DDR3 2133, but it's basically overclocking. Would effect their return rate I guess.

Personally I thing the eSRAM is 6T SRAM after all going down from an Orbis level gpu to the one in Durango you give up 6 CU and 16 ROPS (but perhaps tellingly keeping triangle and tessellation the same). That's what 80-90 mm2 of die space? Going by Jaguar's L2 cache 32MB of SRAM is about 120 mm2.

Also 2133 is the highest DDR3 standard from JEDEC.
 
why couldn't this system basically run the newest games for the next 2-3 years on par with PS4, build on the social, kinnect and media features and then in a couple years add full game streaming, effectively negating the need for OMG! hardware?

Internet connection bandwidth and the fact that you still need the hardware at the other end to run the games. The more power the game requires to run the more power you need at the server side to run it - hence more expense. I'm not saying they couldn't do it but there are definitely barriers to entry which may not make it worthwhile.
 
why couldn't this system basically run the newest games for the next 2-3 years on par with PS4, build on the social, kinnect and media features and then in a couple years add full game streaming, effectively negating the need for OMG! hardware?

Unless Microsoft reinvents the Internet in the next 2-3 years, then I'd say not happening.
 
Also 2133 is the highest DDR3 standard from JEDEC.
Which means nothing ms can buy ddr3 ram at any speed the fab can produce it. Jedec has nothing to do with it. Of the foundry wants to rub cheetah blood on the ram to get the chips to run at 10 ghz and sells them to ms jedec wont care or be involved
 
Frequency for 2133 DDR3 is 1066 MHz. There are GCN cards running at that frequency today, or at least within spitting distance. The 7970 gigahertz edition is at 1050MHz.

Can the memory architecture handle 25% more GPU horsepower? It seems like there would be some diminishing returns at some point if you just scale the GPU without redesigning the whole system.

Currently the GPU seems really similar to a GTX 650 TI, but would have ~3.5 billion transistors instead of 2.5 billion due to the ESRAM. Despite the low flops ratings, I am excited to see what >3.0 billion transistor GPU can do in a closed system.
 
Personally I thing the eSRAM is 6T SRAM after all going down from an Orbis level gpu to the one in Durango you give up 6 CU and 16 ROPS (but perhaps tellingly keeping triangle and tessellation the same). That's what 80-90 mm2 of die space? Going by Jaguar's L2 cache 32MB of SRAM is about 120 mm2.

Also 2133 is the highest DDR3 standard from JEDEC.

From my post earlier in this same thread:

Scaning around the internet it seems that 64MBit SRAM chips have been produced on TSMCs 28nm process with a cell size of 0.127mm2/Mbit.

For standard 6T-SRAM that's 32.5mm² for 32MB. And for 1T-SRAM it would be 6mm².

On 28nm 32MB of eSRAM would be 33mm² for 6T, and about 6-10mm² for the 1T variant.

Nowhere near 120mm².
 
Which means nothing ms can buy ddr3 ram at any speed the fab can produce it. Jedec has nothing to do with it. Of the foundry wants to rub cheetah blood on the ram to get the chips to run at 10 ghz and sells them to ms jedec wont care or be involved
No they can't. If someone sells DDR3 chips that are supposed to work with DDR3 controllers the Jedec standard is involved. And it stops at 2133.
Otherwise it's not DDR3, it's either a custom chip, or an overclocked and overvolted part that's used out of specs.
 
What I find very sad is that for many is it easier to cling to hopes and believe in the existence of a "super secret sauce" for Durango that nobody ever saw, rather then accept that Durango is not a bad machine at all.

Yeah, I don't really think it's "Sad", I think that it's called "Not knowing all of the facts".
Many people are willing to make huge judgments without knowing all of the facts and I find that to be sad.

Jumping to conclusions and not having up-to-the-date facts is what I find to be "Sad"

If the next Xbox turns out that it's under-powered and we get the facts, then that is the final say, otherwise you can't know what you don't know.
 
Can the memory architecture handle 25% more GPU horsepower? It seems like there would be some diminishing returns at some point if you just scale the GPU without redesigning the whole system.

Currently the GPU seems really similar to a GTX 650 TI, but would have ~3.5 billion transistors instead of 2.5 billion due to the ESRAM. Despite the low flops ratings, I am excited to see what >3.0 billion transistor GPU can do in a closed system.
Just as well as the PS4's memory can handle 32 ROPS (theoretical max of 204GB/s, memory: 176GB/s). But the truth is I don't know. The eSRAM in the vgleaks docs is listed as being able to transfer 128 bytes per clock, and it seems to share the GPU clock, so theoretically, if you overclocked the GPU to 1GHz, it would automatically increase the eSRAM max throughput to 128GB/s

I still think it's a long shot and my guess is the MS will just hold the course.
 
No they can't. If someone sells DDR3 chips that are supposed to work with DDR3 controllers the Jedec standard is involved. And it stops at 2133.
Otherwise it's not DDR3, it's either a custom chip, or an overclocked and overvolted part that's used out of specs.

Show me where it ssys so , I can already purchase non jedec standard ram... so it must be magic huh.
 
Status
Not open for further replies.
Back
Top