• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Microsoft claims 200GB/s of bandwidth, gets caught fudging numbers

http://arstechnica.com/gaming/2013/...xbox-ones-internals-while-disclosing-nothing/

Similarly, the company claimed that there was more than 200GB of bandwidth within the system. Again, the number had no context or clarification and if rumors are to be believed, it suggests some rather creative accounting: 68 GB main memory bandwidth, 102GB bandwidth to an embedded SRAM buffer for the GPU, and 30GB bandwidth between the CPU and GPU. While that does add up to 200GB, there are no two parts of the SoC that can communicate with each other at 200 GB/s. The fastest link is believed to be the GPU read performance, which can aggregate across the main memory and SRAM buffer for 170 GB total.

I have two cars, one goes 80mph and the other goes 90mph... I have 170mph cars!
 

Dennis

Banned
Don't worry, Microsoft.

That point will go over the head of 90% of gamers and 99% of games journalists.
 

Klocker

Member
[The fastest link is believed to be the GPU read performance, which can aggregate across the main memory and SRAM buffer for 170 GB total.

so you CAN add those two together for GPU read speed... cool
 

Striek

Member
Why would they even claim 200GB/s. Thats way down on what they claimed last generation (this is from their PR release in E3 2005):
EiyQVAi.gif
 
so you CAN add those two together for GPU read speed... cool

Yes for 32 MB of it...

As soon as you have more than 32MB of data that needs a decent bandwidth you are screwed.

My cpu has 6MB of l3 cache for specific tasks that has over 350GB/sec bandwidth , that doesn't mean the ram bandwidth is 356GB/sec
 
so you CAN add those two together for GPU read speed... cool

For the infinitesimal amount of time it takes you to exhaust 32MB at that rate, then you run out of data so you have to move more in which takes as much bandwidth as just reading directly from DDR3 so you might as well do that instead of setting up pointless aggregated memory sprints.
 
so you CAN add those two together for GPU read speed... cool
"Aggregate"

"An aggregate is a collection of items that are gathered together to form a total quantity."

It's not the same as full read speed from point b to point a. It's like getting point d, c, b to point a, each at diff speeds, and calling the sum the total read speed.
 
Top Bottom