More blogs in Arm Community blogs
Tell us what you think
Tags
Actions
Related blog posts
Related forum threads

Launching Mali-T658: "Hi Five-Eight, welcome to the party!"

Edvard Sørgård
1 minute read time.

Just a day ago we've lifted the veil off the ARM® Mali™-T658, the second Mali GPU based on the Midgard architecture that we launched a year ago. It's been brewing in our labs next to the Mali-T604 and already has key leading licensees, but until now hasn't enjoyed the public presence of its sibling. Now they are both ready to power the leading end-user devices of 2012! Yes you hear right - expect to see them both in stores next year.

So how does the Mali-T658 complement the Mali-T604? By showing off some of the versatility of the Midgard architecture it brings in a compute punch of up to 350 GFLOPS and over 5GPixel/s real fill rate to external memory to power high-end mobile devices with Visual Computing and Augmented Reality, the 4k DTV revolution, and Exascale High Performance Compute. It's another fantastic component in the joined-up computing story from ARM.

You can still find every part of the unrivalled feature set of the Mali-T604 in the Mali-T658, including the true multicore capability allowing partners to choose the number of cores to implement making one product release from ARM equal a whole roadmap of products from others.

I must say I'm proud that our team has been able to develop and deliver these masterpieces; their rigorous focus on our vision and quality and execution has put Mali in an enviable position ahead in time, quality, reputation and performance. You make my dreams come true!

If you can't make it to one of our upcoming Technical Symposia events, join us online to find out more about our latest and greatest Mali GPU.  We'll be holding a webinar entitled "Harnessing the power and flexibility of the Mali Midgard architecture"& on 15 November, at 10am PST and you can register here. 

Anonymous
  • Offline
    My main question is regarding optimizing for efficient data transfer between the GPU compute and GPU based functions. While it should be possible to, say, do surface tessellation in OpenCL to create strips for vertex shading, is there a way to minimize out-of-cache reads/writes on large mesh data? I suppose I want to know how effectively OpenCL can interact with the OpenGL pipeline?

    Also, I have a following on Google+ and blogger contacts, and if you're interested in some added publicity for the T658, I can orchestrate a light interview and hopefully get a very big viewership. Let me know! I completely understand if you would rather not for whatever reason.
  • Offline
    Please, feel free to ask your questions. We're happy help, worst case it just takes a bit of time to get around to it.
  • Offline
    Thank you very much for the confirmation. I am a huge fan and so grateful that yourself and ARM share information with the public!

    I have one or two other questions but I will respect your timeline as I'm sure you are very busy.

    Thanks again!
  • Offline
    This is fantastic news and quite an impressive part. Can you confirm that the GFLOPS rating is correct at 350? I was under the impression that it would be closer to 250.


    [font="Arial"][size="3"]Hi Sean,[/size][/font]
    [font="Arial"] [/font][font="Arial"][size="3"]Yes, 350 GFLOPS is correct when used with the latest 28nm ARM Physical IP for mobile. Remember the actual hardware implementation is done by our partners and will vary depending on their technology and optimization choices, so when ARM quotes numbers we need to do so at “mainstream”, easy to understand technology points we have qualified in order for our partners to make predicable estimates for their own implementations.This can sometimes be confusing for the general public who like to deal with absolute numbers. E.g. your 250 GFLOPS can still be valid, but under different conditions (say for example for use in a slower, aggressively area optimizedapplication). The nice thing for the performance savvy is that fast is in vogue at the moment[/size][/font][font="Arial"][size="2"].[/size][/font]
  • Offline
    This is fantastic news and quite an impressive part. Can you confirm that the GFLOPS rating is correct at 350? I was under the impression that it would be closer to 250.
Graphics, Gaming, and VR blog