orm

16 ms·

1. 31.▲
   Ravi Rajwar: Specialized Evolution of the General – Purpose CPU [pdf](cidrdb.org)

   1 points by orm 11y ago|0 comments
2. 32.▲
   by orm 11y ago

   A lot of matrix concepts sound a lot more natural to me if I first think of what they mean in terms of vector spaces. For example, I think one can derive associativity of matrix multiplication less tediously from the observation that every
3. 33.▲
   by orm 11y ago

   I think one reason would fault tolerance. Is there a fault tolerance layer in GNU parallel? last time I checked their homepage ( a few minutes ago), there was no reference to fault tolerance. Another reason is, perhaps, scheduling.
4. 34.▲
   by orm 11y ago

   Thinking about it again, from my calculation below, if those packets are coming from RAM then you are bound to, say, 25GB/s incoming, 25GB/s outgoing. Since Haswell L1 cache seems to allow a peak bandwidth of 64 byte loads + 32 b
5. 35.▲
   by orm 11y ago

   New to this, but basing such calculation on purely on CPU capacity would imply a bandwidth to memory (or some cache) 64 B / inst * 2 (read and writeback) * 3 inst / cycle * 3 Gcycle/sec > 1000 gigabytes/sec, which
6. 36.▲
   by orm 11y ago

   Some things about this article bother me. It starts by describing what is objectively a problem ascribable to Amazon itself ( letting on-call rotations reach the 3 week mark is terrible ), and then descends to pure appeal to emotion. Here&#