SemiAccurate Forums  

 
Go Back   SemiAccurate Forums > Search Forums

Showing results 1 to 25 of 500
Search took 0.02 seconds.
Search: Posts Made By: aaronspink
Forum: CPUs 04-15-2014, 06:10 PM
Replies: 109
Views: 2,811
Posted By aaronspink
You can roll your eyes all you want, I'll stick...

You can roll your eyes all you want, I'll stick with H&P, 40 years of nomenclature, and commonly accepted practice thanks.
Forum: CPUs 04-15-2014, 06:00 AM
Replies: 109
Views: 2,811
Posted By aaronspink
Its really quite simple, just like you cannot be...

Its really quite simple, just like you cannot be a little big pregnant you cannot be a little bit OoO. You are either OoO or you are not. You viewpoint/belief is completely at odds with accepted...
Forum: CPUs 04-14-2014, 01:08 AM
Replies: 109
Views: 2,811
Posted By aaronspink
yeah. Basically, a throughput oriented...

yeah.

Basically, a throughput oriented design has to have specific design elements that effectively provide latency tolerance. In effect, the hardware to be considered throughput orientated needs...
Forum: CPUs 04-13-2014, 09:27 PM
Replies: 109
Views: 2,811
Posted By aaronspink
In fact, none of these are really correct. ...

In fact, none of these are really correct.



Phi can already do standalone bootable mode, fyi.
Forum: CPUs 04-13-2014, 09:22 PM
Replies: 109
Views: 2,811
Posted By aaronspink
You ever hear the line about being a little big...

You ever hear the line about being a little big pregnant?
Forum: CPUs 04-13-2014, 09:17 PM
Replies: 109
Views: 2,811
Posted By aaronspink
Phi -> likely going OoO GCN -> not enough...

Phi -> likely going OoO
GCN -> not enough information to make a determination unless you work for AMD and are posting non-public info
Nv cores -> not enough information to make a determination...
Forum: GPUs 04-11-2014, 05:40 PM
Replies: 139
Views: 10,349
Posted By aaronspink
Cacheline sizes are pretty set from a usage graph...

Cacheline sizes are pretty set from a usage graph point. For the vast majority of workloads, going to larger line sizes results in lower performance.
Forum: GPUs 04-11-2014, 01:11 PM
Replies: 139
Views: 10,349
Posted By aaronspink
an ARM core does not a CPU make.

an ARM core does not a CPU make.
Forum: CPUs 04-11-2014, 01:08 PM
Replies: 109
Views: 2,811
Posted By aaronspink
If my marketing calls red blue does that make a...

If my marketing calls red blue does that make a red light a blue light?



You keep saying that, I do not think it means what you think it means
Forum: GPUs 04-11-2014, 01:03 PM
Replies: 139
Views: 10,349
Posted By aaronspink
Limited availability, significantly more...

Limited availability, significantly more expensive (no really), not as much performance advantage as claimed (PD-SOI was also suppose to be the best thing ever ever), etc.

Its also worth pointing...
Forum: GPUs 04-11-2014, 12:49 PM
Replies: 139
Views: 10,349
Posted By aaronspink
CPUs generally take longer post first silicon to...

CPUs generally take longer post first silicon to reach a stable and tested state, even for demos than GPUs. CPUs are directly running actual code and have pretty hard requirements, GPUs however are...
Forum: GPUs 04-11-2014, 12:47 PM
Replies: 139
Views: 10,349
Posted By aaronspink
Actually, 2k/4k is generally sub-optimal. Once...

Actually, 2k/4k is generally sub-optimal. Once you've passed the cacheline size, you go rapidly down hill in efficiency. And even approaching the cacheline size runs into various efficiency issues....
Forum: CPUs 04-11-2014, 12:41 PM
Replies: 109
Views: 2,811
Posted By aaronspink
This is also false. A throughput optimized core...

This is also false. A throughput optimized core is a core designed around computation. It can and likely will have OoO hardware. And while perf/watt is important, it cannot be looked at in...
Forum: CPUs 04-11-2014, 12:33 PM
Replies: 109
Views: 2,811
Posted By aaronspink
Easiest way I know how to explain it in relation...

Easiest way I know how to explain it in relation to HPC is that HPC applications generally have two aspects, local computation and global computation/communication. Local computation is pretty much...
Forum: CPUs 04-11-2014, 12:16 PM
Replies: 109
Views: 2,811
Posted By aaronspink
Nvidia's diagrams generally have as much...

Nvidia's diagrams generally have as much reference to actual reality as a video game.






Their *MARKETING* might think otherwise... I'm pretty sure their engineers would agree with me.
Forum: CPUs 04-11-2014, 12:08 PM
Replies: 109
Views: 2,811
Posted By aaronspink
That's not big-little though. Its latency...

That's not big-little though. Its latency optimized and throughput optimized cores on a monolithic die which is completely not big-little.
Forum: CPUs 04-11-2014, 12:06 PM
Replies: 109
Views: 2,811
Posted By aaronspink
Except that A53 isn't a throughput optimized core...

Except that A53 isn't a throughput optimized core at all.
Forum: CPUs 04-11-2014, 03:48 AM
Replies: 109
Views: 2,811
Posted By aaronspink
There is BS, blatant BS, lies, and then GPU core...

There is BS, blatant BS, lies, and then GPU core counts...
Forum: CPUs 04-11-2014, 01:38 AM
Replies: 254
Views: 27,739
Posted By aaronspink
This is actually a fallacy. The issue really...

This is actually a fallacy. The issue really isn't core size, its perf/mm and perf/watt. The reality is that the big cores that currently exist tend to be generalist cores and are able to handle a...
Forum: CPUs 04-11-2014, 01:31 AM
Replies: 254
Views: 27,739
Posted By aaronspink
Most HPC apps break down into local compute and...

Most HPC apps break down into local compute and global compute. Local compute is all about throughput and if the application can be broken down into lots of threads, throughput optimized...
Forum: CPUs 04-11-2014, 01:25 AM
Replies: 254
Views: 27,739
Posted By aaronspink
If makes zero sense for HPC. It would be an...

If makes zero sense for HPC. It would be an absolute nightmare and a performance disaster. For HPC, you basically want binary power management, full on and full off. I think a lot of people fail...
Forum: CPUs 04-11-2014, 01:20 AM
Replies: 254
Views: 27,739
Posted By aaronspink
And if you understood anything about it, you...

And if you understood anything about it, you would know that no one has gotten it close to really working. And if you knew anything about the history of hardware power management, you'd know that...
Forum: Article discussion 03-20-2014, 10:01 PM
Replies: 985
Views: 97,245
Posted By aaronspink
yes the cryengine that CIG have been working with...

yes the cryengine that CIG have been working with for over a year with. They are one of the major drivers of new features for the next cryengine. They are willing to push the technology to the...
Forum: Article discussion 03-20-2014, 10:00 PM
Replies: 985
Views: 97,245
Posted By aaronspink
What hacking an engine? CIG have been working...

What hacking an engine? CIG have been working hand in hand with Crytek on the next generation cryengine for well over a year. CIG's SC has been one of the lead development drivers for the next...
Forum: Article discussion 03-19-2014, 06:57 PM
Replies: 985
Views: 97,245
Posted By aaronspink
this is actually OLD OLD OLD news. Star Citizen...

this is actually OLD OLD OLD news. Star Citizen which is using Cryengine has stated they would have mantle support for quite a while.
Showing results 1 to 25 of 500

 
Forum Jump

All times are GMT -5. The time now is 11:21 PM.


Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2014, vBulletin Solutions, Inc.
SemiAccurate is a division of Stone Arch Networking Services, Inc. Copyright 2009 Stone Arch Networking Services, Inc, all rights reserved.