- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Guys,
I need some reasonable explanation for this before I shoot an SR to Altera. I'm using Quartus 12.1, the greatest supposedly. I have exactly the same design, written in behavioral code, no primitive whatsoever. It's rather small: 1800 ALMs. I compiled for Arria II GX, speed grade C5: the fitter took 2 minutes, the speed achieved is ~165MHz I compiled for Arria V GX, speed grade C5 : the fitter took 24+ minutes, the speed achieved is 128MHz. What is going on here? They don't have good algorithm to fit Arria V or what? Please help me, this doesn't make any sense or I have wrong expectation for Arria V? Thank you, JeffLink Copied
5 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Did you get the same size after fitting for each design ?
If your design uses block ram , are you sure that quartus inferred them correctly for arria 5? If not Quartus may use LEs instead of block ram, leading to long run time and poor lower timing performances.- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
--- Quote Start --- Did you get the same size after fitting for each design ? If your design uses block ram , are you sure that quartus inferred them correctly for arria 5? If not Quartus may use LEs instead of block ram, leading to long run time and poor lower timing performances. --- Quote End --- Thanks. I haven't checked the size but I don't use blockram at all. The settings between compilations are the same. Maybe it's worthwhile to have a look. I'll check it again today. Anyone having the same weird result, please discuss!
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
--- Quote Start --- Thanks. I haven't checked the size but I don't use blockram at all. The settings between compilations are the same. Maybe it's worthwhile to have a look. I'll check it again today. Anyone having the same weird result, please discuss! --- Quote End --- The ALM used by Arria V is significantly less than Arria II (1400 compared to 2000). I guess it is because the number of inputs of Arria V LUT is wider. But still, i don't know why the critical path delay is more in Arria V :( The core has a wide datapath of 256 bits. :D ... it that rings a bell in you.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I think what you are seeing is the effect of that Arria V is built using a LP process. It is the same process as used on Cyclone V. The Arria II was built using the high performance process, which is the same as Stratix III. This is probably why you see the discrepancy. However, power use for the Arria V should be much better than that of Arria II for the same circuit.
Hope that helped- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
--- Quote Start --- I think what you are seeing is the effect of that Arria V is built using a LP process. It is the same process as used on Cyclone V. The Arria II was built using the high performance process, which is the same as Stratix III. This is probably why you see the discrepancy. However, power use for the Arria V should be much better than that of Arria II for the same circuit. Hope that helped --- Quote End --- Wow, thank you !!!! That's eye-opening.
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page