Sadly, VIA hasn’t been too precise regarding performance estimates compared to the desktop competition (Sempron/Brisbane/Conroe-L), however given the overall architecture we would expect IPC to be comparable to AMD’s pre-Barcelona 65nm chips, if not higher (especially in floating point workloads ironically). Here’s a brief summary of the key features in the architecture:

  • Two integer units that handle all operations but multiply and divide (+slightly asymmetrical).
  • Two 128-bit media ports, one handles additions and misc. SSE operations, the other is dedicated to multiplications (including integer multiplications, which means Isaiah can indeed run 2 ADD+1 MUL per clock). Divisions and cryptographic hardware is on the first port, both fully pipelined.
  • One load port, one store address port, and one store data port. 16 loads may be outstanding at one point, and the latency is 4 clocks (->L1). The store data unit also supports 16 stores and includes a write combining buffer of 6 lines, each 64-byte.
  • Supports memory disambiguation that is superior to Barcelona’s and theoretically comparable to Penryn’s (but the predictor’s efficiency in both cases is unknown). Supports prefetching like other modern CPU cores.
  • Supports not only micro-op fusion, but macro-op fusion which even Barcelona doesn’t have. It can decode 3 x86 instructions resulting in up to 7 micro-ops. The branch predictor is also claimed to be very advanced (and large on the die shot!)
  • L1 cache is 2x64KiB 16-way, which is insane. The L2 cache is also 16-way associative (same as Conroe, twice that of Conroe-2M) and exclusive. Data prefetching uses a separate buffer, unlike competing architectures.
  • Optimized not only for peak power, but also average power through a variety of mechanisms including two distinct PLLs (can change power state while continuing execution) and temperature-based voltage adjustments.

Conclusion

Isaiah is a very impressive architecture that will allow them to compete head-on with the big boys in the low-end of the market, something they couldn’t really do with C7. Future iterations are also very promising, with Centaur publicly insisting time and again that many things are specific to this first release and that they’ve learned a lot from it.

While that might also imply IPC might be lower than you’d expect it to be right now, it does bode well for the architecture and company in general as Centaur has an history of getting architectural revisions to market much faster than any other company. It’s not clear whether their next version will already be 45nm or not, but we definitely can’t help but be excited by the (very realistic) prospect of a 45nm dual-core higher-IPC variant!

Glenn Henry, founder of Centaur, has always been of the opinion that there was a market for cheap but ‘good enough’ CPUs and that eventually x86 would become a commodity market. It has been a rough road for them in the last 15 years or so, but Isaiah and the industry trends in general indicate his vision might finally be fully realized in the next few years.

I have personally claimed back in mid-2006 that VIA might have been a viable acquisition target instead of ATI given the much lower price tag (although it didn’t offer the same bundling opportunities or the CE exposure), and that it would be wise for NVIDIA to consider acquiring them (despite the x86 license not being transferable, there’s nothing that prevents them from getting a new one) in the coming year to counter AMD’s strategy.

I was proven right on the former up to a certain extent, but the latter obviously didn’t happen. It still could, however, as VIA’s market capitalization is now lower (and rightfully so given the debacle that is their chipset business) and their product roadmap including Isaiah is potentially very attractive for single-chip integration. No matter what happens, VIA has now been given a new beginning with Isaiah, and a new chance to return to profitability even if they remain on their own. We wish them and their various engineering divisions the best of luck in achieving that goal.

 

You can comment on this article and on Isaiah in general in this thread on our brand new Processor & Chipset Technology forum!