By: Wilco (Wilco.Dijkstra.delete@this.ntlworld.com), August 9, 2014 2:30 pm
Room: Moderated Discussions
Alberto (git.delete@this.git.it) on August 9, 2014 2:13 pm wrote:
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on August 9, 2014 1:49 pm wrote:
> > Michael S (already5chosen.delete@this.yahoo.com) on August 9, 2014 12:38 pm wrote:
> > > anon (no.delete@this.thank.you) on August 9, 2014 10:17 am wrote:
> > > > juanrga (nospam.delete@this.juanrga.com) on August 9, 2014 6:43 am wrote:
> > > > > Aaron Spink (aaronspink.delete@this.notearthlink.net) on August 9, 2014 3:44 am wrote:
> > > > > > juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 10:49 am wrote:
> > > > > >
> > > > > > > Take a modern A57 core. According to AMD the A57 Opteron is faster than jaguar based Opteron but
> > > > > > > consumes less power. The ARM core performance is ~40% faster, and consumes roughly one half.
> > > > > > >
> > > > > > Faster at what? Its a bunch of market point pointing to nothingness.
> >
> > Jaguar has about the same IPC as Cortex-A15, so is it really that hard to believe A57 is much faster?
>
> I don't see proof behind your claim. Looking how performs Jaguar vs.
> Silvermont, i can say Jaguar is pretty neck to neck with A57.
Eh how did you come to that (incorrect) conclusion? Jaguar and Silvermont are similar, but A57 is in a different league.
> > > I can believe it. Jaguar is low-power low-frequency design with comfort zone around 1.5 GHz.
> > > If you take Jaguar to 2GHz (as in Opterons) it is very probable that
> > > it's going to lose to A57 both in power and in performance.
> > > That's why AMD builds overwhelming majority of their Opterons around Buldozer and Piledriver cores now and
> > > likely going to migrate to Steamroller in the near future. If I am not mistaken, out of dozens of Opteron
> > > models only 2 are based on Jaguar, i.e. even less that there are Xeon models based on Silvermont.
> > > And their main selling point of Jaguar-based Opterons is
> > > low price rather than high performance or low power.
> > > In fact, power-wise Jaguar-based Opterons are pretty uninteresting - 17W for
> > > quad-core GPU-less 2 GHz part. For comparison, Intel Avoton has 20 W octacore
> > > at 2.4 GHz. But the later costs (list) $171 while the former only $64.
> > >
> > > Still, although I believe it, I'd really like to see published SPECInt
> > > numbers for ARM-based Opterons instead of internal lab estimates.
> >
> > Anand quoted 80 SPECint_rate for Seattle and 28.1 for X2150, ie. ~43% more SPECint per core. That's
> > likely with the same uncore as AMD would reuse their existing L2/L3 cache and DRAM controllers. Note
> > also these are old results, ARM recently updated the A57 performance estimates showing better results,
> > and GCC work on ARM64 is obviously still ongoing, so actual results should be better than claimed.
> >
> > As for power, the A57 version uses 25W while Jaguar uses
> > 22W, so when running full-out at the same performance
> > A57 uses 40% of the power of Jaguar (with frequency scaling it should be far better than that).
> >
> > Wilco
>
> It's hard to compare a consumer SKU with a device "done" for server space. Obviously there
> isn't much effort in X2150, well motivated AMD can do much better looking at Mullins.
Both of these are server SKUs, if you have proof otherwise, please do tell.
> "Seattle" is like to say "i have not resources", better buy (not so good) IP from ARM to do multicore cloud cpus.
> Same thing are doing few little companies around the world trying to grab some market share from Intel.
AMD has already announced a custom ARM64 server core for 2016, so yes it will be like one of those little companies such as Broadcomm designing Haswell-class server CPUs.
Wilco
> Wilco (Wilco.Dijkstra.delete@this.ntlworld.com) on August 9, 2014 1:49 pm wrote:
> > Michael S (already5chosen.delete@this.yahoo.com) on August 9, 2014 12:38 pm wrote:
> > > anon (no.delete@this.thank.you) on August 9, 2014 10:17 am wrote:
> > > > juanrga (nospam.delete@this.juanrga.com) on August 9, 2014 6:43 am wrote:
> > > > > Aaron Spink (aaronspink.delete@this.notearthlink.net) on August 9, 2014 3:44 am wrote:
> > > > > > juanrga (nospam.delete@this.juanrga.com) on August 8, 2014 10:49 am wrote:
> > > > > >
> > > > > > > Take a modern A57 core. According to AMD the A57 Opteron is faster than jaguar based Opteron but
> > > > > > > consumes less power. The ARM core performance is ~40% faster, and consumes roughly one half.
> > > > > > >
> > > > > > Faster at what? Its a bunch of market point pointing to nothingness.
> >
> > Jaguar has about the same IPC as Cortex-A15, so is it really that hard to believe A57 is much faster?
>
> I don't see proof behind your claim. Looking how performs Jaguar vs.
> Silvermont, i can say Jaguar is pretty neck to neck with A57.
Eh how did you come to that (incorrect) conclusion? Jaguar and Silvermont are similar, but A57 is in a different league.
> > > I can believe it. Jaguar is low-power low-frequency design with comfort zone around 1.5 GHz.
> > > If you take Jaguar to 2GHz (as in Opterons) it is very probable that
> > > it's going to lose to A57 both in power and in performance.
> > > That's why AMD builds overwhelming majority of their Opterons around Buldozer and Piledriver cores now and
> > > likely going to migrate to Steamroller in the near future. If I am not mistaken, out of dozens of Opteron
> > > models only 2 are based on Jaguar, i.e. even less that there are Xeon models based on Silvermont.
> > > And their main selling point of Jaguar-based Opterons is
> > > low price rather than high performance or low power.
> > > In fact, power-wise Jaguar-based Opterons are pretty uninteresting - 17W for
> > > quad-core GPU-less 2 GHz part. For comparison, Intel Avoton has 20 W octacore
> > > at 2.4 GHz. But the later costs (list) $171 while the former only $64.
> > >
> > > Still, although I believe it, I'd really like to see published SPECInt
> > > numbers for ARM-based Opterons instead of internal lab estimates.
> >
> > Anand quoted 80 SPECint_rate for Seattle and 28.1 for X2150, ie. ~43% more SPECint per core. That's
> > likely with the same uncore as AMD would reuse their existing L2/L3 cache and DRAM controllers. Note
> > also these are old results, ARM recently updated the A57 performance estimates showing better results,
> > and GCC work on ARM64 is obviously still ongoing, so actual results should be better than claimed.
> >
> > As for power, the A57 version uses 25W while Jaguar uses
> > 22W, so when running full-out at the same performance
> > A57 uses 40% of the power of Jaguar (with frequency scaling it should be far better than that).
> >
> > Wilco
>
> It's hard to compare a consumer SKU with a device "done" for server space. Obviously there
> isn't much effort in X2150, well motivated AMD can do much better looking at Mullins.
Both of these are server SKUs, if you have proof otherwise, please do tell.
> "Seattle" is like to say "i have not resources", better buy (not so good) IP from ARM to do multicore cloud cpus.
> Same thing are doing few little companies around the world trying to grab some market share from Intel.
AMD has already announced a custom ARM64 server core for 2016, so yes it will be like one of those little companies such as Broadcomm designing Haswell-class server CPUs.
Wilco