By: rwessel (robertwessel.delete@this.yahoo.com), January 30, 2013 4:07 pm
Room: Moderated Discussions
Mark Roulo (nothanks.delete@this.xxx.com) on January 30, 2013 3:51 pm wrote:
> rwessel (robertwessel.delete@this.yahoo.com) on January 29, 2013 10:15 pm wrote:
> > Mark Roulo (markroulo.delete@this.yahoo.com) on January 29, 2013 6:08 pm wrote:
> > > someone (someone.delete@this.somewhere.com) on January 29, 2013 8:58 am wrote:
> > > > Richard Cownie (tich.delete@this.pobox.com) on January 29, 2013 4:42 am wrote:
> > > > > Very interesting analysis, thanks.
> > > > >
> > > > > I think the argument about area and cost - and Intel's processing advantage -
> > > > > is all correct. What may be missing is the impact of the different business
> > > > > strategies: Intel really wants to sell those server cpus at very high margins -
> > > > > say 300mm2 of silicon for $1500.
> > > >
> > > > Market research companies like Mecury have indicated Intel server MPU ASP is
> > > > around $300. The vast majority of units sold are two socket server MPUs that
> > > > differ from PC variants of the same device only by feature fusing. The high end
> > > > Intel server MPUs with 4 figure prices sell in relatively small quantities and using
> > > > them as a strawman for Intel competitiveness in low end servers is fallacious.
> > >
> > >
> > > I have seen this ...
> > >
> > > http://investorvillage.com/mbthread.asp?mb=476&tid=10759150&showall=1
> > >
> > > Can't vouch for the accuracy, but the claim is that the
> > > average ASP for Intel server chips in Q2 of 2011 was ~$570.
> > >
> > > The numbers look plausible ...
> >
> >
> > Right, but unless you think that the microserver vendors are likely to be going head to
> > head with Intel 2S and larger systems, the relevant datum is the $252 ASP for Xeon UPs.
>
> Unless we think that these customers need the RAS (including ECC), the relevant ASP might be
> the Value Desktop chips at $57.92. I'm guessing that these are single socket, quad core, dual
> memory channel chips. I'd expect the individual cores to be faster than current ARM cores, so
> you might need two Tegra4 (quad core) chips (minus the graphics and wireless) to match throughput
> performance [Tegra4 would be clocked lower and I expect that Intel gets more done per clock because
> of better branch predictor, better hardware prefetch, and deeper OoO execution].
>
> The ARM chips would (I think?) have an edge in power (which translates into operating costs),
> but not a huge edge in purchase price. Two Tegra4s would maybe run $25x2 = $50.
>
> $50 vs $58 isn't a huge win by itself.
>
> The operating cost win of ARM might be ... but the Intel chip is probably ~65W TDP (which probably means
> more like 30-35W under typical load ... I don't expect the servers these things go into to be hammering
> the SSE/AVX units much). If the two Tegras need something like 10W total, then the ARM edge is between
> 20W and 50W per motherboard equivalent. After adding in the power of Ethernet, DRAM, hard drives, etc.
> does the power advantage for the ARM chips matter enough to switch without a noticeable purchase price advantage?
> Especially if one expects Haswell chips to come in at less than SandyBridge power budgets?
Just to toss a couple of numbers out there, 50W for three years, assuming 20 cents per kW-H (to cover both power and cooling costs), is about $260. 20W is about $105. OTOH, the cheap Intel CPUs are not the best for performance/watt.
> rwessel (robertwessel.delete@this.yahoo.com) on January 29, 2013 10:15 pm wrote:
> > Mark Roulo (markroulo.delete@this.yahoo.com) on January 29, 2013 6:08 pm wrote:
> > > someone (someone.delete@this.somewhere.com) on January 29, 2013 8:58 am wrote:
> > > > Richard Cownie (tich.delete@this.pobox.com) on January 29, 2013 4:42 am wrote:
> > > > > Very interesting analysis, thanks.
> > > > >
> > > > > I think the argument about area and cost - and Intel's processing advantage -
> > > > > is all correct. What may be missing is the impact of the different business
> > > > > strategies: Intel really wants to sell those server cpus at very high margins -
> > > > > say 300mm2 of silicon for $1500.
> > > >
> > > > Market research companies like Mecury have indicated Intel server MPU ASP is
> > > > around $300. The vast majority of units sold are two socket server MPUs that
> > > > differ from PC variants of the same device only by feature fusing. The high end
> > > > Intel server MPUs with 4 figure prices sell in relatively small quantities and using
> > > > them as a strawman for Intel competitiveness in low end servers is fallacious.
> > >
> > >
> > > I have seen this ...
> > >
> > > http://investorvillage.com/mbthread.asp?mb=476&tid=10759150&showall=1
> > >
> > > Can't vouch for the accuracy, but the claim is that the
> > > average ASP for Intel server chips in Q2 of 2011 was ~$570.
> > >
> > > The numbers look plausible ...
> >
> >
> > Right, but unless you think that the microserver vendors are likely to be going head to
> > head with Intel 2S and larger systems, the relevant datum is the $252 ASP for Xeon UPs.
>
> Unless we think that these customers need the RAS (including ECC), the relevant ASP might be
> the Value Desktop chips at $57.92. I'm guessing that these are single socket, quad core, dual
> memory channel chips. I'd expect the individual cores to be faster than current ARM cores, so
> you might need two Tegra4 (quad core) chips (minus the graphics and wireless) to match throughput
> performance [Tegra4 would be clocked lower and I expect that Intel gets more done per clock because
> of better branch predictor, better hardware prefetch, and deeper OoO execution].
>
> The ARM chips would (I think?) have an edge in power (which translates into operating costs),
> but not a huge edge in purchase price. Two Tegra4s would maybe run $25x2 = $50.
>
> $50 vs $58 isn't a huge win by itself.
>
> The operating cost win of ARM might be ... but the Intel chip is probably ~65W TDP (which probably means
> more like 30-35W under typical load ... I don't expect the servers these things go into to be hammering
> the SSE/AVX units much). If the two Tegras need something like 10W total, then the ARM edge is between
> 20W and 50W per motherboard equivalent. After adding in the power of Ethernet, DRAM, hard drives, etc.
> does the power advantage for the ARM chips matter enough to switch without a noticeable purchase price advantage?
> Especially if one expects Haswell chips to come in at less than SandyBridge power budgets?
Just to toss a couple of numbers out there, 50W for three years, assuming 20 cents per kW-H (to cover both power and cooling costs), is about $260. 20W is about $105. OTOH, the cheap Intel CPUs are not the best for performance/watt.