With the economy on the mend - at least by gross measures that may not mean a hill of beans to people on the street - Intel thinks it is putting its "Westmere-EP" Xeon 5600 processors into the field at precisely the right time. Not only are people more willing to spend money than they were a year ago, but their aged servers are …
aged servers 1 year closer to death
yep, the Compaq Proliants I bought in 1999 are gonna die any day now...
Imitation is the most sincere form of flattery
All hail the great Intel innovation as present in all Via CPUs for the last 8 years. All hail the great Intel achievement of finally putting AES on the CPU. I have been using it for VPNs and encrypted backups for years now.
Now can we have that where it really matters please. An enterprise banking app does not run its SSL on the web server CPU anyway. It is on a front-end load balancer supplied by the like of Juniper, F5 or Cisco. Doing it on the CPU in a server class hardware is a waste of silicon.
The place where it really matters is the desktop where it can support our VPNs, SSL browsing or even WiFi which Intel has long moved to the realm of software crypto. Can we have it there please.
And considering how long Via has had it, for Intel to claim any form of superior achievement here is frankly preposterous.
Depends on the Via chip
It's true that Via have had it for ages, but unfortunately it's not terribly good. OpenBSD did a few calculations a while back about x86 CPU vs crypto extensions and found that using the special extensions was actually slower..
Fixed in newer chip designs I believe, but certainly at one point, actually worse than nothing.
Did a couple of (not very precise) openssl speed tests on my via c7, command lines copied from the kd85 openbsd guy, and found that for block sizes of 16 bytes going to the hardware (and incurring context switches and so on) was slower than software, but the larger the block size, the bigger the gain.
You may be thinking of the via c3, which is a bit slower. Or maybe about the hifn mini-pci cards found in soekris boxes, that have decent throughput for pci cards and are definately a boon for the souped-up 486-type cores found in the geode chips in those soekris boxes, but being pci cards can't compete with on-chip stuff.
And then there's the sun opensparc chips with built-in crypto, traditionally the home of java apps. Not to mention the stupidly fast POWER chips that have a market in banking. Again, intel isn't setting the trend here, they're solidly following.
Like comparing new kit with kit from 2005, though. But what I'd really want to know: How does their kit compare to 1997, the last year there were still single core supers in the top-500? Incidentally typing this on a 1997 thinkpad; my newer laptop died. Spare me a newer? *grin*
If your buying new gear why might you want to buy the 6-core systems instead of the upcoming 8-core systems? Or better yet upcoming 8 and 12 core AMD systems?
I can understand if your really tight on capacity perhaps on existing 5500 systems you may want to switch out the CPUs(may not be worth it unless you find a way to get credit for the CPUs or re-sell them to someone else), but really don't see the point in going with 5600 otherwise.
Maybe upgrading from 5500 to 5600 in a year or so may make more $$ sense.
Well, They would say that wouldn't they
Nothing like blowing your own trumpet is there Intel.
Of course we'd like to throw out all that old x64 hardware. You know that slinky box that was delivered less than a year ago.
Good luck trying to pitch this to the bean counters who at the monent have a big fat ZERO in their wallets for this kinda thing.
Around here, we all have to use 5yr old Stinkpads T43's. Nowt newer on the horizon sadly.
I don't understand how it is unreasonable...
..for Intel to compare the fastest Westmere to the fastest Nehalem.
The x5570 is/was an expensive processor. A lot of people bought them anyway, and many more bought the x5550 or x5560 instead, to get most of the value at significant discounts.
Anyway, companies will buy the x5670. When you are piling dozens of VMs on there it's not a bad idea. If you're installing Oracle database enterprise edition, who cares about an extra $2,000 tied up in processors when your database license was a few hundred thousand?
It isn't unfair to compare them as long as you make an adjustment for the higher clock speed (in brackets maybe) when comparing performance otherwise you are not showing people what they pick up purely by design rather than just by overclocking.
As the author stated, most companies are unlikely to use a 130W design over a 95W design so comparing the two 2.93s would make more sense.
Just consider paying a bit more for you database, because db vendors (at least a couple of major ones) charge you not per CPU SOCKET, but CORE. So, more cores in the same box - more money out of you pocket. Say $150K instead of current $100K ;) Exactly, what is that $2K at all?
Just to everyone who is is not familiar with the database licensing policy. Database vendors (mostly) count their license per PROCESSOR CORE, not per SOCKET. So you need more performance per CPU core to get more work done with the same software budget. If you'll decide to go to 6 or 8 cores instead of 4 - get your wallet ready!
For some reason, Intel used the SPEC_int_rate2006 test
Erm, could that reason be that the integer test allows Intel to use almost all of the non clock speed impacting power management features in the processor and turn off much of the core (very much enhanced in recent versions of their silicon). The SPEC Integer test is not going to work much of actual core real estate or memory I/O due to its nature. Of course if they ran a mixed workload benchmark there would be a lot more core turned on and much more happening in memory I/O (yes that uses more power than just keeping the RAM refreshed). So, the "some reason" will be that it is the benchmark that gave the best comparison between the old and new processor designs.
Re: Imitation is the most sincere form of flattery
If you put it into the server chips, you start to make it more attractive to do your load balancing with nice cheap servers, rather than buying expensive black boxes from the aforementioned vendors.
If you want it on the desktop, buy a Core i5 (or a i7-980X).
Via may have had acceleration for a while, but Intel chips were probably still faster even without.
The 8-core chips from Intel aren't entirely comparable. (aka Beckton, Nehalem-EX or Xeon 7500)
They're more akin to being a pair of 5500s in a single package. So they don't have all the architectural advantages of the 5600s.The other main difference is they use different sockets. 5600s go into boards that only take 2 processors, Beckton will go into 4-way and above.
More to encryption than bit-twiddling
If off-board crypto gear is done away with, where do the random numbers for session keys come from? VIA provides these in the CPU, Intel doesn't. Presumably there will be a future Intel chipset with a random number generator, but until then most designs which currently run off-board cryptography will need to retain it, at the very least as a robust source of random numbers.
Touting a fast processor for virtualisation makes me look to the see if the chipsets support the sort of massive I/O required to run the CPUs at full tilt. As your SAP results suggest, the architecture is limited by I/O throughput.
I/O and everything else?
And pray tell, where is this magical giant leap in I/O performance for your applications going to come from when you consolidate 50 servers into 3?
Are we all meant to trust and afford 256GB SSDs in our servers now? I think not.
- Geek's Guide to Britain INSIDE GCHQ: Welcome to Cheltenham's cottage industry
- 'Catastrophic failure' of 3D-printed gun in Oz Police test
- Game Theory Is the next-gen console war already One?
- Analysis Spam and the Byzantine Empire: How Bitcoin tech REALLY works
- Apple cored: Samsung sells 10 million Galaxy S4 in a month