Sun Microsystems is expected to revamp its core Unix server line on Tuesday with the debut of its high-end UltraSparc IV+ processor. The big UltraSparc IV+ improvement is the inclusion of 2MB of built-in high-speed cache memory where UltraSparc IV had none. The initial clock speed of the dual-core chip is expected to be 1.5GHz.
What a bad idea, if you ask me.
The original USIV has 16 MB of cache, IIRC. The summary is poorly written.
The new USIV+ has even more cache.
Only Intel is dumb enough to sell modern processors without cache (original Celeron?).
The article is a little misleading. The L2 cache was off-chip in the US-IV, but there was 8M/core (16M/socket). The new chip will have on-chip 2M L2 cache, as 32M off-chip.
Sun has also shrunk the die from 130nm to 90nm, which is what has allowed them to add the on-chip cache.
Most of that new circuitry is memory. Where the UltraSparc IV had 16MB of cache memory packaged separately from the chip–8MB for each processor core–the IV+ will have 2MB directly on the chip shared by the two cores and 32MB packaged separately, Greenley said. On-chip cache is faster than cache packaged separately.
The first link (highlighted UltraSPARC IV+) points to another cnet article that has the paragraph above. Granted the original article missed the point of the 32MB L3 cache.
is a top-of-the-line sparc…. if sun goes out of buisness i’ll buy the best model they produced used or something. *sigh*
It looks like Sun has a really (excuse the pun) bright future. Partnering with Fujitsu and going massively multi-core on the high end while using Opterons in their low end offerings seems like a killer combination. Sun has excellent engineers and excellent technology. They may never gain back their dot com status but they aren’t going away any time soon.
The UltraSPARC III was almost a good idea. The idea was that, while competitors were creating whole new architectures with advanced features that could require long debugging and refinement times, Sun planned to do a quick re-implemetation of the UltraSPARC II which just cleaned up all the bottlenecks without any new risky technology. This would allow them to finish the design in less time, and have a product faster than the previous generation (but slower than the next), half a generation ahead of everyone else.
The problem was that it was still a completely new design, and it turns out that it takes the same amount of time to debug a completely new design whether it uses brand new technology or not. So UltraSPARC ended up taking an entire generation of development time anyway, but only delivering half a generation of performance improvement. Oops.
Fujitsu, developing like the other competitors, produced a SPARC with performance comperable to POWER, Itanium, and Alpha (which was always a generation ahead anyway, until now).
Sun now has a 1.5Ghz dual core chip with 1MB cache per core? That’s 500 Mhz slower than a consumer grade AMD chip and over 1000 Mhz slower than Intel’s.
It will be interesting to see some benchmarks, because RISC is supposed to be faster per clock, IIRC. IBM PPC chips at that speed run faster than comparable AMD Opterons for some benchmarks..
As a consumer I have no incentive to move to the Sparc architecture. As a business I would choose price/performance and compatibility over superior design any day, specially in the tech industry where performance doubles every 18 months or so and prices fall faster than expected.
Is Sun still the dot in dotcom?
The MHz/GHz argument died several years ago. Why keep bringing it up? MHz is a meaningless number, and everyone is ending up reducing their clock speeds to go with dual core (or larger, as Niagara will have).
Sun isn’t after the consumer market with these, they’re after the enterprise market. As far as price/performance and compatibility, they’ve got quick a compelling story now. Solaris across both SPARC and X64, from workstations to 1RU 1 way boxes all the way up to 128 ways; X64 boxes that are certified for Solaris, Linux, and Windows (Tier 1 support). The new Opteron workstations and Galaxy servers have the lowest entry price of any mainline vendor.
Sun may have struggled for the past couple of years, but it looks like they are coming back with a vengence.
One more thing just to add to what spotter has said. If this upgrade should be like the USIV one it will just be board replacement into the current systems. That is a huge incentive since customers don’t buy new systems they just populate the servers with new boards and double thier existing performance. That is huge in the Data center.
Why not introduce these at the recent “network computing” event? Did they not want to steal Galaxy’s thunder (not that that amount of thunder could be dampened)?
I want to put a server in the rack that I can put my most critical business processes on and then go to sleep at night knowing that it will all be running today, tomorrow and next year. I have Sun servers in my server room that go all the way back to Solaris 2.5.1 on Sparc20 and I replace them when they look to be dying. They refuse to die. I have to spill coffee into the cooling ports to “try” to kill a Sun server. My V880 runs around the clock for years now and there has never been a minute of downtime, not one. People cry for patch updates but the Solaris OS on those Sparc boxes just never stops. Speed? I don’t care about an extra percentage point in speed when I can get perfect uptime. For blazing speed I put in a stack of V20z units ( and some Appro boxes too ) and then run Solaris 10 on them. Zone some of them and run VMware on RedHat on the rest and everyone is happy.
Incorrect; UltraSPARC have always had cache, both level 2 and 3 – where that cache was located, well, thats a different story, and how fast it is, is another story as well. IIRC, they’ve moved the cache onto the chip and to full speed but still have level 3 cache if one desires to boost it up a notch.
I fear this is too little too late for Sun. Right now sparc machines are about dead-last in the price/performance arena. Not that they are bad machines, they just aren’t fast. We still have a lot of Sun servers because for some uses their other qualities are more important than price and speed. While Solaris is my favorite version of *NIX we are in the process of moving some compute intensive processes to IBM/AIX and most of our new web work goes to clusters of Linux/Intel servers. Unless Niagra and Rock deliver big-time I doubt we will have any Sun servers in our data center 5 years from now.
“in the process of moving some compute intensive processes to IBM/AIX”
Ah, you bought into IBM’s Linux strategy.
They don’t even need new boards; going by the article, if your board supports UltraSPARC III or IV (previous generation), it will be simply a matter of yanking out the old CPU, sending that back to SUN, who sometimes give a discount if you’re willing to swap CPUs, then install the new one.