OEMs Adopt Intel's 'Tulsa' ChipBy Jeffrey Burt | Print
Re-Imagining Linux Platforms to Meet the Needs of Cloud Service Providers
The new high-end Xeon processor offers massive 16MB cache and twice the performance of the current "Paxville MP" model, according to the chip maker.
A host of OEMs are lining up to offer servers running on Intel's latest processor, the dual-core Xeon MP 7100 series for systems running four or more chips.
Dell, Hewlett-Packard and IBM highlight the list of server makers looking to take advantage of the massive 16MB Level 3 cache and other features rolling out Aug. 29 in the processor, formerly code-named "Tulsa."
The series is the latest in a slew of new chips introduced by Intel over the past few months as it works to slow the momentum gained by rival Advanced Micro Devices and its Opteron processor.
Tulsa takes aim at the high-end of the x86 space, a segment that has been especially kind to Opteron. Intel officials say the new processor not only offers greater performance but also performance-per-watt than the current "Paxville" Xeon MP chip and AMD's technology, a key metric at a time when businesses are becoming more aware of rising power and cooling costs in their data centers.
"The performance and performance-per-watt improvements [in the 7100 series] are key and continue our drive to lead the industry in those categories," said Jay Parker, director of PowerEdge servers for Dell.
Not all industry observers are sold on the chip, but there is enough promise to entice users to take a look.
Jevin Jensen, director of IS at Mohawk Industries, in Dalton, Ga., said he wants to do a side-by-side comparison between a Tulsa-based server and one running AMD's newest "Rev F" Opterons. Of particular interest is how the chipswhich both offer hardware-based virtualization technologyperform with VMware virtualization software.
"I have contacted both vendors about getting four-way demo units of each," Jensen said. "On paper, it appears AMD may still have a slight advantage, especially since Tulsa is based on the old NetBurst architecture, but I will let our real-world tests decide the outcome."
Intel is phasing out its NetBurst architecture in favor of its new Core platform, which not only offers better performance but is more energy-efficient. Tulsa is the last of the NetBurst chips that Intel will make.
Dell, of Round Rock, Texas, is putting the Tulsa chips into its four-socket PowerEdge 6800 and 6850 servers, which will be available immediately. Dell officials expect the new chips to give their systems a 123 percent performance boost, and a 129 percent jump in performance per watt.
HP, of Palo Alto, Calif., is putting the new chips into its four-socket ProLiant DL 580 G4 and ML 570 G4 servers, and will price them aggressively, starting at $5,799, said John Gromala, director of server marketing. Gromala said he expects to see a 62 percent performance increase because of the new Intel technology.
For its part, IBM, of Armonk, N.Y., next month will start shipping System x3850with volume availability in late Octoberand System x3800 and x3950, with first customer shipments in October. System x3800 and x3850 will also include an updated I/O subsystem supporting both PCI-X 2.0 and PCI Express slots.
Fujitsu Computer Systems, of Sunnyvale, Calif., is putting the chip into its four-socket Primergy RX600 S3 rack-mounted and the TX600 S3 tower systems.
Unisys and Supermicro Computer also will offer systems with the Tulsa chips. Supermicro, of San Jose, Calif., is shipping the ultra-dense 1U (1.75-inch) SuperServers 8014T-8 and 8014T-I and the 4U (7-inch) SuperServer 8044T-8R servers. Unisys in the fourth quarter will roll out the chips in its ES7000/one systems, which can run both Xeon and Itanium processors from Intel.
In all, Intel expects more than 40 systems makers to roll out servers with the 7100 series chips.
Mark Feverston, director of enterprise server marketing for Unisys, of Blue Bell, Pa., pointed to the large Level 3 cachesecond in size in Intel's processor lineup only to the 24MB offered in the latest dual-core "Montecito" Itanium2 9000 seriesas a key enhancement in the new chips.
"It allows us to really be able to provide significant scaling, which will allow us to do more work per server," Feverston said.
Tom Kilroy, vice president and general manager of Intel's Digital Enterprise Group, said the Santa Clara, Calif., company's new 65-nanometer manufacturing technology is a key to enabling Intel to offer such a huge memory cache, which speeds the amount of time it takes the processor to retrieve data. Most chip makers are still using a 90-nm manufacturing process.
The large shared cache is further enhanced by Intel's Cache Safe Technology, which isolates errors and writes them out to protect the rest of the cache from exposure. The technology also is offered in the Montecito chip.
Next Page: Huge performance boost.
The new chip offers twice the performance, three times the performance-per-watt and four times the performance-per-CPU as the current Paxville MP processor, Kilroy said.
Intel, which also is offering Tulsa technology in its four-socket SR4850HW4x and SR6850HW4x servers, is aiming the chip at such heavy-duty workloads as databases, supply chain management and IT consolidation.
The chip comes in several models, ranging from the largest, the 7140M, at 3.4GHz and 16MB of cache, to the smallest, the 7110, at 2.5GHz and 4MB. Prices range from $856 per 1,000 units shipped to $1,980.
The 7140M also has a large power envelope of about 150 watts, though Intel also offers a versionat 3GHz with 4MB of cachethat comes in at 95 watts. However, Kilroy said he expects the larger models to sell better in a space where performance is key.
"I don't see energy efficiency really as important in this segment," he said.
That is a key difference between Intel and AMD, according to AMD officials.
"We don't require customers to choose between high performance and great power savings," John Fruehe, worldwide market development manager for the AMD Server/Workstation Business, said in a statement. "We believe they should be able to get both in the same platform. ... Tulsa is a huge power consumer, which will represent a challenge for customers in both powering and cooling the server."
Though still dominant, Intel has seen its smaller rival cut deeply into its market share over the past three years, in large part due to AMD's ability to beat Intel in such key areas as 64-bit computing in x86 systems, dual-core technology and energy efficiency. According to analyst firm IDC, of Framingham, Mass., AMD, of Sunnyvale, saw its worldwide market share of x86 servers climb to 20.2 percent in the second quarter.
AMD now has chips running in servers from three of the top four vendors, and Delllong an Intel-only OEMlater this year will begin selling two- and four-socket Opteron-based systems.
However, while admitting to mistakes over the past three years, Intel is aggressively rolling out new system chips in what it has dubbed the "summer of servers." Those have included not only the Tulsa and Montecito chips, but also two families of dual-core Xeon DP chips for servers running one and two processors.
Intel also is expecting to come out with quad-core chips later this year, ahead of AMD, which won't release a quad-core Opteron until next year.
Check out eWEEK.com's for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.