i975X: Intel offers dual-PEG for the mainstream market
by Peter Mayerhofer on September 10, 2005 4:00 PM EST- Posted in
- Bulldozer
Intel Chipset Roadmap Update
Intel is preparing a new high-end Pentium 4 chipset, the 975X Express chipset. Due for release early in 2006 - possibly January - this chipset marks Intel's entry into the mainstream dual PCI Express (PCIe) Graphics market. (Intel supported dual X8 slots with the Tumwater chipset, but that was targeted at workstations and servers.) NVIDIA has remained unchallenged in this market since its introduction of nForce4 Intel Edition this past spring. ATI will be also be entering this market in the fall with its Crossfire chipset.
The Intel 975X Express chipset is the evolution of the 955X chipset. The primary difference with the 955X chipset is that 975X chipset can subdivide its PCIe X16 into two PCIe X8 lanes, supporting flexible I/O implementations including SLI and Crossfire video cards. However, NVIDIA in August of this year released their latest SLI chipset for socket 775, the nForce4 SLI X16 MCP, that supports 2 X16 PCIe slots. (That's two X16 data paths - PCIe is architected so that you can have a physical X16 slot with anything from an X1 data path all the way up to X16.) Our sources indicate that at launch Intel expects to have two X16 lanes instead of two X8 lanes to compete with NVIDIA.
If you're not familiar with X8 and X16 terminology, here's the recap: Each PCIe lane sends one bit per clock in each direction. A single PCIe lane is designated as X1. Connecting eight PCIe lanes together to one device is called a PCIe X8 port. Similarly connecting 16 PCI Express lanes gives a PCIe X16. High-end video cards, such as NVIDIA's 7800GTX, as well as budget SLI video cards that do not use a direct bridge between cards, such as NVIDIA's 6600LE, currently don't need more than PCIe X8. Future video cards, especially mainstream non-bridged SLI video cards, will likely require the X16 configuration for optimal performance. NVIDIA's high-end video cards handle SLI by using a direct bridge between the two video cards that bypasses the PCIe ports.
Intel PCIe Chipsets |
||||
Manufacturer | Chipset | Configuration | Launch | |
ATI | Crossfire | 2 X8 PCIe | Soon | |
Intel | 945/955 | 1 X16 PCIe | April 2005 | |
Intel | 975X | 2 X8 PCIe | Q1'06 | |
NVIDIA | nForce4 SLI Intel | 2 X8 PCIe | April 2005 | |
NVIDIA | nForce4 SLI X16 | 2 X16 PCIe | August 2005 |
The 975X Express chipset is designated as a high performance platform and is aimed at power users, including multimedia users and high-end gamers. It will also be used for entry level single processor workstations and servers. Other finer points of the 975X chipset include support for dual-core CPUs with HT, 800/1066MHz FSB, 533/667 DDR2 memory, 8 GB RAM maximum with optional ECC support, and Intel Memory Pipeline Technology. The real kicker is that our roadmaps also indicate 975X will be introduced at the same price as the 955X, so the "Flexible PCI Express Configuration" doesn't cost motherboard manufacturers anything extra. More importantly, ATI and NVIDIA have both internally mentioned support for Intel's 975X motherboard, since it should only require a few driver tweaks to get SLI and Crossfire working on 975X. (There is no word yet on ATI and NVIDIA supporting each other's chipsets, however.)
Going forward, the 975X chipset is expected to support all NetBurst Desktop processors including the new Intel Pentium Extreme Edition 955 processor, the Pentium D 900 series and the Pentium 6X1 series (such as the 641). The Pentium D 900 series are dual core processors scheduled to be released in January 2006. These processors are produced on 65nm, have Virtualization Technology (VT) capability, run at 800FSB and have 2 x 2MB L2 cache, ranging from 2.8GHz to 3.6GHz in clock speed at launch. The Intel Pentium Extreme Edition 955 processor is expected to be released in Q1 2006. This dual core CPU runs at 3.46GHz. It is similar to the Pentium D 900 series, but adds HyperThreading Technology and has a faster 1066MHz FSB.
The similarities between the chipset and processor numbering may cause some confusion. For example there is the 955X chipset and the Pentium Extreme Edition 955. There is no direct relationship between a chipset number and Intel's CPU rating, although we might see some clever marketing with overlapping CPU and chipset names. (Ed: Hopefully not.)
In other chipset news, the Intel road map shows continuing support for the 945 chipset for mainstream Pentium 4 and Pentium D systems. Three new variants of the 945 chipset are scheduled to be released. In Q4 2005, the 945PL Express chipset will be launched. This is a value version of the current 945P but with the support of only 2 DIMMs (1 per channel), and no support of DDR 667 or FSB 1066. In Q1 2006, the 945GZ will be release. It is similar to the 945PL but with integrated Intel GMA950 graphic and without the support for a discrete PCIe graphics card. The other 945 chipset launched in Q1 2006 is the Intel 945GT Express chipset. (Yeah, "GT", that's not overused at all....) This chipset expands on the existing 945G chipset with a custom system profile for All-In In-One and entertainment PC platforms.
The 975X chipset, with the 945G, 945P and 955X chipsets are part of the product requirements for Intel's new Viiv platform. Intel states, "Intel Viiv technology is a set of the latest PC technologies designed to maximize the enjoyment of digital entertainment in the home." AnandTech will publish an article describing Viiv in more detail in the near future. Intel's Broadwater chipset versions are also expected to launch within the first half of 2006 and will also support Viiv. In all likelihood, Broadwater will be renamed to 965 G/P/etc. It is not clear whether Broadwater will support multiple PEG implementations in the same manner 975X, though the relationship between the past 865/875, 915/925, and 945/955 makes it a distinct possibility. 975X may end up with dual X16 lanes and Broadwater with dual X8 lanes, or Intel may choose to simply disable that feature on "non-Extreme" chipsets. What we find particularly important is that 975X and Broadwater will both support the new Conroe processors. That should be welcome news for anyone that got burned by 915/925 not supporting Pentium D!
With the 975X chipset, Intel will finally enter the high-end, dual PCIe graphics chipset market. This should occur early next year and will allow them to compete with NVIDIA's nForce4 and ATI's Crossfire (eventually) offerings.
21 Comments
View All Comments
xsilver - Saturday, September 10, 2005 - link
hmm... interesting newsoh wait -- its intel --- yawn :P
there wont be any real competition until conroe shows up
where intel will release another chipset that supports it
intel mobo's havent been good for upgradability since the BX chipset -- and that wasnt even officially supporting 133fsb
danidentity - Sunday, September 11, 2005 - link
Did you even read the article?2nd to last paragraph:
"What we find particularly important is that 975X and Broadwater will both support the new Conroe processors."
IntelUser2000 - Sunday, September 11, 2005 - link
Well I HOPE its true, since Yonah is rumored to be S480 or something, and probably Conroe is based on it. And Anandtech hasn't been correct recently on roadmaps.
IntelUser2000 - Sunday, September 11, 2005 - link
I can't believe they are going for this "We have 2x8 configuration while our competitor has 16x1, 1x1 PCI-e config, since that gives performance benefits"Now when since going to AGP2x give performance increase. The last time I have seen, you hardly get better performance going beyond AGP2x. PCI to AGP did make a vast difference, as my personal test using Geforce2 MX gave 50% increase at fastest setting in Quake 3. I wanted to replace to AGP, and it was more than worth it. However, beyond 2x does NOTHING. We have all seen proof of that with PCI-e in Anandtech's article about it. I bet there is NO advantage by going 2 16x from 1 16x, 1 1x.
mlittl3 - Sunday, September 11, 2005 - link
This article reads a little like a theinquirer.net article. Interesting.Anyway, in response to inteluser2000, you are right about no performance gains going from AGP 2x/4x to higher bus speeds, however, you are completely incorrect about saying there is NO advantage going to 2 16x slots. This is what we call SLI if we use Nvidia's jargon. There is a definite performance improvement going to two video card slots which allow two video cards to run simultaneously.
Now whether or not those two slots should be 1x, 2x, etc. is another questions indeed but I don't think having two AGP slots was feasible from a cost perspective as its generally easier to wire devices with serial connections rather than parallel. PCI-e now allows this.
So I think we can all agree that from a multiple GPU standpoint, you need PCI-e (given the current technology) over AGP. Now how fast that PCI-e slot must be (1x, 2x, etc.), I leave up to the reviewers.
KristopherKubicki - Monday, September 12, 2005 - link
TheWho?
Kristopher
Tarx - Monday, September 12, 2005 - link
One advantage of the PCI-E 2 x16 is the ability to run more powerful graphic cards (not yet available) in SLI mode without using a direct bridge between the cards.JarredWalton - Tuesday, September 13, 2005 - link
On the flipped side, the current setup where one X16 connection comes from the NB and the seconds comes from the SB is... well, odd to say the least. Last I looked, no NB-to-SB connection was anywhere near X16 bandwidth. So they've got X16 going over what amounts to an X2 or X4 connection, as far as I can tell. Did I miss something, or is that the actual current configuration for two X16 lanes from NVIDIA? If Intel does two X16 lanes that are directly linked at full X16 speed, that would actually have potential use.danidentity - Tuesday, September 13, 2005 - link
The nForce 4 doesn't have a NB and SB. It's a single chip.The new nForce 4 SLI X16 chipset has 16 lanes going to each slot.
IntelUser2000 - Wednesday, September 14, 2005 - link
Nforce 4 Intel Edition does. I heard the 2 PCI-E x16 version for both AMD AND Intel has NB and SB(or two chips for that matter).