For $600-ish, the C2750D4I is an 8-core Silvermont Atom. For $800-ish you get an 8-core Broadwell based SoC with integrated 10G networking, up to six PCIe 3.0 x4 slots and Xeon E3 v3 levels of control, all within 45W.
I wonder how cheap DELL and the other vendors are going to offer up Xeon D, because they could make some nice single purpose home rack servers (VPN, AD/LDAP, Mail, Compute) at pretty sweet price points.
Even the throwaway 2013-era Avoton parts are on par or even faster than the fastest ARM server parts that are even on the market, and ARM's own optimistic roadmaps are only promising a 50% improvement in efficiency... that would still lose to outdated Atom systems in performance/watt. The Xeon D is going to make any ARM server manufacturer's life very difficult.
"What Xeon D also brings to the table...". Dream on....or read the 'xeon-d-1500-specification-update" that buries on their site somewhere. It lists a bunch of errata that contradict this vaporware hype. Most features were buggy so are disabled on the production chips in the market.:( Even the 'good-ol' TSX-NI abracadabra magic developers like is mia (just like the old broadwells!!! surprise surprise). The paranoid amongst us might wonder why intel seems to rely on security researchers to find design flaws on their silicone rather than revealing Intel QA findings to IT press and purchasers.
Really odd, I thought the point of Silvermont in servers was that they were going to have a lot of cores on one package (say 32 or more), 8 cores that perform worse than 4 Haswell cores with little cost or power savings doesn't make much sense to me.
Read the article again and try to explain how $800 isn't adequately priced...this makes for a pretty powerful server (nowhere near Atom class) with 10GBE being icing on the cake. You could put together a VERY compact yet quite powerful server for under $2000 with this board, enterprise-class SSD and all.
True. It's been coming down in price a little over the years but it might become mainstream relatively soon if they start putting it on entry-level server boards like this.
Intel is not giving anything away for business class that is going to trash another Intel product line. This is between single socket desktop Xeons and Atom. If you're in the market for 10 GbE the prices are right there with better efficiency. Bear in mind there is two 10 GbE ports and Intel is pricing it because of that. They don't want to cannabalize their add-on networking.
As far as performance, think laptop with 8 cores/16 threads or 4 cores / 8 threads. This has some definite possibilities, but you will need to price out total system cost versus Atom and Xeon E3 lines. It is likely too high for me in price.
Broadwell core (quite fast per clock) and dual 10G networking makes for quite the upgrade. Seems like a good fit for higher end NAS, network equipment (router, firewall, IDS, etc), or microservers. Imagine say 24 of these on small cards in a chassis that integrates a 10G switch inside it.
The Avoton stuff already runs storage fine, this would be more than adequate. Asrock and Supermicro both have C2550/2750 boards with a dozen SATA connectors which are basically perfect for low end NAS - mine happily runs 16Tb (6x4Tb in RAIDZ2).
I'm thinking primarily storage, but not only storage. I do tend to run a single server at home, and I'd like it to run a VPN client/server, occasional x264 encoding, maybe some IP cameras or a few VMs, etc. It's a little more than a handful of Atoms can handle. The only accessory needed beyond the now standard dual GbE, though, is storage. The rest is just some extra CPU grunt, which this has plenty of.
Has a socketed XEON ever been overclockable? I don't understand how this wouldn't be fast enough for it's target market. I don't even understand how a sub-$200 Xeon like the E3-1230v3 with 8 cores at 3.3GHz wouldn't be fast enough for just about any workstation to deem overclocking necessary.
More is always better in a workstation. If your workloads are running for minutes to hours, every bit counts, yet it may not be practical to transition to server clusters. BTW: E3-1230v3 has 4 cores with hyperthreading, so 8 threads. That's just plain old Haswell, a bit slower than the i7 4770.
We're talking about the same thing here: when I say the chip IS Haswell I mean the same die, which you mean by calling them identical. You also acknowledge that it's lower clocked than i7 4770 - that's why they are not identical as a product and why I called it "a bit slower than i7 4770". I sure understand your post and am just replying because I'm a bit puzzled that you seemed not to understand mine the way it was intended.
|Only the older models can be overclocked easily, ie. those that relied on bclk multipliers (X58 and P55), or FSB before that. I have an X5570 that oc's quite well.
I intended to imply since Sandybridge (last 4 years) there has never been a "Xeon-K" or the like. But yes, the Core 2-based Xeon's were pretty good overclockers. X58 had some ok headroom but they ran warm stock (130w+ TDP) leaving only a few hundred MHz. I never got my i7-920 passed 3.2GHz but I did get my i7-950 to 3.5GHz.
Unlock the multiplier, put it onto a mainboard which allows for at least 100 W and has regular HSF mounting points and it could see quite a few fans. Tuning such a system could be really fun! Disable the 10GBit ethernet and ECC if you're afraid of canibalizing server sales.
Has ASRock said what the 14pin power connector provides? This model isn't up on their site yet, and when I poked around I couldn't find any existing models which weren't using conventional 24+8 pin power delivery.
My question was about its pinout/etc. While the MT-C224 uses an 18 pin connector, 4 of the pins aren't used; so it's possible this is just a rationalized version of that one. That said, it appears to be playing a bit fast and loose with the AX spec. The MT-C224 identifies 9 ground pins and 5 for DC in. Looking at the pinout on the ATX PSU adapter cable from Newegg's gallery, 4 of the DC in pins are for 12V, and the 5th appears to be the PS-ON signal. This means it's playing a bit fast and loose with the ATX spec, because it's ignoring the power-good (gray wire) signal; which is normally used by the PSU to signal when its voltages have stabilized after powering on and to signal an operational fault (hopefully soon enough to keep the mobo from smoking itself). It can (probably) get away with that by waiting past the maximum in spec wait for voltage stabilization before attempting to post and by using an onboard power monitoring circuit.
what makes you think this is atom class? it uses broadwell cores at a low clock this isnt baytrail or cherrytrail - its the full fat broadwell cores "Xeon D, also known as Broadwell-DE, combines up to eight high performance Broadwell desktop cores and the PCH onto a single die"
as far as i can tell its broadwell-e with the southbridge on the same substrate
I could use this in a combination NAS/PC/mini-workstation, since I unfortunately don't play games. I could run Windows for personal use and Linux for the server side, and not feel too guilty about leaving it turned on 24 hours a day.
My other reaction is that combined with, say, an NVIDIA K80 or two, this would provide fantastic computing per watt, much more efficiently than a single node in the Titan at Oak Ridge National Laboratory, which uses AMD 12-16 core CPUs if I remember correctly, running at around the same frequency as this CPU, and NVIDIA Tesla cards on a PCIe Gen. 2 interface. The networking on those nodes may be a bit faster perhaps, but 10G is not bad.
OK, so they've built the chip, it's now time to ramp up pressure on Intel to package this for laptops, and finally get rid of the 16GB RAM, no ECC and 4-core limits artificially imposed on mobile workstations.
I've never understood why the Xeon E3 wasn't repackaged into a mobile form factor much like a Core i7 is...
I really don't see anything great here. 10Gbe is still expensive because of the switches, not so much the adapters. A chip that runs 8 x 2.0Ghz is really not that impressive in the server world at these prides, if it could do SMP or something then it would get some attention. But as another commenter pointed out. Intel is only competing with itself in the high end server market and they aren't going to under-cut themselves by offering something at a much lower price point with the same computational horsepower as current Xeon's.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
38 Comments
Back to Article
Michael Bay - Monday, March 9, 2015 - link
800$ for Atom-class performance, even if it is in ITX?Someone please explain.
Ian Cutress - Monday, March 9, 2015 - link
For $600-ish, the C2750D4I is an 8-core Silvermont Atom.For $800-ish you get an 8-core Broadwell based SoC with integrated 10G networking, up to six PCIe 3.0 x4 slots and Xeon E3 v3 levels of control, all within 45W.
Sekkyo - Monday, March 9, 2015 - link
I wonder how cheap DELL and the other vendors are going to offer up Xeon D, because they could make some nice single purpose home rack servers (VPN, AD/LDAP, Mail, Compute) at pretty sweet price points.ThePegasi - Tuesday, March 10, 2015 - link
The next Microserver from HP could offer this as a higher end version, it'd be the dream homelab machine.Michael Bay - Monday, March 9, 2015 - link
I guess "within 45W" and "Atom" just don`t gel in my mind after using Baytrail for some time.CajunArson - Monday, March 9, 2015 - link
If by 3x faster than Atom with 70% better performance per watt you mean "Atom Class" then I guess...Did you bother to read this article? http://anandtech.com/show/8357/exploring-the-low-e...
Even the throwaway 2013-era Avoton parts are on par or even faster than the fastest ARM server parts that are even on the market, and ARM's own optimistic roadmaps are only promising a 50% improvement in efficiency... that would still lose to outdated Atom systems in performance/watt. The Xeon D is going to make any ARM server manufacturer's life very difficult.
Montel B - Saturday, October 10, 2015 - link
"What Xeon D also brings to the table...".Dream on....or read the 'xeon-d-1500-specification-update" that buries on their site somewhere.
It lists a bunch of errata that contradict this vaporware hype.
Most features were buggy so are disabled on the production chips in the market.:(
Even the 'good-ol' TSX-NI abracadabra magic developers like is mia (just like the old broadwells!!! surprise surprise).
The paranoid amongst us might wonder why intel seems to rely on security researchers to find design flaws on their silicone rather than revealing Intel QA findings to IT press and purchasers.
Flunk - Monday, March 9, 2015 - link
Really odd, I thought the point of Silvermont in servers was that they were going to have a lot of cores on one package (say 32 or more), 8 cores that perform worse than 4 Haswell cores with little cost or power savings doesn't make much sense to me.patrickjp93 - Sunday, March 15, 2015 - link
That's what Knight's Landing is about, except it's 72, not 32.Samus - Monday, March 9, 2015 - link
Read the article again and try to explain how $800 isn't adequately priced...this makes for a pretty powerful server (nowhere near Atom class) with 10GBE being icing on the cake. You could put together a VERY compact yet quite powerful server for under $2000 with this board, enterprise-class SSD and all.mapesdhs - Tuesday, March 10, 2015 - link
Possibly a lot of people don't realise how expensive 10GigE normally is.Ian.
Samus - Tuesday, March 10, 2015 - link
True. It's been coming down in price a little over the years but it might become mainstream relatively soon if they start putting it on entry-level server boards like this.eanazag - Tuesday, March 10, 2015 - link
Intel is not giving anything away for business class that is going to trash another Intel product line. This is between single socket desktop Xeons and Atom. If you're in the market for 10 GbE the prices are right there with better efficiency. Bear in mind there is two 10 GbE ports and Intel is pricing it because of that. They don't want to cannabalize their add-on networking.As far as performance, think laptop with 8 cores/16 threads or 4 cores / 8 threads. This has some definite possibilities, but you will need to price out total system cost versus Atom and Xeon E3 lines. It is likely too high for me in price.
spikebike - Tuesday, March 10, 2015 - link
Broadwell core (quite fast per clock) and dual 10G networking makes for quite the upgrade. Seems like a good fit for higher end NAS, network equipment (router, firewall, IDS, etc), or microservers. Imagine say 24 of these on small cards in a chassis that integrates a 10G switch inside it.icrf - Monday, March 9, 2015 - link
I'd like to see a board targeted at storage, using all those extra PCIe labs for SATA, or maybe some onboard SAS ports.liu_d - Monday, March 9, 2015 - link
Depending on how much CPU power you need, Asus makes a pretty excellent ITX board for this purpose: http://www.newegg.com/Product/Product.aspx?Item=N8...ats - Tuesday, March 10, 2015 - link
Marvell controllers are not well liked even in Linux. Its pretty much LSI or go home.wintermute000 - Tuesday, March 10, 2015 - link
The Avoton stuff already runs storage fine, this would be more than adequate. Asrock and Supermicro both have C2550/2750 boards with a dozen SATA connectors which are basically perfect for low end NAS - mine happily runs 16Tb (6x4Tb in RAIDZ2).icrf - Tuesday, March 10, 2015 - link
I'm thinking primarily storage, but not only storage. I do tend to run a single server at home, and I'd like it to run a VPN client/server, occasional x264 encoding, maybe some IP cameras or a few VMs, etc. It's a little more than a handful of Atoms can handle. The only accessory needed beyond the now standard dual GbE, though, is storage. The rest is just some extra CPU grunt, which this has plenty of.thetuna - Tuesday, March 10, 2015 - link
I'd say you want an i3 with ECC.Really, the only attractive feature I see on this Xeon D is the built-in 10GbE.
Peter Cordes - Friday, March 27, 2015 - link
Working TSX is attractive until fixed silicon appears in other models.boogerlad - Monday, March 9, 2015 - link
Why can't they release this socketed? This is the CPU I want. Perfect number of lanes, cores. Just enable overclocking and I will be happy!Samus - Monday, March 9, 2015 - link
Has a socketed XEON ever been overclockable? I don't understand how this wouldn't be fast enough for it's target market. I don't even understand how a sub-$200 Xeon like the E3-1230v3 with 8 cores at 3.3GHz wouldn't be fast enough for just about any workstation to deem overclocking necessary.MrSpadge - Tuesday, March 10, 2015 - link
More is always better in a workstation. If your workloads are running for minutes to hours, every bit counts, yet it may not be practical to transition to server clusters.BTW: E3-1230v3 has 4 cores with hyperthreading, so 8 threads. That's just plain old Haswell, a bit slower than the i7 4770.
Samus - Tuesday, March 10, 2015 - link
The i7 4770 is "plain old Haswell"The E3-1230v3 is identical to the i7 4770 except for clock speed and ECC support. IDENTICAL.
MrSpadge - Wednesday, March 11, 2015 - link
We're talking about the same thing here: when I say the chip IS Haswell I mean the same die, which you mean by calling them identical. You also acknowledge that it's lower clocked than i7 4770 - that's why they are not identical as a product and why I called it "a bit slower than i7 4770". I sure understand your post and am just replying because I'm a bit puzzled that you seemed not to understand mine the way it was intended.mapesdhs - Tuesday, March 10, 2015 - link
|Only the older models can be overclocked easily, ie. those that relied on bclk multipliers (X58and P55), or FSB before that. I have an X5570 that oc's quite well.
Ian.
Samus - Tuesday, March 10, 2015 - link
I intended to imply since Sandybridge (last 4 years) there has never been a "Xeon-K" or the like. But yes, the Core 2-based Xeon's were pretty good overclockers. X58 had some ok headroom but they ran warm stock (130w+ TDP) leaving only a few hundred MHz. I never got my i7-920 passed 3.2GHz but I did get my i7-950 to 3.5GHz.MrSpadge - Tuesday, March 10, 2015 - link
Unlock the multiplier, put it onto a mainboard which allows for at least 100 W and has regular HSF mounting points and it could see quite a few fans. Tuning such a system could be really fun! Disable the 10GBit ethernet and ECC if you're afraid of canibalizing server sales.DanNeely - Monday, March 9, 2015 - link
Has ASRock said what the 14pin power connector provides? This model isn't up on their site yet, and when I poked around I couldn't find any existing models which weren't using conventional 24+8 pin power delivery.Lonearchon - Tuesday, March 10, 2015 - link
It the main power input. Asrock has used smaller power input connector before such as MT-C224 board.DanNeely - Tuesday, March 10, 2015 - link
My question was about its pinout/etc. While the MT-C224 uses an 18 pin connector, 4 of the pins aren't used; so it's possible this is just a rationalized version of that one. That said, it appears to be playing a bit fast and loose with the AX spec. The MT-C224 identifies 9 ground pins and 5 for DC in. Looking at the pinout on the ATX PSU adapter cable from Newegg's gallery, 4 of the DC in pins are for 12V, and the 5th appears to be the PS-ON signal. This means it's playing a bit fast and loose with the ATX spec, because it's ignoring the power-good (gray wire) signal; which is normally used by the PSU to signal when its voltages have stabilized after powering on and to signal an operational fault (hopefully soon enough to keep the mobo from smoking itself). It can (probably) get away with that by waiting past the maximum in spec wait for voltage stabilization before attempting to post and by using an onboard power monitoring circuit.f0d - Tuesday, March 10, 2015 - link
what makes you think this is atom class?it uses broadwell cores at a low clock
this isnt baytrail or cherrytrail - its the full fat broadwell cores
"Xeon D, also known as Broadwell-DE, combines up to eight high performance Broadwell desktop cores and the PCH onto a single die"
as far as i can tell its broadwell-e with the southbridge on the same substrate
f0d - Tuesday, March 10, 2015 - link
was supposed to be response to michael bay but something got messed upKtracho - Tuesday, March 10, 2015 - link
I could use this in a combination NAS/PC/mini-workstation, since I unfortunately don't play games. I could run Windows for personal use and Linux for the server side, and not feel too guilty about leaving it turned on 24 hours a day.My other reaction is that combined with, say, an NVIDIA K80 or two, this would provide fantastic computing per watt, much more efficiently than a single node in the Titan at Oak Ridge National Laboratory, which uses AMD 12-16 core CPUs if I remember correctly, running at around the same frequency as this CPU, and NVIDIA Tesla cards on a PCIe Gen. 2 interface. The networking on those nodes may be a bit faster perhaps, but 10G is not bad.
MrHorizontal - Tuesday, March 10, 2015 - link
OK, so they've built the chip, it's now time to ramp up pressure on Intel to package this for laptops, and finally get rid of the 16GB RAM, no ECC and 4-core limits artificially imposed on mobile workstations.I've never understood why the Xeon E3 wasn't repackaged into a mobile form factor much like a Core i7 is...
DanaGoyette - Tuesday, March 17, 2015 - link
It seems odd that they list both VT and SR-IOV, yet don't list VT-d.How useful is SR-IOV without VT-d? (I've used VT-d, but not SR-IOV.)
Supercell99 - Monday, March 30, 2015 - link
I really don't see anything great here. 10Gbe is still expensive because of the switches, not so much the adapters. A chip that runs 8 x 2.0Ghz is really not that impressive in the server world at these prides, if it could do SMP or something then it would get some attention. But as another commenter pointed out. Intel is only competing with itself in the high end server market and they aren't going to under-cut themselves by offering something at a much lower price point with the same computational horsepower as current Xeon's.