The issue is that none of the large OEM embrace or dare to allign with AMD on a new workstation build with EPYC or THreadripper which would result in total demolishing of the intel monopoly in this business.... no options at all
I mean if you're an enterprise buyer looking for workstations for a particular workload with a specific set of applications, you're going to buy the fastest part, not necessarily the part with the best value.
At the end of the day, the simple fact is that a large chunk of those applications still gain large benefit from Intel's clock and slight IPC lead, despite TR being able to offer great value in pure high-thread loads like rendering.
It's very good to see AMD pushing hard on the pro-level stuff, as the more common they become, the more incentive software companies like Adobe will have to better optimize for that hardware. At least we can be thankful that pressure has finally forced Intel to start adjusting instead of completely abusing their dominant market position.
If the workload can use a large number of threads, and you're not willing to spend say, $40k per workstation x the number of employees you have - even big corporations balk at that sort of thing - then AMD's chips do still have an advantage, even with slightly lower IPC.
I suspect that they are getting big discounts from Intel in exchange for both buying in bulk, and using Intel as an exclusive supplier for workstations. They also know that most enterprise customers pick a vendor such as Dell or HP and use them as an exclusive source for simplicity, support, and maintenance reasons.
I think you've got a typo in your article. You said " Intel sees the workstation market slightly differently to AMD, in that it sees workstations as mission critical hubs for commercial and corporate workflows. " but I think you meant " Intel sees the workstation market slightly differently to AMD, in that it sees an excuse for artificial market segmentation and to charge businesses extra. "
Probably still too expensive compared to the upcoming Threadripper which also supports ECC. They do have the LRDIMM support for them though, which AMD still maintains as an EPYC exclusive. Rumor's the WR8X chipset will provide such support to Threadripper though (along with 8-channel memory), so we'll see how it compares in price. Still a significant disadvantage on the PCI-E side, with less lanes at 3.0 instead of 4.0. And those quad-core SKUs... looks like Intel still hasn't completely understood that we're past that now.
Are we, though? There may still demand for low-end 'workstations' with ECC support, which you wouldn't necessarily want to do the big calculations on, but on which you might prepare scientific reports or engage in trading. Imagine if stock transactions went awry because of a flipped bit?
Maybe it's not the CPU you'd choose for yourself - increased wattage combined with decreased performance suggests these are very low-quality parts - but I could see enterprises buying them.
I work at a trading firm. No actual trading is done on their workstations, just tracking data and configuring settings. All actually trading is done on servers that are co-located at the exchange. And over the past year or so every new machine we've spun up is a custom built Threadripper machine. We would LOVE it if we could get some actual workstations from HP or Dell.
Chipset handling RAM connection days are long time gone fortunately. WR8X or not, RAM will be supported only what TR supports. It's possible that TR family would be more fragmented, but still chipset has nothing to do with RAM connected directly to CPU.
I know that CPUs have had IMCs since the last decade, but I suspect that AMD is using the chipset to enable or disable CPU features. The TRX80 & WRX80 chipsets are rumored to enable 8-channel memory, while overclocking is only available on the TRX chipsets. Those are CPU features, why would they have anything to do with which chipset it is associated with unless AMD is switching CPU features on and off depending the chipset just for market segmentation purposes?
There are rumors Intel is losing money (or barely making any) on these W series chips at these lowered prices. The size of the silicon for the W-series is huge, even with great yields they'd need a pretty high price just to break even.
Xaxaxaxa. What a ripoffs of a monopoly, These 7nm Snapdragons and Kirins with 8 cores, 10 billion transistors, built for $10-20 each, sold for $25 have 8-channel (!!!) memory while ~5-B transistor 8-core 10-12nm 4 channels Intel Neanderthals probably built for $7 each and sold for $700!
"Intel sees the workstation market slightly differently to AMD, in that it sees workstations as mission critical hubs for commercial and corporate workflows."
So AMD doesn't see them as "mission critical hubs"?
"require full uptime, full performance, and full support for a wide array of applications and hardware integration." "this level of hardware stability"
I can understand the mention about full support, but aren't the other three features offered by Core X? Are they less stable versions of the Xeons? what about AMD CPUs?
"perhaps to be in line with the competition" Does it have a name? The competition.
"Some cynics might not that this stops direct comparisons to AMD’s 16-core offering." No, it's just reality. Intel is going from 4 cores up to 18 cores with a steady step of 4 and suddently they decide that there is no reason for a 16 core? Oh, please.
PS that W 2223 looks like a terrible model of the worst kind of silicon that could not pass quality control. So they lowered the frequencies upped the voltage as much as needed and throw it out there to be used in "mission critical hubs" with a high "level of hardware stability" as the main feature.
Man, that W-2223 is getting the hate. Poor thing. But you did make me notice that its TDP is higher than its X-2225 sibling even though it has lower specs across the board. I would be curious to meet the people who buy that runt.
No, they are correct. That W 2223 looks like a low quality chip that needs extra voltage to work at those lower frequencies, compared with the3 other models. That's why it comes with a higher TDP.
Well, what's nice to see is that finally you get clocks that follow the silicon capabilities. In the old days, high core CPUs would never be allowed to clock high, even if a workload only used a couple of them: And if you wanted higher clocks, you needed to go with chips that had fewer cores enabled and thus couldn't go wide and slow, when the use case made that a better choice.
Now, no matter what your workload, you'll get whatever the silicon can deliver without exceeding the thermal limits.
Of course the next step would be the ability to control TDP or the energy budget nice and easy from command lines and library, so you datacenter control plane can optimize energy efficiency depending on the workload, slower nightly batches and better response timed during day-time peak hours. It's not just gamers who want to fiddle with clocks!
I always assumed that the fixed binning employed earlier was to prevent the thermal stress in the huge dies, when high-core islands of heat might expand and contract in seconds, causing issues with soldered heat spreaders solder bumpers on the organic substrate.
Perhaps they got better at that game or it's simply that even a 14nm 18 core is a little smaller than the 22nm Haswell equivalent.
Sadly, if you want live-migrations between hosts, it's not so easy to switch to AMD, unless you can afford to change everything at once.
The only company to really use this line up are a handful of OEM's, most notably Apple in their iMac Pro. So iMac Pro update in coming? I wonder if Apple will drop the price since the CPU side has gotten significantly cheaper.
Realistically Apple probably should hope over to the AMD side of things as announced but not yet shipping new Mac Pro is already trounced by Epyc chips for system configs that are twice as fast for nearly half the price. That makes the only interesting part of that platform ironically quad Vega 20 config using Infinity Fabric. Apple probably could negotiate with AMD to provide an Infinity Fabric link between the Epyc CPU and each GPU in such a config as each Epyc chip has four independent Infinity Fabric links.
AMD CEO Dr. Lisa Su told at the Hot Chips Keynote at 05:25PM EDT - 'These technologies might be designed for the high end HPC systems, these technologies filter down to commercial systems and next gen CPU/GPU.' But I don't expect it to be on Apple systems, but on Windows system we can expect to see next gen CPU/GPU interconnected by Infinity Fabric.
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.
Facebook
Twitter
RSS
30 Comments
Back to Article
duploxxx - Monday, October 7, 2019 - link
The issue is that none of the large OEM embrace or dare to allign with AMD on a new workstation build with EPYC or THreadripper which would result in total demolishing of the intel monopoly in this business.... no options at allduploxxx - Monday, October 7, 2019 - link
half the price... ripoff for yearsGreenReaper - Monday, October 7, 2019 - link
Ryzen's like Opteron, back in '06 or '7 — long ago, but I remember...Crimson swirl and vert;
Red and green competing.
Thirteen years of heat...
Big cats kept us gaming.
duploxxx - Tuesday, October 8, 2019 - link
stating 1TB support.... as it was deliberately removed by Intel as if it is a feature....Marketing up side down.... limit artificially something, remove limit afterwards and call it an enhancement - feature
karmapop - Tuesday, October 8, 2019 - link
I mean if you're an enterprise buyer looking for workstations for a particular workload with a specific set of applications, you're going to buy the fastest part, not necessarily the part with the best value.At the end of the day, the simple fact is that a large chunk of those applications still gain large benefit from Intel's clock and slight IPC lead, despite TR being able to offer great value in pure high-thread loads like rendering.
It's very good to see AMD pushing hard on the pro-level stuff, as the more common they become, the more incentive software companies like Adobe will have to better optimize for that hardware. At least we can be thankful that pressure has finally forced Intel to start adjusting instead of completely abusing their dominant market position.
twtech - Thursday, October 10, 2019 - link
If the workload can use a large number of threads, and you're not willing to spend say, $40k per workstation x the number of employees you have - even big corporations balk at that sort of thing - then AMD's chips do still have an advantage, even with slightly lower IPC.notashill - Monday, October 7, 2019 - link
Dell does sell some TR machines, too bad they're only in the form of ridiculous-looking "gaming" machines where TR makes no sense.edzieba - Tuesday, October 8, 2019 - link
OEMs build what they have demand for. Plus for Hyperscale deployments everyone is moving towards OCP-type custom boards and skipping OEMs entirely.twtech - Thursday, October 10, 2019 - link
I suspect that they are getting big discounts from Intel in exchange for both buying in bulk, and using Intel as an exclusive supplier for workstations. They also know that most enterprise customers pick a vendor such as Dell or HP and use them as an exclusive source for simplicity, support, and maintenance reasons.kpb321 - Monday, October 7, 2019 - link
I think you've got a typo in your article. You said " Intel sees the workstation market slightly differently to AMD, in that it sees workstations as mission critical hubs for commercial and corporate workflows. " but I think you meant " Intel sees the workstation market slightly differently to AMD, in that it sees an excuse for artificial market segmentation and to charge businesses extra. "ZoZo - Monday, October 7, 2019 - link
Probably still too expensive compared to the upcoming Threadripper which also supports ECC.They do have the LRDIMM support for them though, which AMD still maintains as an EPYC exclusive. Rumor's the WR8X chipset will provide such support to Threadripper though (along with 8-channel memory), so we'll see how it compares in price.
Still a significant disadvantage on the PCI-E side, with less lanes at 3.0 instead of 4.0.
And those quad-core SKUs... looks like Intel still hasn't completely understood that we're past that now.
GreenReaper - Monday, October 7, 2019 - link
Are we, though? There may still demand for low-end 'workstations' with ECC support, which you wouldn't necessarily want to do the big calculations on, but on which you might prepare scientific reports or engage in trading. Imagine if stock transactions went awry because of a flipped bit?Maybe it's not the CPU you'd choose for yourself - increased wattage combined with decreased performance suggests these are very low-quality parts - but I could see enterprises buying them.
jawknee530 - Monday, October 7, 2019 - link
I work at a trading firm. No actual trading is done on their workstations, just tracking data and configuring settings. All actually trading is done on servers that are co-located at the exchange. And over the past year or so every new machine we've spun up is a custom built Threadripper machine. We would LOVE it if we could get some actual workstations from HP or Dell.kgardas - Monday, October 7, 2019 - link
Chipset handling RAM connection days are long time gone fortunately. WR8X or not, RAM will be supported only what TR supports. It's possible that TR family would be more fragmented, but still chipset has nothing to do with RAM connected directly to CPU.ZoZo - Monday, October 7, 2019 - link
I know that CPUs have had IMCs since the last decade, but I suspect that AMD is using the chipset to enable or disable CPU features. The TRX80 & WRX80 chipsets are rumored to enable 8-channel memory, while overclocking is only available on the TRX chipsets. Those are CPU features, why would they have anything to do with which chipset it is associated with unless AMD is switching CPU features on and off depending the chipset just for market segmentation purposes?Supercell99 - Monday, October 7, 2019 - link
Lack of PCIe 4.0 and PCIe 128 lanes make AMD not to mention price. Intel should cut the prices in half again.rahvin - Monday, October 7, 2019 - link
There are rumors Intel is losing money (or barely making any) on these W series chips at these lowered prices. The size of the silicon for the W-series is huge, even with great yields they'd need a pretty high price just to break even.csell - Monday, October 7, 2019 - link
I wonder if Intel also will drop the prices for the Cascade Lake 3000 Series Xeon W CPUs, which Apple plans to use in the their comming Mac Pro?SanX - Monday, October 7, 2019 - link
Xaxaxaxa. What a ripoffs of a monopoly, These 7nm Snapdragons and Kirins with 8 cores, 10 billion transistors, built for $10-20 each, sold for $25 have 8-channel (!!!) memory while ~5-B transistor 8-core 10-12nm 4 channels Intel Neanderthals probably built for $7 each and sold for $700!diehardmacfan - Monday, October 7, 2019 - link
There's so much wrong in this post it hurts my brain.yannigr2 - Monday, October 7, 2019 - link
Let's see."Intel sees the workstation market slightly differently to AMD, in that it sees workstations as mission critical hubs for commercial and corporate workflows."
So AMD doesn't see them as "mission critical hubs"?
"require full uptime, full performance, and full support for a wide array of applications and hardware integration."
"this level of hardware stability"
I can understand the mention about full support, but aren't the other three features offered by Core X? Are they less stable versions of the Xeons? what about AMD CPUs?
"perhaps to be in line with the competition"
Does it have a name? The competition.
"Some cynics might not that this stops direct comparisons to AMD’s 16-core offering."
No, it's just reality. Intel is going from 4 cores up to 18 cores with a steady step of 4 and suddently they decide that there is no reason for a 16 core? Oh, please.
PS that W 2223 looks like a terrible model of the worst kind of silicon that could not pass quality control. So they lowered the frequencies upped the voltage as much as needed and throw it out there to be used in "mission critical hubs" with a high "level of hardware stability" as the main feature.
yannigr2 - Monday, October 7, 2019 - link
"a steady step of 4"a steady step of 2, not 4 obviously
ZoZo - Monday, October 7, 2019 - link
Man, that W-2223 is getting the hate. Poor thing.But you did make me notice that its TDP is higher than its X-2225 sibling even though it has lower specs across the board. I would be curious to meet the people who buy that runt.
ZoZo - Monday, October 7, 2019 - link
(typo, meant W-2225 sibling, not X-2225... where's the edit feature)abufrejoval - Monday, October 7, 2019 - link
Looks like you got the wattage for the lowest two SKUs reversed.yannigr2 - Monday, October 7, 2019 - link
No, they are correct. That W 2223 looks like a low quality chip that needs extra voltage to work at those lower frequencies, compared with the3 other models. That's why it comes with a higher TDP.abufrejoval - Monday, October 7, 2019 - link
Well, what's nice to see is that finally you get clocks that follow the silicon capabilities. In the old days, high core CPUs would never be allowed to clock high, even if a workload only used a couple of them: And if you wanted higher clocks, you needed to go with chips that had fewer cores enabled and thus couldn't go wide and slow, when the use case made that a better choice.Now, no matter what your workload, you'll get whatever the silicon can deliver without exceeding the thermal limits.
Of course the next step would be the ability to control TDP or the energy budget nice and easy from command lines and library, so you datacenter control plane can optimize energy efficiency depending on the workload, slower nightly batches and better response timed during day-time peak hours. It's not just gamers who want to fiddle with clocks!
I always assumed that the fixed binning employed earlier was to prevent the thermal stress in the huge dies, when high-core islands of heat might expand and contract in seconds, causing issues with soldered heat spreaders solder bumpers on the organic substrate.
Perhaps they got better at that game or it's simply that even a 14nm 18 core is a little smaller than the 22nm Haswell equivalent.
Sadly, if you want live-migrations between hosts, it's not so easy to switch to AMD, unless you can afford to change everything at once.
ksec - Monday, October 7, 2019 - link
It would be competitive if OEM have further 20 -30% discount from that list. Otherwise I cant think of reason to buy Intel instead of EPYC 2.Kevin G - Tuesday, October 8, 2019 - link
The only company to really use this line up are a handful of OEM's, most notably Apple in their iMac Pro. So iMac Pro update in coming? I wonder if Apple will drop the price since the CPU side has gotten significantly cheaper.Realistically Apple probably should hope over to the AMD side of things as announced but not yet shipping new Mac Pro is already trounced by Epyc chips for system configs that are twice as fast for nearly half the price. That makes the only interesting part of that platform ironically quad Vega 20 config using Infinity Fabric. Apple probably could negotiate with AMD to provide an Infinity Fabric link between the Epyc CPU and each GPU in such a config as each Epyc chip has four independent Infinity Fabric links.
csell - Tuesday, October 8, 2019 - link
AMD CEO Dr. Lisa Su told at the Hot Chips Keynote at 05:25PM EDT - 'These technologies might be designed for the high end HPC systems, these technologies filter down to commercial systems and next gen CPU/GPU.'But I don't expect it to be on Apple systems, but on Windows system we can expect to see next gen CPU/GPU interconnected by Infinity Fabric.
From this link: https://www.anandtech.com/show/14762/hot-chips-31-...