Intel is Serving Major Xeon Discounts to Combat AMD EPYC

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
22,027
11,607
136
The only one with single channel is the Lenovo Ideapad 330, all the rest is dual channel...

https://geizhals.de/?cat=nb&xf=9690_Raven+Ridge

I haven't looked at any German sites recently, but I did look at Dell's desktop lineup, and they still haven't put AM4 Raven Ridge in any of their systems, for anything. All you can get is Stoney Ridge, an A10-9700 (yes, that's the only BR they sell), or Ryzen. And it's all 1x8GB by default. I didn't look seriously at their notebooks, mind.

To get back OT no sure that Intel s quasi unlawfull behaviour wont backfire at some point, server CPUs discount here are not made at the OEM level but at the retail one, as if Bestbuy would make a discount paid by Intel on a laptop to a consumer that would want to buy a Raven Ridge equipped laptop..

As others have articulated here, backroom deals seem to be de rigueur in the server world. The only thing that can/will stop Intel from shutting down EPYC sales may well be their inability to serve the market with product. People are not going to care what deals Intel can offer them if they are looking at months worth of delays to get access to hardware.

ARM is almost non-existent at the moment and has other issues such as the switch from x86 to ARM which might frighten some companies so I'm not sure Intel has to offer anything here :) OTOH, as we all know, Intel heavily subsidized companies that took Intel chips for tablets.

I was thinking about the historical RISC players in the server market (MIPS, POWER, SPARC, etc.). But after all they might have collapsed just because they were too expansive and didn't (couldn't) properly react to Intel prices, even official prices.

Remember what happened to Qualcomm? They came out big with an impressive first-effort chip, only for Qualcomm to bag on the entire operation and shut it down due to cost concerns. It looked like they were ready to compete (alongside Cavium, albeit with chips that were sufficiently dissimilar that the two required different support within the Linux kernel to function properly). Then all of a sudden, splat, no more Qualcomm server chips. The funny thing is that the server chip was probably the first thing Hock Tan would have cut from the operational budget had Broadcom bought out Qualcomm. When the Feds killed that buyout, it looked like the server chip had a chance of making at run on the server market. Nope! Didn't happen. Now it's down to Cavium.

But between Cavium, MIPS, IBM POWER, and whatever is left of SPARC, I would give Cavium either the best or second best chance of taking share from x86 in the server market. And I would only say second best when taking OpenPOWER into account.

TBH, the biggest market for this kind of cpu is cloud providers, and they already have major discount, and intel also sold it directly to them.

Hard to tell. ODM sales are shrouded in mystery. A lot of the cloud providers get their hardware this way.

AFAIK, the 14nm >10nm transition was started and caused a few fabs to be in the process of being upgraded plus chipsets are starting to being fabbed on 14nm. Rumor (?) has it that Intel will be using TSMC for some product.

This 10nm fiasco is a lot more convoluted than just a lack of 10nm CPUs.

edit:
Also, the last 14nm process for Coffee Lake is rumored to be less dense than previous versions to sustain higher clocks. Result is less die/wafer. I suspect the latest 14nm process is even less dense for the i9 5GHz clocks.

I was aware of the density issue, but still, Intel should have had all their 32nm and 22nm capacity switched to 14nm by now. If Intel is relying on "low density" 14nm for their future Xeons (Cascade Lake, etc) then they are getting fewer dice per wafer. In fact, Intel seems to be relying on 14nm++/14nm+++ for everything now . . .

And I was not aware that they had actually cut down on the number of fabs they had tooled for 14nm as a part of their botched 10nm rollout. What a fiasco.

Virtually all of their products are on one node now, this is the first time in recent history this has been the case for Intel. Plus the die sizes of products are getting bigger; they sell quad cores when they used to sell duals for instance; Skylake Server is considerably bigger than Broadwell-EP..

See above, though I will repeat that Intel "should have" been just fine if all their 32nm and 22nm capacity had switched to 14nm by now. Seems like they were not prepared to lean on the 14nm node for this long.
 

Nothingness

Diamond Member
Jul 3, 2013
3,063
2,047
136
Remember what happened to Qualcomm? They came out big with an impressive first-effort chip, only for Qualcomm to bag on the entire operation and shut it down due to cost concerns. It looked like they were ready to compete (alongside Cavium, albeit with chips that were sufficiently dissimilar that the two required different support within the Linux kernel to function properly). Then all of a sudden, splat, no more Qualcomm server chips. The funny thing is that the server chip was probably the first thing Hock Tan would have cut from the operational budget had Broadcom bought out Qualcomm. When the Feds killed that buyout, it looked like the server chip had a chance of making at run on the server market. Nope! Didn't happen. Now it's down to Cavium.
I don't think Intel had a lot to do with Qualcomm demise in the server business. I'm afraid it's just them wanting to cut costs where something could not bring enough money in a short time. That's utterly stupid, they had a good chip (though it was single socket only, the second version should support two sockets, and might be released) and you don't take a market by storm with your first try, no matter how good that try is. Perhaps they realized they didn't have enough money to sustain a multi year effort. Quite disappointing.

But between Cavium, MIPS, IBM POWER, and whatever is left of SPARC, I would give Cavium either the best or second best chance of taking share from x86 in the server market. And I would only say second best when taking OpenPOWER into account.
Just a note: Cavium has been bought by Marvell.

They have some good support at the moment (HPE, Cray). Let's see how that turns out.
 

R0H1T

Platinum Member
Jan 12, 2013
2,582
162
106
That's a risk indeed. But you could also see things optimistically: AMD is good enough to present a danger (and Intel management publicly admitted it some weeks ago) that needs to be countered. And if AMD is that good then some will buy their server chips.
Well they said the same about the current POTUS, how did that turn out for US of A?

The P&N forum is down the hall,
and to the left.

Leave political discussion out of the
tech areas.

AT Mod Usandthem
 
Last edited by a moderator:
Mar 10, 2006
11,715
2,012
126
Remember what happened to Qualcomm? They came out big with an impressive first-effort chip, only for Qualcomm to bag on the entire operation and shut it down due to cost concerns. It looked like they were ready to compete (alongside Cavium, albeit with chips that were sufficiently dissimilar that the two required different support within the Linux kernel to function properly). Then all of a sudden, splat, no more Qualcomm server chips. The funny thing is that the server chip was probably the first thing Hock Tan would have cut from the operational budget had Broadcom bought out Qualcomm. When the Feds killed that buyout, it looked like the server chip had a chance of making at run on the server market. Nope! Didn't happen. Now it's down to Cavium.

One of the first things that Hock Tan did when Broadcom and Avago merged was to sell original Broadcom's Vulcan CPU efforts (now known as Cavium Thunder X2) to Cavium :)
 

moinmoin

Diamond Member
Jun 1, 2017
5,064
8,032
136
Reading the (for this particular article surprisingly many) comments at STH I think this is seriously good news in disguise to AMD. Intel doesn't change the MSRP, just quote depending on customers' intend. To get any discounts customers have at the very least to plan with AMD, Intel will likely increase the requirements or reduce the discounts the more customers start fishing for them. Of course it's not ideal for AMD for them toying with AMD chips but ending up in Intel's lap again. But the more publicity this story gets the more customers will put AMD on their map at least for trying to get the discounts, many of them previously completely ignoring AMD. This is a win in mindshare for AMD, something that will bear fruits if they are able to keep executing their ambitious roadmap where Intel continues to stagnate.
 

DrMrLordX

Lifer
Apr 27, 2000
22,027
11,607
136
I don't think Intel had a lot to do with Qualcomm demise in the server business. I'm afraid it's just them wanting to cut costs where something could not bring enough money in a short time. That's utterly stupid, they had a good chip (though it was single socket only, the second version should support two sockets, and might be released) and you don't take a market by storm with your first try, no matter how good that try is. Perhaps they realized they didn't have enough money to sustain a multi year effort. Quite disappointing.

I agree that it was disappointing. Perhaps they set unrealistic goals for adoption of their early chips. Regardless, I can't help but wonder . . . there seem to be a lot of server room guys out there pulling for an alternative like ARM to take the market by storm. For whatever reason, OpenPOWER isn't doing it for them.

Just a note: Cavium has been bought by Marvell.

They have some good support at the moment (HPE, Cray). Let's see how that turns out.

I'll be watching, to the extent that I'm able.

One of the first things that Hock Tan did when Broadcom and Avago merged was to sell original Broadcom's Vulcan CPU efforts (now known as Cavium Thunder X2) to Cavium :)

Not surprising. Tan doesn't seem to want anything to do with the server market.
 

slashy16

Member
Mar 24, 2017
151
59
71
Even before these price cuts the cost of a dual socket server filled up was about the same whether you chose Intel or AMD. Intel has massive brand recognition which isn't going away. Open any Dell or HP server and you will see Intel CPU's, chipsets, storage controllers, HBA cards, NIC cards and storage. I know management at our company is weary of AMD servers parts. Many felt burned when they switched to Opteron based servers only for their support to be shit because of their small market share.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,552
10,171
126
And I was not aware that they had actually cut down on the number of fabs they had tooled for 14nm as a part of their botched 10nm rollout. What a fiasco.
Seems... slightly weird that they would do that, shut down 14nm current production, to get their 10nm line(s) operating, because of the fact, that not too long ago, they actually mothballed some of their fabs that were scheduled for expansion years ago, so, they should have had the facilities and physical space to build out 10nm, without cutting into 14nm production. Also, what doesn't make sense, is Intel (rarely) ever shut down lines in the past, I think, they just moved them to other products, that didn't require cutting-edge lithography. Also, isn't Intel moving chipsets to 14nm too? This all seems so... strange.
 

maddie

Diamond Member
Jul 18, 2010
4,881
4,951
136
Seems... slightly weird that they would do that, shut down 14nm current production, to get their 10nm line(s) operating, because of the fact, that not too long ago, they actually mothballed some of their fabs that were scheduled for expansion years ago, so, they should have had the facilities and physical space to build out 10nm, without cutting into 14nm production. Also, what doesn't make sense, is Intel (rarely) ever shut down lines in the past, I think, they just moved them to other products, that didn't require cutting-edge lithography. Also, isn't Intel moving chipsets to 14nm too? This all seems so... strange.
That was a reply to what I posted, but AFAIK they didn't cut 14nm to go to 10nm, but they are migrating die from larger nodes to smaller ones. Chipsets, etc are going to smaller nodes and thus the overall demand for 22 and 14 are increasing without more fabs coming online as some of the 14nm products should have migrated by now to 10nm thus leaving more available space at 14nm. Having to keep fabbing CPUs on 14nm is causing the production bottleneck.
 
  • Like
Reactions: Nothingness

Abwx

Lifer
Apr 2, 2011
11,542
4,325
136
Some new old news...

AMD counters and is extremely confident that Intel will have to do much more in the future than just adjust prices to a minimum in order to " take the roadmap's weaknesses ", AMD explains. The goal for this year therefore remains to nibble about five percent of Intel's formerly over 99 percent market share.

https://translate.google.fr/translate?sl=de&tl=en&js=y&prev=_t&hl=fr&ie=UTF-8&u=https://www.computerbase.de/2018-09/amd-epyc-cpu-rome/&edit-text=

4-630.3191044902.png


7-630.1097020912.png
 
  • Like
Reactions: lightmanek

Abwx

Lifer
Apr 2, 2011
11,542
4,325
136
Well now that is some absurd problem, you get a discount price but Xeons are in the "jumping and high-speeding fog".

https://semiaccurate.com/2018/09/07/intel-cant-supply-14nm-xeons-hpe-directly-recommends-amd-epyc/

HP Enterprise (HPE) has recently been said to have recommended its partners adopt its server products using AMD's platforms to avoid the impact of Intel processor shortages. But Digitimes sources from the upstream supply chain have indicated that no other server players have seen issues with supply of Intel's server processors


https://www.digitimes.com/news/a20180920PD201.html?chid=9