Tuesday, January 28th 2025

Intel Cuts Xeon 6 Prices up to 30% to Battle AMD in the Data Center

Intel has implemented substantial price cuts across its Xeon 6 "Granite Rapids" server processor lineup, marking a significant shift in its data center strategy. The reductions, quietly introduced and reflected in Intel's ARK database, come just four months after the processors' September launch. The most dramatic cut affects Intel's flagship 128-core Xeon 6980P, which saw its price drop from $17,800 by 30% to $12,460. This aggressive pricing positions the processor below AMD's competing EPYC "Turin" 9755 128-core CPU both absolute and per-core pricing, intensifying the rivalry between the two semiconductor giants. AMD's SKU at 128 cores is now pricier at $12,984, with higher core count SKUs reaching up to $14,813 for 192-core EPYC 9965 CPU based on Zen 5c core. Intel is expected to release 288-core "Sierra Forest" Xeon SKUs this quarter, so we can get an updated pricing structure and compare it to AMD.

Additionally, Intel's price adjustments extend beyond the flagship model, with three of the five Granite Rapids processors receiving substantial reductions. The 96-core Xeon 6972P and 6952P models have been marked down by 13% and 20% respectively. These cuts make Intel's offerings particularly attractive to cloud providers who prioritize core density and cost efficiency. However, Intel's competitive pricing comes with trade-offs. The higher power consumption of Intel's processors—exemplified by the 96-core Xeon 6972P's 500 W requirement, which exceeds AMD's comparable model by 100 W—could offset the initial savings through increased operational costs. Ultimately, most of the data center buildout will be won by whoever can serve the most CPU volume shipped (read wafer production capacity) and the best TCO/ROI balance, including power consumption and performance.
Source: Tom's Hardware
Add your own comment

35 Comments on Intel Cuts Xeon 6 Prices up to 30% to Battle AMD in the Data Center

#1
Vya Domus
It's wild it took them so long to finally start competing on pricing.
Posted on Reply
#2
kondamin
Vya DomusIt's wild it took them so long to finally start competing on pricing.
It’s a heck of a lot more expensive to produce one of these compared to and epyc soc, I kinda doubt those cost much more than $500 to make
Posted on Reply
#5
Daven
Slowly but surely Intel is realizing they are no longer chipzilla. Coming to this self awareness has been painfully slow to watch.
Posted on Reply
#6
_roman_
So Intel can reduce prices? really? Well after reducing the hardware quality, software quality and marketing quality they have money left to reduce the price.

Most Intel desktop branded products are in my point of view overpriced.
Posted on Reply
#7
L0stS0ul
Now Intel should cut prices of Ultra's - 30% :pimp:
Posted on Reply
#8
Chaitanya
Too little and a little to late given its not just AMD but ARM as well that has entered the competition at this point.
Posted on Reply
#9
Tek-Check
It's not enough to cut prices.

EPYC Turin actually crushes new Xeons in performance, as comprehensively measured by Phoronix, leaving Intel almost two generations behind in server CPU performance.
Posted on Reply
#10
EatingDirt
DeathtoGnomes30% cut is huge, AMD has to respond.
This just seems like Intel finally adjusting their prices because AMD and Epyc have finally reached the tipping point of market penetration and, I assume, processor availability. Meaning no one is buying the 6980P because they can get their hands on the 9755.

www.phoronix.com/review/amd-epyc-9965-9755-benchmarks

Don't have time to scroll the entire article, but it seems like the 9755 is better than the 6980P in the majority of tests.
Posted on Reply
#11
AnarchoPrimitiv
DeathtoGnomes30% cut is huge, AMD has to respond.
Do they? Zen 5 performs so much better and is more efficient, so I'm not sure if AMD has to make drastic price cuts...Intel isn't "discounting" their chips as much as they're just bringing their prices back to reality....this is how they should have been priced all along.
Posted on Reply
#12
BoggledBeagle
Change in overall pricing does not rule out further discounts to make sales happen.

I guess it appeared too stupid to give 50% discounts and thus they lowered official prices.
Posted on Reply
#13
Wirko
kondaminIt’s a heck of a lot more expensive to produce one of these compared to and epyc soc, I kinda doubt those cost much more than $500 to make
I don't know about manufacturing cost, I'll just make a comment about very large compute dice compared to AMD. Intel apparently has been smart (and knew about poor yields in advance) so it built the large Granite Rapids-AP processors with three dice with a total of 132 cores and 528 MB of L3. That's quite a redundancy. Xeons with the most cores are the 6980P (128 cores, 2.0 GHz base, $12,460) and the 6979P (120 cores, 2.1 GHz base, $11,025). Both have 504 MB of L3 and same 3.9 GHz turbo clock. If we go by those recommended prices, a customer gets 6% fewer cores, 5% higher base clock and full L3 for 11% less money. That doesn't seem bad from the customer's point of view. For Intel it means that a die with up to four defects in core logic and two defects in L3 can still be turned into a product with good margins.

Compared to that, AMD needs a huge and complex switch with 16 ports in the middle to make a 128-core Epyc. Obviously it works very well for them but their solution has its own bottlenecks, and 12-core or 16-core CCDs with a smaller, simpler IOD would make a better system overall.
Tek-CheckIt's ot enough to cut prices.

EPYC Turin actually crushes new Xeons in performance, as comprehensively measured by Phoronix, leaving Intel almost two generations behind in server CPU performance.
Xeons have those DSA/UAA/QAT/DLB/TLA/ETC/WTH accelerators, Phoronix made a few brief comments about them here but I didn't see them doing any benchmarks. I'm wondering what's the situation with accelerators, has Intel mostly given up regarding their usability or do they offer substantial support to large customers, which the general public doesn't know or care about?
Posted on Reply
#14
MacZ
I, for one, am quite happy that only totalitarian oppressive countries prevent free expression that displease them.
Posted on Reply
#15
kondamin
Intel has those dies sit on a huge interposer EMIB
WirkoCompared to that, AMD needs a huge and complex switch with 16 ports in the middle to make a 128-core Epyc. Obviously it works very well for them but their solution has its own bottlenecks, and 12-core or 16-core CCDs with a smaller, simpler IOD would make a better system overall.
That switch is relatively small
Posted on Reply
#16
ncrs
WirkoI don't know about manufacturing cost, I'll just make a comment about very large compute dice compared to AMD. Intel apparently has been smart (and knew about poor yields in advance) so it built the large Granite Rapids-AP processors with three dice with a total of 132 cores and 528 MB of L3. That's quite a redundancy. Xeons with the most cores are the 6980P (128 cores, 2.0 GHz base, $12,460) and the 6979P (120 cores, 2.1 GHz base, $11,025). Both have 504 MB of L3 and same 3.9 GHz turbo clock. If we go by those recommended prices, a customer gets 6% fewer cores, 5% higher base clock and full L3 for 11% less money. That doesn't seem bad from the customer's point of view. For Intel it means that a die with up to four defects in core logic and two defects in L3 can still be turned into a product with good margins.
Since AMD is using the same CCD design for every segment they can bin accordingly and reuse even more than Intel can with their CPU+MC die in Xeon 6. There's even an EPYCs with only 1 active core per CCD - 9175F.
WirkoCompared to that, AMD needs a huge and complex switch with 16 ports in the middle to make a 128-core Epyc. Obviously it works very well for them but their solution has its own bottlenecks, and 12-core or 16-core CCDs with a smaller, simpler IOD would make a better system overall.
Every Xeon 6 also contains two IO dies on Intel 7 process measuring 241 mm² each. EPYC Genoa/Turin IO die is around 420 mm² manufactured on TSMC N6.
The smaller IOD is available on AM5 with EPYC 4000 series which physically are basically Ryzens, but there are some features unavailable on the desktop variants. 16-core CCDs might be available in EPYC 4005 when Zen 5 enters that segment, it would actually be quite interesting to see.
WirkoXeons have those DSA/UAA/QAT/DLB/TLA/ETC/WTH accelerators, Phoronix made a few brief comments about them here but I didn't see them doing any benchmarks. I'm wondering what's the situation with accelerators, has Intel mostly given up regarding their usability or do they offer substantial support to large customers, which the general public doesn't know or care about?
Accelerators provide a tangible benefit only if software is specifically written for them. Intel is providing a lot of that, but adoption is not at the level they want it at. In order to develop and test those solutions developers have to have access to hardware containing them, and historically Intel has been segmenting their availability (not the case with Granite AP so far).
On the other hand AMD's improvements to their cores affect every piece of x86 software, and features like AVX-512 are available even in their laptop chips.
Posted on Reply
#17
MaMoo
More worried about Intel workstation / HEDT market. They are still stuck on the uncompetitive Sapphire Rapids WS at a high cost while AMD's Shimada Peak Zen 5 parts are due to arrive soon and slots into existing mobos.
Posted on Reply
#18
TristanX
Lol, Intel lowering prices on their falgship server CPU. Where is old Intel, with their absolute prices ?
Now they joined to queue of economical cost-competitive corporations, finally.
Posted on Reply
#19
Daven
Intel's main problem is that they got so far behind in core counts that they thought 128 cores was something special that could be sold for historically high margins like in the past. But AMD has been cultivating the data center CPU market for years with high core counts for two socket servers.

Epyc 7601 - 32 cores - $4,200 (2 sockets)
Epyc 7742 - 64 cores - $6,950 (2 sockets)
Epyc 7763 - 64 cores - $7,890 (2 sockets)
Epyc 9654 - 96 cores - $11,805 (2 sockets)
Epyc 9755 - 128 cores - $12,984 (2 sockets)

versus Intel

Xeon Platinum 8180M - 28 cores - $13,011 (8 sockets)
Xeon Platinum 8284 - 28 cores - $15,460 (8 sockets)
Xeon Platinum 8380 - 40 cores - $8,099 (2 sockets)
Xeon Platinum 8490H - 60 cores - $17,000 (8 sockets)
Xeon Platinum 8593Q - 64 cores - $12,400 (2 sockets)
Xeon 6 6980P - 128 cores - $17,800 (2 sockets)

Intel was charging a premium for low core counts but high socket counts but AMD destroyed the high socket count market with high core counts that Intel stopped releasing 4/8 socket products altogether. However, Intel didn't stop charging a premium for Xeon 6 as the two socket SKUs were priced like 8 socket SKUs from the past. That is until today when they realized that a 128 core count Xeon is nothing special and they can't charge a premium over the competition.

Edit: Now that I'm looking at the pricing above, it looks like Intel thought that AMD would go up in pricing between Zen 4 and 5 based on the increase in cores. AMD did so between Zen 3 and 4. This would have meant a $17,800 price for the Epyc 9755. Instead, AMD kept the pricing roughly the same. Sneaky!!!
Posted on Reply
#20
eidairaman1
The Exiled Airman
_roman_So Intel can reduce prices? really? Well after reducing the hardware quality, software quality and marketing quality they have money left to reduce the price.

Most Intel desktop branded products are in my point of view overpriced.
They always have been
Posted on Reply
#21
TheinsanegamerN
DeathtoGnomes30% cut is huge, AMD has to respond.
They'll respond by continuing to take marketshare with far superior performance.
Posted on Reply
#22
igormp
Denverwww.phoronix.com/review/intel-xeon-6980p-performance/2

The problem is that it's more competitive against Zen4 than Zen5.
So what's the point of trying to sell something inferior for much more?
If you don't require the absolute best CPU perf, Intel provides (marginally) better IO, and more scalability options up to 8S systems, which would allow one to have way more memory per node.
If IO and memory are more important for you, and you can get an intel system cheaper than an AMD one, I guess that'd be the point.

In most other cases an AMD setup would make more sense.
Posted on Reply
#24
Visible Noise
Weird, the same people that complain about Intel having high prices complain when they lower them. But don’t complain when AMD lowers prices.

I always thought competition was good.

Quite strange, hmmmm…..
Posted on Reply
#25
ncrs
igormpIf you don't require the absolute best CPU perf, Intel provides (marginally) better IO, and more scalability options up to 8S systems, which would allow one to have way more memory per node.
Only Sapphire Rapids supports 4S/8S while Emerald Rapids and Granite Rapids do not. This puts such systems at a disadvantage where performance matters since both Xeon 6 and EPYC Turin support faster RAM speeds along with better core performance.
Posted on Reply
Add your own comment
Jan 29th, 2025 21:54 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts