Retailers quietly slash prices of AMD's and Intel's latest EPYC and Xeon CPUs by up to 50% — inexplicable price drops left unexplained
Let the price war begin?

Demand for high-performance server CPUs is quite high these days, both because of the AI boom and because traditional cloud service providers continue to expand their fleets. However, something unusual is happening on the retail market of data center processors in the U.S., as the latest EPYC 9005-series CPUs from AMD and Xeon 6 Performance-series CPUs from Intel are available with discounts of around 50%. We have inquired with both AMD and Intel about why their latest CPUs are available below list prices at retailers in the U.S. and are awaiting their response.
Neither AMD nor Intel recently announced price cuts for their server offerings, but as we discovered today, virtually all high-end EPYC and Xeon CPUs — including the latest generation and previous generation SKUs — are now being sold well below their prices for direct customers at retail.
If server processors from AMD and Intel were available below their list prices at a small retailer with no reputation, we would suspect the listings were merely fraud. However, two respected retailers — ShopBLT and Newegg — offer the latest EPYC and Xeon CPUs at steep discounts from their pricing guidelines, which means that these products can indeed be obtained relatively cheaply, or rather below their list prices (1000-unit prices for AMD, recommended customer price in case of Intel).
Flagship server CPUs from AMD and Intel are quite expensive: the 192-core EPYC 9965 costs $14,813, and the 128-core Xeon 6980P is priced at $12,460 (after Intel had already slashed its price from $17,800 in early January) when purchased in 1000-unit quantities by its direct customers.
However, ShopBLT is selling both processors with 35% and 53% discounts for $9,713 and $5,836, respectively.
ShopBLT is not alone, as Newegg offers Intel's Xeon 6980P for $6,190, a 50% reduction from Intel's recommended customer pricing (RCP).
Model | Retail Price | 1kU Price | Original 1kU Price | Price Per Core | Cores/Threads | Base/Boost (GHz) | TDP | L3 Cache (MB) | cTDP (W) |
---|---|---|---|---|---|---|---|---|---|
EPYC 9965 (Turin) | $9,713 | $14,813 | $14,813 | $77 | 192 / 384 | 2.25 / 3.35 | 500W | 384 | 450-500 |
EPYC 9845 (Turin) | $8,896 | $13,564 | $13,564 | $85 | 160 / 320 | 2.10 / 3.70 | 390W | 320 | 320-400 |
EPYC 9825 (Turin) | $8,030 | $13,006 | $13,006 | $90 | 144 / 288 | 2.20 / 3.70 | 390 | 384 | 320-400 |
EPYC 9755 (Turin) | $8,516 | $12,984 | $12,984 | $101 | 128 / 256 | 2.70 / 4.10 | 500 | 512 | 450-500 |
EPYC 9655 (Turin) | $6,968 | $11,852 | $11,852 | $123 | 96 / 192 | 2.60 / 4.50 | 400 | 384 | 320-400 |
EPYC 9565 (Turin) | $5,728 | $10,486 | $10,486 | $145 | 72 / 144 | 3.15 / 4.30 | 400 | 384 | 320-400 |
Row 6 - Cell 0 | Row 6 - Cell 1 | Row 6 - Cell 2 | Row 6 - Cell 3 | Row 6 - Cell 4 | Row 6 - Cell 5 | Row 6 - Cell 6 | Row 6 - Cell 7 | Row 6 - Cell 8 | Row 6 - Cell 9 |
EPYC Genoa 9654 | $6,533 | $11,805 | $11,805 | $123 | 96 / 192 | 2.4 / 3.7 | 360W | 384 | 320-400 |
EPYC Genoa 9634 | $5,538 | $10,304 | $10,304 | $123 | 84 / 168 | 2.25 / 3.7 | 290W | 384 | 240-300 |
EPYC Genoa 9554 | $5,442 | $9,087 | $9,087 | $142 | 64 / 128 | 3.1 / 3.75 | 360W | 256 | 320-400 |
AMD's top-of-the-line 192-core EPYC 9965 processor is officially priced at $14,813 when sold in batches of 1,000 units. However, ShopBLT offers it for $9,713 (about 35% below the official 1kU pricing) with an ETA of 15 days. A slightly less powerful 160-core EPYC 9845 is sold for $13,564 in 1kU quantities by AMD, but ShopBLT sells it for $8,896, a 35% reduction from AMD's 1kU pricing.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
A more mainstream 72-core EPYC 9565 can be obtained for $5,728, down 45% from the official price for direct customers. AMD's previous-generation Zen 4-based EPYC 9004-series CPUs are also sold well below their recommended prices for direct clients.
Model | Retail Price | RCP | Old RCP | Price Per Core | Cores/Threads | Base/Boost (GHz) | TDP | L3 Cache (MB) | cTDP (W) |
---|---|---|---|---|---|---|---|---|---|
Xeon 6980P (GNR) | $5,836 | $12,460 | $17,800 | $97 | 128 / 256 | 2.0 / 3.9 | 500W | 504 | - |
Xeon 6979P (GNR) | $5,166 | $11,025 | $15,750 | $92 | 120 / 240 | 2.1 / 3.9 | 500W | 504 | - |
Xeon 6978P (GNR) | ? | $11,025 | - | $92 | 120 / 240 | 2.1 / 3.9 | 500W | 504 | 400-500 |
Xeon 6972P (GNR) | $4,789 | $10,220 | $11,805 | $106 | 96 / 192 | 2.4 / 3.9 | 500W | 480 | - |
Xeon 6962P (GNR) | ? | $9,925 | - | $138 | 72 / 144 | 2.7 / 3.9 | 500W | 432 | - |
Xeon 6952P (GNR) | $4,274 | $9,115 | $11,400 | $95 | 96 / 192 | 2.1 / 3.9 | 400W | 480 | ? |
Xeon 6960P (GNR) | $4,511 | $9,625 | $13,750 | $134 | 72 / 144 | 2.7 / 3.9 | 500W | 432 | - |
Intel Xeon 8592+ (EMR) | $2,280 | $11,600 | $11,600 | $181 | 64 / 128 | 1.9 / 3.9 | 350W | 320 | - |
Intel's Xeon 6 6900P-series CPUs are sold at even steeper discounts from their RCPs. The flagship 128-core Xeon 6980P is available for $5,166 (53% lower than RCP), the 96-core Xeon 6972P is priced at $4,789 (53% lower than RCP), whereas the Xeon 6960P can be purchased for $4,511 (53% off from the RCP). The previous-generation flagship — the 64-core Xeon 8592+ — is priced at 'affordable' $2,280, which is five times less than its list price.
AMD's 1kU prices and Intel's RCPs are prices for their direct customers buying in batches of 1,000 units. In general, these are benchmarks, not fixed retail prices. Real-world prices depend on many factors, including negotiated contracts, order sizes, and distributor channel incentives, among other things, so it's quite common for large cloud service providers (CSPs) or OEMs to get their CPUs below list prices.
Retail is another matter, though. For now, we could see two possible reasons why brand-new server CPUs are available below their list prices.
Intel and AMD are in fierce competition for data center market share, which may drive prices well below official RCPs. Intel had to slash Xeon 6 pricing to stay relevant against AMD's higher-core-count EPYC 9005 CPUs earlier this year, while AMD discounts to prevent hyperscalers and enterprises from defaulting to Intel, which still dominates server chip sales. This head-to-head pressure could force distributors and retailers to undercut list prices, making high-end CPUs far cheaper in the open market.
Also, both AMD and Intel are producing CPUs aggressively to secure market share in data centers. If supply briefly overshot immediate OEM demand, distributors may be offloading excess stock into retail channels, and to move volume, they could be discounting well below list prices.
Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
-
Stomx AMD EPYC made out of many tiny less than 1 square cm chiplets while Intel has much smaller number of them so it is AMD which can substantially drop the price and even not notice that because its yield is much larger. Chiplets tech is pretty old already and well established, AMD uses it since 2015. And EPYC is also well established and reliable tech.Reply
Interesting is that if we divide the recent price TSMC charges per developed 30x30 cm waver $30k by approximately 1000 chiplets on it, we get $30 per 8-core chiplet. For 128 core processor we need 16 of them so the cost of making such processor is just ~$500, 17x less than the selling price. 2-3 years ago TSMC charged twice less but as you can see it does not matter for AMD. Sure now AMD can afford to drop the price even more and compensate this with larger selling volumes. -
jp7189 This is what makes Threadripper (esp. Pro) a non starter.Reply
The niche that can't be served by Ryzen desktop and also can't make use of EPYC's extra resources is nearly non-existant. -
Stomx Why AMD and Intel not following NVIDIA which "spices" their processors with the fast memory, packs them into bunches and sells them like hot cakes ?Reply
Difficult to pack processors and 12 channels of very bulky user configurable RDRAM memory ?
Then tell motherboard companies to "spice" their products with the fast interconnect which will allow users to buy initially 1, then upgrade to 2, then 4 or more motherboards and processors working in parallel
Compared to how successful NVIDIA is with their GPUs the CPU and motherboard companies look like the effin losers while their capabilities are actually not any worse if not better. In supercomputer world it absolutely does not matter if your computer is two times faster or slower. What matter is also your total access time which is always tiny fraction of total. -
Stomx
Threadripper is essentially same EPYC with cherrypicked parameters allowing higher clock. To get higher clock the new Threadrippers have even decreased number of cores, 96 maximum. My impression also is that it is harder to find cheap Threadripper than EPYC, am i wrong?jp7189 said:This is what makes Threadripper (esp. Pro) a non starter.
The niche that can't be served by Ryzen desktop and also can't make use of EPYC's extra resources is nearly non-existant.
If AMD made it possible for user to buy a motherboard with 1, later add into empty socket second, third and up to 8 EPYC processors (ideally like on NVIDIA masterpiece motherboard) there would be no Threadripper niche at all (it has one more restriction -- it allows only two chips work in parallel).
But unfortunately, despite EPYC allows multiprocessor configurations, currently only one or two processors can work in parallel be it EPYC or Threadripper. There are no other options on the market. Neither AMD or Intel, or any motherboard manufacturers care to get more profit. -
jp7189
You're right that Threadripper has a higher single core boost, but nearly the same if the workload is multi-threaded, so there is no practical advantage. EPYC gets the preference for better binned cores, but doesn't allow overclocking, or at least I've never seen a platform with OC options exposed, so there is the possibility of TR pulling ahead with high-end cooling and power delivery.Stomx said:Threadripper is essentially same EPYC with cherrypicked parameters allowing higher clock. To get higher clock the new Threadrippers have even decreased number of cores, 96 maximum. My impression also is that it is harder to find cheap Threadripper than EPYC, am i wrong?
If AMD made it possible for user to buy a motherboard with 1, later add into empty socket second, third and up to 8 EPYC processors (ideally like on NVIDIA masterpiece motherboard) there would be no Threadripper niche at all (it has one more restriction -- it allows only two chips work in parallel).
But unfortunately, despite EPYC allows multiprocessor configurations, currently only one or two processors can work in parallel be it EPYC or Threadripper. There are no other options on the market. Neither AMD or Intel, or any motherboard manufacturers care to get more profit.
EPYC uses a different IOD with more memory channels, more PCIe, and with interconnects that enable 2 socket operation. Despite that, EYPC motherboards sell for less than TR motherboards.
Regarding multi-socket, now that each socket is paired with 12 (or 24) memory slots, there's no room in standard form factors to do 4S. Things get exotic with ram mounted on vertical daughter boards and PCIe moved to a mezzanine; the cost to develop such a platform is high and the volume is low. AMD is content to cover the 2S market which now has access to 384 cores, 768 threads. -
reflex25 AMD's AI APUs offer great price / performance (aka bang per buck), and will likely transform the market. You may have noticed PC makers unloading stock of PCs and laptops based on earlier generation chips at knock down prices too.Reply -
93QSD5 Stomx said:Why AMD and Intel not following NVIDIA which "spices" their processors with the fast memory, packs them into bunches and sells them like hot cakes ?
Difficult to pack processors and 12 channels of very bulky user configurable RDRAM memory ?
Then tell motherboard companies to "spice" their products with the fast interconnect which will allow users to buy initially 1, then upgrade to 2, then 4 or more motherboards and processors working in parallel
Compared to how successful NVIDIA is with their GPUs the CPU and motherboard companies look like the effin losers while their capabilities are actually not any worse if not better. In supercomputer world it absolutely does not matter if your computer is two times faster or slower. What matter is also your total access time which is always tiny fraction of total.
Because NVIDIA's success is driven by hype around a segment of market that has yet to provide any tangible real world utilization.
Nothing has changed, A.I in medicine and science existed before Nvidia's hype.
It's just yet another place where big money can shove its abundant piles of untaxed cash into (same with housing btw, in case people still believe it's a natural market...).