Servnet Research · Data study

Same silicon, different box

Every server comparison starts with the badge. So we took the processor and accelerator lists out of all 108 current Dell, HPE and Lenovo models — straight from the manufacturers' own documents — and checked how much of the hardware is genuinely different. The answer reframes the question: most of the time you are not choosing compute. You are choosing a chassis, a management stack and a support contract.

68%

of 239 distinct processors are sold by all three vendors

30

processors exclusive to a single vendor

8 of 25

accelerators available from all three

108

models analysed, from their own technical documents

Updated 12 September 2026 · every figure computed from the manufacturers’ own documents · method and dataset below

01 — The finding

162 of 239 processors are on all three price lists

Reduce every processor in every catalogue to the part it actually is, then ask which vendors sell it. 68% are available from all three; 47 from two; only 30 from one. The exclusives are spread evenly — Dell 7, HPE 9, Lenovo 14 — and are mostly parts tied to one platform rather than parts one vendor has locked up.

Scroll the chart sideways to see all of it →

Vendors write the same part three different ways — “Intel Xeon 6 6952P”, “Intel Xeon 6952P”, “Intel Xeon Gold 6544Y”. Matching printed names would report almost everything as exclusive, so parts are matched on maker plus part designator. The method under-counts overlap, so this is a floor.

Processor	Vendors	Models offering it
AMD EPYC 9575F	Dell · HPE · Lenovo	24
AMD EPYC 9845	Dell · HPE · Lenovo	21
AMD EPYC 9825	Dell · HPE · Lenovo	21
AMD EPYC 9475F	Dell · HPE · Lenovo	21
Intel Xeon 6767P	Dell · HPE · Lenovo	20
AMD EPYC 9455	Dell · HPE · Lenovo	20
Intel Xeon 6787P	Dell · HPE · Lenovo	19
Intel Xeon 6747P	Dell · HPE · Lenovo	19
AMD EPYC 9335	Dell · HPE · Lenovo	19
AMD EPYC 9255	Dell · HPE · Lenovo	19
AMD EPYC 9135	Dell · HPE · Lenovo	19
AMD EPYC 9745	Dell · HPE · Lenovo	19

Scroll the table sideways to see every column →

The twelve most widely offered parts. If your build is specified around one of these, the processor is not the thing you are choosing between vendors.

02 — What you are actually choosing

The chassis, the ceiling and the management stack

If the silicon is shared, the decision moves to everything around it. These are the characteristics that genuinely differ across the three catalogues, computed from the same documents.

	Models	Drive bays (max)	Drive bays (median)	DIMM slots (max)	Largest supply	Models with GPU support	Liquid-cooled models	Management
Dell PowerEdge	45	40	16	64	3200 W	35	7	iDRAC9
HPE ProLiant	32	48	8	64	3200 W	26	1	HPE iLO 7
Lenovo ThinkSystem	31	24	8	128	3800 W	26	1	XClarity Controller 3

Scroll the table sideways to see every column →

Read the maximums as “what this catalogue can reach”, not as a score — a 48-bay chassis and a 24-bay chassis are different products, not better and worse ones. The median is the more useful number for a mainstream build, and there the three are close. Where they are not close is management: iDRAC9, HPE iLO 7, XClarity Controller 3 are genuinely different products, with different licensing and different automation, and that is the choice an estate lives with for years.

03 — The part nobody standardised

The catalogue is a 25-gigabit catalogue

Across every model’s network options, 25 Gigabit is where the mass sits — 629 options against 284 at 100G, 99 at 400G and 3 at 800G. Mainstream servers have not moved to AI fabric speeds, which is why GPU clusters end up specified — and cabled — separately from the rest of the estate.

Scroll the chart sideways to see all of it →

04 — One more spread worth knowing

Memory per core runs from 4 GB to 85.3 GB

With every slot filled and the largest module each document lists, the catalogue spans 4 GB per core (ThinkEdge SE100) to 85.3 GB per core (PowerEdge MX760c), with a median of 24 GB. That is a twenty-fold spread inside the same three catalogues, and it decides whether a machine suits dense virtualisation, in-memory workloads or inference long before the badge does.

Sizing a machine for a workload rather than for a maximum? The HCI node sizing calculator works memory, cores and storage together, and the comparison pages show the ceiling for any two models side by side.

05 — Method

What was counted, and what it does not claim

How these figures were produced

Population: all 108 current Dell PowerEdge, HPE ProLiant and Lenovo ThinkSystem models for which we hold a verified specification extracted from the manufacturer's own technical document. Every processor, accelerator and network option those documents list is in scope.
Parts are identified by maker plus part designator, because the three vendors write the same processor three different ways. A part counts as shared when it appears in more than one vendor's catalogue, whatever the wording.
The sharing figures are a floor. Where a part is worded unusually enough that the designator cannot be read, it is counted as exclusive — so the true overlap is at least this high and probably higher.
Twelve processor entries produce no designator (two Pentium parts, NVIDIA Grace superchips that are not sold as standalone processors, and one Dell entry that prints only a part number). They are excluded rather than guessed at.
This is a study of what is on the price lists, not of performance. Nothing here says one vendor's implementation of a part is faster than another's; none of the three publishes comparable benchmarks for these machines.
No prices appear anywhere in this study, and no vendor sponsored it.

Download the dataset (JSON)Compare any two of them

Free to reuse with attribution to Servnet and a link to this page.

06 — Questions

Choosing between server vendors, answered

Is there any real difference between Dell, HPE and Lenovo servers?

Not in the silicon, mostly. Of 239 distinct processors across the three catalogues, 162 — 68% — are offered by all three vendors, and only 30 are exclusive to one. The differences are in the chassis: how many drives it takes, how many memory slots it has, how much power it can draw, how it is managed, and whether a liquid-cooled version exists.

Do the three vendors sell the same GPUs?

Largely. Of 25 distinct accelerators, 8 appear in all three catalogues — including the parts most AI builds are specified around. 12 are exclusive to a single vendor, which is usually a matter of which platforms have been certified rather than of exclusivity deals.

So does the server brand matter at all?

It matters for everything that is not the processor. Management tooling is the clearest example — iDRAC, iLO and XClarity are genuinely different products with different licensing and different automation stories, and an estate that standardises on one is easier to run. Drive-bay counts, memory ceilings, GPU support and the availability of a liquid-cooled variant differ per model, and support terms differ per vendor.

Which vendor has the most drive bays?

On the models in this study, Dell PowerEdge tops out at 40 bays, HPE ProLiant tops out at 48 bays, Lenovo ThinkSystem tops out at 24 bays. The medians are closer than the maximums, which is the more useful comparison for a mainstream build.

What about networking?

The catalogue's mass sits at 25 Gigabit: 629 options against 284 at 100G and only 99 at 400G. Mainstream servers have not caught up with AI fabric speeds, which is one reason GPU clusters are specified separately from the rest of the estate.

How was this measured?

Every processor and accelerator named in each model's own manufacturer document was reduced to its maker and part designator — vendors write the same part three different ways — and then matched across the three catalogues. The method under-counts overlap where a part is worded unusually, so the sharing figures are a floor, not a ceiling.

Take it further

Study: the rack runs out of power, not spaceWhat these machines ask of a rack, and how little of it fitsOpen →Head-to-head comparisonsEvery model in this study, compared spec by specOpen →All 108 data sheetsThe verified specifications this study is built fromOpen →Build oneConfigure a server against its own vendor documentOpen →