NVIDIA Levels Explained: IC1 to IC8, Pay by Job Family, and How Leveling Actually Works

No. Reported data covers IC1 through IC8, with IC8 corresponding to Senior Distinguished Engineer. A lot of guides stop at IC7, which is where the widely quoted Distinguished Engineer title sits, but the ladder formally extends one rung further.

In practice the distinction matters to almost nobody. There are so few reports at IC8 that no meaningful compensation figure exists for it, and nobody plans a career around reaching it.

This is genuinely disputed and we are not going to pretend otherwise. Levels.fyi labels IC3 as Senior Engineer, some aggregators put Senior at IC4, and NVIDIA's internal titling is reportedly loose enough that one Senior Engineer title spans both.

The practical consequence is that you should always ask for the IC number rather than relying on the title. The median gap between IC3 and IC4 is close to sixty thousand dollars in the United States, which is far too large to leave ambiguous in an offer conversation.

The gap is real and it is substantial. At IC3 a software engineer's reported median is around $319k against roughly $223k for a hardware engineer, and the difference persists at every rung until it narrows at IC6.

The cause is market competition rather than any judgement about the work. Software engineers with accelerated computing experience are being bid for by every large technology company and every well funded AI startup. The market for silicon design engineers is smaller and moves less aggressively, so the benchmark NVIDIA pays against is lower even though the work is arguably harder to hire for.

Because the money moves into stock. Reported bonuses collapse toward zero from roughly IC3 upward, and the equity share of the package climbs from under ten per cent at IC1 to well over half at IC5 and above.

This makes NVIDIA offers look weaker than they are when you compare only base and bonus against a company with a large cash bonus culture. It also means your compensation is tightly coupled to the share price, which has worked spectacularly well recently and is not a guarantee about the future.

Grants are called NSUs, or NVIDIA Stock Units, in the offer letter, and they vest over four years. The exact schedule has varied: older grants commonly vested evenly, some carry a one year cliff followed by quarterly vesting, and many recent grants are front loaded at roughly forty per cent in year one, then thirty, twenty, and ten.

Annual refreshers are granted on top based on performance and also vest over four years. From about year three these refreshers matter more to your total compensation than the joining grant did, so comparing offers on first year numbers alone will mislead you.

It is one of the better ones in the industry and it is a genuine part of compensation rather than a perk. The plan offers a fifteen per cent discount applied to the lower of the price at the start of the offering period or at the purchase date, across an offering period of roughly twenty four months with semi annual purchases.

In a rising stock that lookback provision is worth substantially more than the headline fifteen per cent, because you buy at a price fixed as far back as two years earlier. Contributions are capped at fifteen per cent of salary and subject to the statutory annual limit, so the absolute benefit is bounded, but it is real money that most offer comparisons ignore entirely.

IC3 and IC4 between them account for the bulk of experienced hiring, and IC3 has by far the most reported data of any level. If you are joining with several years of experience, that band is where the conversation is likely to start.

Getting into the upper half of that band depends less on your years and more on whether you can describe a system you owned outright, including the decisions that were yours and what happened when it failed.

Yes. The same IC numbers run across product management, technical program management, solution architecture, data science, and other functions, which is why an IC5 in one part of the company and an IC5 in another are notionally peers.

The pay attached to a given number varies considerably by function though. An IC4 product manager's reported median sits around $300k against roughly $377k for an IC4 software engineer. The level describes scope; it does not describe a single salary.

Roughly, IC1 lines up with Google L3, IC3 with L5, IC5 with L6, and IC7 with L8. The mapping is approximate and gets looser in the middle because NVIDIA's IC3 and IC4 both overlap the ground Google covers with L5.

The bigger practical difference is the shape of the package rather than the ladder. Google offers carry a meaningful cash bonus at every level, where NVIDIA pushes almost everything into equity from IC3 upward. Two offers with identical headline totals can behave very differently over four years.

No. Promotions depend on demonstrated scope, and it is entirely normal for engineers to hold the same level for many years without that indicating a problem.

The two early moves, IC1 to IC2 and IC2 to IC3, are reasonably predictable. Everything above IC4 depends on whether your organisation has problems large enough to justify the level, which is why internal transfers are a legitimate strategy rather than an admission of defeat.

No. NVIDIA does not publish a level to title map, a leveling rubric, or compensation bands, and internal promotion criteria are not shared externally.

Every figure on this page comes from crowdsourced compensation reporting and community accounts. It is reliable enough to plan a negotiation around and specific enough to be wrong in an individual case, which is why we have flagged the places where sources genuinely disagree rather than papering over them.

No, though it is more common at NVIDIA than at most software companies because a large share of the work sits close to research in graphics, silicon architecture, and machine learning systems.

What actually gates the senior levels is demonstrated impact at the required scope. A PhD helps if it produced work the field noticed; it does nothing on its own.