DeadTier

Device benchmark · nothing to install

What GPU does my phone have, and is it any good?

Your phone tells the browser which graphics chip it has, so it can be read without installing anything. What the name means, which class it falls in, why the answer is sometimes hidden, and why no browser can honestly benchmark a GPU.

A phone does not hide its graphics chip. When a page asks the browser to draw anything, the graphics driver identifies itself, and that name can be read without installing a thing. So the question "what GPU is in this phone" has an answer you can get in a browser, on a device you do not own, in seconds.

The question after it — is it any good — also has an answer, but a narrower one than most people expect, and the honest part of this page is the limit rather than the capability.

Where to find it

Run the test on this site and open System specs on the results screen. The row marked Hardware Class is the one you want. It shows the family the driver reported and which of four classes it belongs to: entry, mid, upper-mid or flagship.

That row is filled in whether or not you care about the rest of the page. It is not a score and it is not derived from your results — it is a name being looked up in a table.

What the name tells you

Mobile graphics chips come in a small number of families, and within a family the model number usually tracks the tier. Qualcomm's Adreno line is the clearest example: an Adreno 750 is flagship silicon, while an Adreno 610 is the entry part that ships in budget handsets. Arm's Mali line works the same way — a Mali-G57 is entry silicon.

The boundaries are narrower than they look. An Adreno 610 and an Adreno 619 sit in genuinely different tiers despite the nine between them, which is the sort of detail a rough "six-hundred series means budget" rule of thumb gets wrong. This site had to learn that one the hard way: the classifier originally treated the whole range as a single band and collapsed two real tiers into one.

What makes any of this quotable rather than invented is that the families are published by the companies that make them. You can check the claim. That is a different kind of statement from a benchmark score, and it is worth keeping the two apart in your head.

Sometimes the answer is "unknown", and that is correct

Two things commonly hide the chip. Safari reports a flat "Apple GPU" for every device it has ever run on — an iPhone from years ago and the newest iPad return the same string, so it identifies nothing. And several browsers deliberately mask the value, because a detailed hardware string is one of the components used to fingerprint people across sites.

When that happens this site says unknown rather than guessing. A masked string is not a weak signal to be squeezed for a hint; it is an absence, and treating an absence as evidence is how confident wrong answers get made.

Why no browser can properly benchmark a GPU

This is the part worth being blunt about, because plenty of pages promise otherwise.

The graphics test on this site draws twenty thousand sprites on a 2D canvas and times every frame. That measures the rasteriser and the cost of issuing draw calls, and both of those lean heavily on the processor. A phone with a weak graphics chip and a strong processor can do well at it. That is why the panel is called Render test and not a GPU score.

It has not always been called that. An earlier version of this site labelled the very same measurement "GPU Score", which was wrong, and renaming it was the fix — the number was real, the label was a claim the number could not support. A figure with the wrong name on it is worse than no figure, because everything else on the page inherits the doubt.

A real graphics benchmark pushes 3D geometry, shaders and texture bandwidth, and it needs access a web page is not given. If you want one, that is what the installable benchmarks are for. Anyone telling you a browser tab measured your GPU is telling you something the browser did not do.

What the chip class is actually used for here

The hardware class is not shown as a verdict. It is used to check the verdict — and it can disagree with it in two directions, which are not equally interesting.

If a device reports flagship silicon but measures below it, that is the finding: it is performing below what its class normally manages. A hot device does this. So does a background app eating the processor, and so does an aged battery that can no longer supply peak current. It is a real observation about this device today, and it is never proof that a phone is not what it claims to be — the ordinary explanations are far more common, and the honest claim is "performing below typical for its class".

The other direction points at this site rather than at your phone. If entry silicon somehow scores in a high band, the likeliest explanation is not that the chip is secretly fast — it is that the scale is too generous. The results screen says so in those words when it happens, because a benchmark that only doubts the device and never itself is not being careful, it is being flattering.

What to do with the answer

If you are buying, the chip class is the most durable thing on the page. Frame rates move with temperature and background load; the silicon does not change. A phone advertised as high-end whose chip reads as entry is worth a second look before you pay, and that check takes seconds on a demo unit.

If you already own the device, the useful comparison is the class against the measurement. Matching is unremarkable. A gap is a reason to look at what else is running, how warm the phone is, and how old the battery has got — in that order, because that is roughly how often each one turns out to be the cause.

Measure the device you are holding

About a minute, nothing installed, and it works on a device you do not own.

Run the test

Related