July Steam Survey: RTX 5000 surge, new top GPU, 4 in 10 participants using AMD CPUs

Posted by gurugabrielpradipaka@reddit | hardware | View on Reddit | 83 comments

[-]

Firefox72@reddit

Its probably been since like 2007 since AMD last had a 40% CPU share on Steam.

Although i'm not sure the Steam survey even existed then. The wayback machine only goes back to late 2008 with AMD then having 35% which obv slipped more and more to the eventual low point of 8.2% in November 2017.

[-]

dudemanguy301@reddit

If anything this just shows how slow change is in consumer habbits. It took AMD almost a decade to slip from 35% to 8% even though the were hardly competitive for most of that time outside of maybe some well priced Phenom II parts in the early 2010's. And then it took them almost 8 years of consistently quality chips to get from that low point back to 40%

This is why I always roll my eyes when I see the sentiment “there’s no point in competing, AMD had the best card and people just bought Nvidia”, it takes multiple generations of excellence to move the needle, you can’t just have one off victories and expect a landslide shift in market share.

[-]

Any-Ingenuity2770@reddit

this also majority takes at most 8 years to upgrade their kit

[-]

Earthborn92@reddit

CPU Upgrades are also generally slower than GPU upgrades.

[-]

Jon_TWR@reddit

I’m on a 5700x3d and 4080 Super…I’ll probably upgrade my CPU, mobo, and RAM when AM6 comes out (assuming it’s using DDR6). Then my GPU, when I financially recover from that upgrade…and the new 80 series GPU isn’t single digit percent points better than the last one.

[-]

Pijany_Matematyk767@reddit

>and the new 80 series GPU isn’t single digit percent points better than the last one.

TBH i dont really get this argument. If youre upgrading from a 4080 why would you care about what the difference is between a 7080 and an 8080? The difference between 4080 and 8080 and the 8080's actual performance in software would be the only relevant things

If the gpu market by then is any similiar to the current by the time 8080s are available 7080s will be either out of stock or available at basically the same prices as their next-gen counterparts so its not even like picking between upgrading to 7080 or 8080 would make sense

[-]

PaulTheMerc@reddit

he would have the option of 7080 used vs 8080 new, which would make it relevant.

[-]

Strazdas1@reddit

used comes with its own can of worms most people dont want to deal with.

[-]

SovietMacguyver@reddit

Because it matters of you are being utterly shafted, perhaps?

[-]

Toojara@reddit

Because the difference compounds and the higher the starting point is the more relevant small differences are.

As an example:

	Performance	Perf over 4080
4080	1.00	0
5080	1.10	10%
6080	1.38	38%
7080	1.52	52%
8080	1.67	67%

The increase from the 7080 to 8080 is just 10% but is makes a large difference in the overall upgrade.

[-]

Jon_TWR@reddit

Well, I’m upgrading from a 4080 Super, and I’ve already seen barely any improvement from my card to the 5080. If there’s similarly low improvement moving forward, that means it will be even longer before I upgrade. It’s a pretty simple and straightforward concept, what don’t you get?

[-]

Dave_Wein@reddit

How is it still only 40%. I don’t know anyone building PCs with intel

[-]

teutorix_aleria@reddit

It's a sampled survey of all machines on steam

It includes the entire steam userbase not just new machines/logins for the month. So even fairly large changes in current market share would not reflect on the hardware survey as similarly large gains.
Again it includes all machines on steam, which means laptops, OEM PCs, and other prebuilts, all of which skew heavily towards intel.

[-]

Pimpmuckl@reddit

Importantly, something most users here completely miss:

It includes Chinese LAN Cafes.

There is some severe sampling issue in some of these data points. Remember when Windows XP had some insane spike? Or when the Chinese language suddenly went to like 50% or so in February's survey?

Commenters and tech "press" loves steam hardware surveys to extrapolate some bullshit so they can run headlines that have nothing to do with reality.

Where as this was never intended to be a true representation of the global PC user base, it was mostly for Valve internally to verify trends and do some soft market research without even trying to be some huge ground truth.

[-]

SunfireGaren@reddit

If people are gaming on Steam in Chinese LAN cafes, why would you want to exclude them from the survey?

[-]

Pimpmuckl@reddit

They absolutely should to be included.

But if you have almost random spikes where your chinese language doubles in one sampling month, then something is wrong with the data collection.

[-]

PaulTheMerc@reddit

they have the power, just poll every account that has logged in in the last 3 months.

[-]

teutorix_aleria@reddit

Why would they, its not intended to be market research. It's a tool for valve and developers to see what hardware users actually have, not what's selling well.

[-]

INITMalcanis@reddit

Amongst people who make a specific decision about buying a CPU, AMD is probably at >90%

But most PC are prebuilts sold by OEMs.

[-]

UsernameAvaylable@reddit

I don’t know anyone building PCs with intel

Well, guess what, they are bot building PCs, they are just buying them. Or rather, most of them are buying laptops.

[-]

CapsicumIsWoeful@reddit

Laptops

[-]

MeedLT@reddit

because new systems make up a minority of these surveys.

[-]

Sevastous-of-Caria@reddit

And oems still dish out intel chips to laptops

[-]

wankthisway@reddit

People are probably still using the same machines from years ago, and the vast majority of laptops had Intel chips for the last decade or so.

[-]

Jon_TWR@reddit

Prebuilts.

[-]

rubiconlexicon@reddit

If anything this just shows how slow change is in consumer habbits.

Possibly, but I think there's also an effect of the Intel numbers being inflated and sticky due to all the ancient Intel prebuilts out there. I'd bet if you could somehow filter for just enthusiast DIY builds, AMD would be a lot higher than 40% right now.

[-]

996forever@reddit

That’s not “inflated”, that’s simply the reality of the pc world being OEM dominated. Even if you had somehow had a way to filter by enthusiast DIY builds you would just be living in a bubble if you do that.

[-]

jasonwc@reddit

I generally agree with your point, but many visually-demanding AAA titles have minimum and recommended targets that exclude a significant number of older PCs. Indiana Jones and the Great Circle and Doom: The Dark Ages both require hardware accelerated RT for ray-traced global illumination, so you need an RTX 2000/RX 6000 or newer GPU. There are other games that don't have such a hard cutoff but, in practice, require a relatively new GPU and 8GB+ VRAM for a decent experience. Newer DIY builds, and even prebuilts, are more likely to have Ryzen CPUs. In contrast, the most popular live service games are designed to run on almost any hardware, so even a relatively ancient prebuilt with a discrete GPU would likely be fine.

The Steam hardware survey provides a picture regarding the entire Steam user base, but that's not necessarily the target for every PC release. It shows 29.3% of Steam users have 4 GB of VRAM, or less, which will preclude most recent and even some older AAA titles, and around 33.5% have 6 GB of VRAM or less. Several games, including Indiana Jones, require a minimum of 8 GB of VRAM, and plenty require 8 GB, or more, for a decent experience.

[-]

UsernameAvaylable@reddit

Its probably been since like 2007 since AMD last had a 40% CPU share on Steam.

No, it was never that high. Back when there was peak athlon reign intel still had all their retailers 100% in their grasp.

[-]

SmartOpinion69@reddit

well the PC community is always asking "should i upgrade?" while the casual community will only upgrade when their rig is actually too slow or if it stops working. what amd and intel has been doing as of late is beating each other by small margins, but if you want to dominate your share, i think you need to beat the competition so hard that lenovo/dell/hp puts your chip in low supply

[-]

DXPower@reddit

The Steam Survey was created to aide the development of Half Life 2, so I think its earliest rendition would have been around 2002 or so. Valve later made it public, though I'm not sure when that was.

[-]

jeremiah_wright_@reddit

Although i'm not sure the Steam survey even existed then.

it's been around since at least 2005 since I remember getting it in the app but they only showed you results on steam back then, it wasn't a standalone web page you could view whenever you want until later.

[-]

nukleabomb@reddit

Lovelace's increasing popularity illustrates how the cards are finally seeing their stock levels improve. More importantly, there are also some cards available at their MSRP, including the RTX 5070 on Newegg ($549).

Beyond the prices, Lovelace cards have faced plenty of criticism for their minor performance upgrades compared to the previous-gen equivalents. But it appears plenty of people who are upgrading are still opting for Nvidia's latest and most powerful products.

Lovelace??

Do they mean Blackwell?

[-]

Pijany_Matematyk767@reddit

ChatGPT constantly confuses the 2 and says Lovelace when asked about 5000-series cards

[-]

VastTension6022@reddit

Is it because the old data it's trained on says lovelace is the newest arch or is there something else too?

[-]

Pijany_Matematyk767@reddit

Yeah, it had a lot of training data from when lovelace was the newest and if asked about newest nvidia cards will often talk about 4000-series

If you ask it something about blackwell cards it will often say that theyre hypothetical and speculate on their specs instead of printing out the actual specs of the real cards

it has the same problem with amd 9000-series cards, treats them as hypotheticals

gotta start the chat by telling it to look up the specs of these cards for it to start treating them as real and giving some data about them

[-]

Pimpmuckl@reddit

The knowledge cut-off is pretty evident.

If those journalists would suck at their jobs, they could feed this type of "base information" into the LLM which then would generate better (not perfect, but at least a bit better) articles.

But that would require manual work.

So ya.

[-]

Techhead7890@reddit

Editor? Never heard of her!?

Vanishes from internet

[-]

Pijany_Matematyk767@reddit

>If those journalists would suck at their jobs, they could feed this type of "base information" into the LLM which then would generate better (not perfect, but at least a bit better) articles.

Or they could like... write the damn article? I dont need them to copy paste what the ai told them, if i wanted to get info from an ai id have asked the ai myself, all theyre doing is just collecting ad revenue in exchange for doing nothing of value

[-]

Pimpmuckl@reddit

Of course, but that would require even more work, don't be ridiculous.

[-]

PaulTheMerc@reddit

claude seemed to not have an issue(just tested it) though it did refer to it as just Ada initially.

[-]

shroudedwolf51@reddit

The fun thing about regurgitative "AI" is that you could ask it and it gets the question right and you could ask it again and it might get it right, might tell you the opposite, or might tell you something entirely else.

[-]

Pimpmuckl@reddit

That depends on the temperature setting used.

You can have almost full determinism with a zero setting (based on same hardware, floating point errors etc).

But then you'd have to actually use the API to request that and not just chug shit into a web prompt.

[-]

angry_RL_player@reddit

AI generated article

[-]

GOMADGains@reddit

It's not a smoking gun, but if you use the date range and search for the editors name, Rob Thubron, he seems to mysteriously have fallen in love with em dashes after 2021 on his techspot articles.

Weird huh...?

[-]

Ethrealin@reddit

Not a smoking gun indeed, since ChatGPT launched in Nov 2022 and only really gained writer traction in winter.

[-]

LaM3a@reddit

And ChatGPT didn't invent em dashes, journalists are completely right to use them.

In a casual reddit comment, now that's weirder.

[-]

Strazdas1@reddit

i use them on comments too, i just like them. I guess i just learned from the books i read?

[-]

comperr@reddit

alright, that's it. Journalism is extinct. You can't convince me AI can put together a nice article about something new. All it can do is jumble together some words about the past. Previously written words.

[-]

angry_RL_player@reddit

Speaks volumes that these people can't even be bothered to correct it either.

[-]

Strazdas1@reddit

proofread it? we fired the editor in 2015.

[-]

Not_Daijoubu@reddit

The obvious lack of human effort from beginning to end is such a terrible issue, because LLMs can perform well with strategic guidance. These "journalists" are probably putting out 1st drafts from ChatGPT after only prompting them "Make an article about this info I ripped from another website".

[-]

nukleabomb@reddit

Isn't techspot related to HUB?

I had higher hopes for them.

[-]

Carrot2244@reddit

No, not really related to them. They are freelance writers for Techspot.

[-]

N1NJ4W4RR10R_@reddit

The Hardware Unboxed guys just contribute to it with articles as far as I know. Wouldn't have any control over what other writers contribute.

[-]

psi-storm@reddit

The 7600XT suddenly appears this month, 18 months after it's release, and the 7800XT doubled it's numbers from April to July even though everyone is buying the new cards instead. Yes, these numbers totally make sense.

[-]

Strazdas1@reddit

people selling off their old cards used.

[-]

_elijahwright@reddit

16 GB being the most used RAM config is very telling for DDR4 lol. but I will say that it is going down

[-]

No-Dust3658@reddit

Why get more than 16GB?

[-]

Bluemischief123@reddit

Why do you think 16GB of RAM these days is good to have? Literally 0 reasons to limit yourself.

[-]

DirectFrontier@reddit

~~4 GB~~ ~~8 GB~~ ~~16GB~~ 32 GB will defo be enough for your lifetime!!

[-]

Bluemischief123@reddit

Ah well. Remember when SSDs were becoming more and more popular so you'd get a tiny SSD to load windows onto and an HDD for everything else. Now almost every game is a required SSD and some people lose their minds over that lol.

[-]

DirectFrontier@reddit

Oh shit I remember that.

[-]

No-Dust3658@reddit

Because I have been using 16GB for like 6 years and no game yet has maxed it out

[-]

Professional-Tear996@reddit

Some interesting data points:

AMD CPU share went up by 0.74%. SSE4a was up by 0.73% but AVX-512F was up by only 0.45%. Meaning 40% of AMD CPUs that were added this month are Zen 3 or older.
10 core and 20 core CPUs was up slightly - most likely Intel is doing well with the the 265K price drop and some of the non-K Raptor Lake i5s.
CPUs with 14, 16 and 24 cores all went down by the highest amount (except for 6-core, but I would think that is due to people upgrading to higher core count). People are really getting rid of 14th gen and older Intel CPUs.
The difference between 6 and 8 core CPUs in the share of the user base is less than 5%. I would think it would take at least one year for 8-cores to be the most popular CPU core count that gamers use. Just in time for next year's CPU launches.

[-]

Alive_Worth_2032@reddit

and some of the non-K Raptor Lake i5s.

Also 14700K since we are talking 20 cores. If you have a slower LGA1700 CPU (like those non K i5s) and want to do a in socket upgrade. Then the 14700K is what makes sense as the "EOL upgrade" for the platform, especially for gamers where the 14900K offers almost nothing extra.

[-]

996forever@reddit

And the 14700HX which is/was quite common and upper mid range gaming laptops.

[-]

Alive_Worth_2032@reddit

Ye, Intel is even supply constrained for 7nm since RPL is selling better than anticipated at this point in its lifespan. Gaming laptops is a one reason since RPL is very competitive in that segment still.

Since AMD doesn't have a mobile X3D chip and is mostly just found in bulky desktop replacements. And mobile Zen 4 and 5 doesn't really hold up as well as their desktop counterparts in gaming performance. Since they cut out some of the L3 for mobile.

[-]

996forever@reddit

AMD's monolithic APUs have always been terrible for dGPU gaming. Even against intel 14nm, Cezanne lost to the ancient i7-10870H in dGPU gaming. Now Strix Point not only has cut down L3 but also extremely high inter core and inter CCX latency. To the point the Razer Blade 16 with 160w GPU can lose to Zephyrus G16 (ARL-H45) with 120w GPU at 1200p gaming.

Just terrible.

[-]

Professional-Tear996@reddit

I don't think that is true because CPUs with HT are also down. People are more likely to ditch the platform rather than upgrade if they had got into it with Alder Lake unless they happen to have a high-end motherboard.

[-]

Alive_Worth_2032@reddit

There's far more people with HT than just RPL. All the HT decline means is that there's more people abandoning older Intel platforms than investing into new HT platforms.

Someone moving from a i3 12100 to a 14700k will no affect the HT statistics at all. Since they are just moving from HT to HT.

[-]

Professional-Tear996@reddit

They aren't moving up from Alder and Raptor Lake i5 to i7 or i9.

And the HT loss is compensated for in the 20-core CPU category, which could mean that 265K bundles are selling better than expected.

People forget that Alder and Raptor Lake had DDR4 support and entry-level customers were more likely to have DDR4 boards, so upgrading from an i5 12400 or 13400 to a 14700K is not going to help at all if you are going to be stuck with DDR4 and 40-50 GB/s of memory bandwidth for a 20-core CPU.

[-]