Wednesday, November 23rd 2016

AMD and NVIDIA Add-in-Board GPU Market Share from 2002 to Q3/2016

The folks over at 3dcenter.org have compiled comprehensive historical GPU AIB market share data for our digestion. While we recently reported on Q3'16 and its comparison to the quarter before and the same period last year, this information spans a near 14 year quarter-on-quarter time frame. The compilers have quite helpfully included points of reference along the timeline which highlight the two major GPU manufacturers milestone desktop product line debuts.

It is worth noting that their exact numbers differ slightly to the ones Jon Peddie Research provided as 3dcenter have also cited the work of Mercury Research, which appears more conservative. The figures provided in their own graph split the difference between the two sources to give us a more impartial look at the market.

Back in 2002, the two GPU manufacturers were in the earlier stages of their rivalry to dominate the desktop graphics card market. A typical PC powered by either would have utilized an "ATI" (Before their purchase by AMD in 06') Radeon 8/9000 Series or GeForce 2 MX/Ti series graphics card. The average amount of on-board video memory was 64 or 128MB accompanied by 2, 4 or if you were lucky, 8 pixel shader units. The most popular screen resolution was 1024x768, and probably on a CRT monitor. Leading API's at the time were DirectX 8.1 and OpenGL 1.3.

Fast forward to anno 2016 and some 10-15 generations of GPU later, your average on-board video memory is closer to 4 or 8GB, and now far more complex clusters of computational units that have superseded shaders number in the thousands. A typical gamers screen resolution is 1920x1080 or in many cases, beyond - not to mention the advent of VR. Both major API's are evolving too, now aiming to have much more efficient low level hardware access and less resource overhead with Microsoft's DirectX 12 and the Successor to OpenGL 4.5, Vulkan.
Add your own comment

45 Comments on AMD and NVIDIA Add-in-Board GPU Market Share from 2002 to Q3/2016

#26
bug
Ac
Brusfantomet
G92 used in:
8800GTS
9600GTX
GTS250

That is as bad as AMD have been, yet because of the Nvidia marketing it started the Ati/AMD decline.
Actually, that's not as bad as AMD. 8800GTX was G80, built on 90nm, 9600GTX was G92(4?) built on 65 or 55nm. Like I said, more like a tick-tock strategy. In taking GPUs and bringing them to a new node, some of them kept existing configurations. However, virtually the entire lineup consisted of a single generation. When AMD mixes various GCN revisions under the same series, it means you actually get different capabilities from different GPUs (as in one offers hardware decoding for something while the next one may not).
What was horrible was the naming. I'm a hardware enthusiast and I could never find my way around the GT, GTS, GTX, GS, GSO and GTO nomenclature. That was a clusterfuck by itself.
Posted on Reply
#27
bug
john_
Nvidia is doing it with it's pockets full. AMD's pockets are full of debts. I love how people don't give any excuses to a company that is fighting two fronts, at the same time, against stronger companies and with billions in debts and at the same time they will find plenty of excuses for companies that have the upper hand and billions of cash to spare. And as always the GT730 example is ignored because (usual excuse) it can't run modern titles at 4K @60fps with everything at ultra. I mean we have an example where a company sells three totally different cards under the same name, and NO ONE CARES. But AMD, that sells cards under different names and at price points where the cards are competitive if not better value options, it's guilty.

Anyway I stop here before a certain moderator starts getting worried.
As if it's my problem AMD got themselves into that position.

And I don't think this is a case of cutting anyone slack. Users unrealistically expect entire new lineups each year. That's not economically feasible and it doesn't happen. But of the two strategies, I find Nvidia's less confusing (at least for non-techies), that's all.
Posted on Reply
#28
Ubersonic
bug
What was horrible was the naming. I'm a hardware enthusiast and I could never find my way around the GT, GTS, GTX, GS, GSO and GTO nomenclature.
Basically: GTX > GTS > GT > GS/GSO.

The 7900GTO was a downclocked 7900GTX sold on the cheap to clear out the stock before the 8800 made it worthless.
Posted on Reply
#29
cdawall
where the hell are my stars
bug
Ac

Actually, that's not as bad as AMD. 8800GTX was G80, built on 90nm, 9600GTX was G92(4?) built on 65 or 55nm. Like I said, more like a tick-tock strategy. In taking GPUs and bringing them to a new node, some of them kept existing configurations. However, virtually the entire lineup consisted of a single generation. When AMD mixes various GCN revisions under the same series, it means you actually get different capabilities from different GPUs (as in one offers hardware decoding for something while the next one may not).
What was horrible was the naming. I'm a hardware enthusiast and I could never find my way around the GT, GTS, GTX, GS, GSO and GTO nomenclature. That was a clusterfuck by itself.
The mixing gcn in the same line up yet keep ignoring nvidia doing essentially the same thing

8800GTS 320/640MB vs the 8800GTS 512. Same line up, same name, different chip. The 8800GTS 512 then became the 9800GTX, 9800GTX+, GTS250 and GTS250G. Nvidia actually took something sold and marketed as a upper midrange and convinced everyone it was a top tier card the next generation. I'm not even mad that's impressive, but on that note if AMD would have taken the R9 290 and resold it as the R9 390x you would have ostracized them and accused them of everything under the sun.

Move onto a couple gens later and we have the 580 vs the 480 which was purely a single SP unlocked and a mature process.

This isn't a new trick. Companies rebrand crap, hell nvidia still sells the 8400GS and god knows what GPU it's actually based off this week.
Posted on Reply
#30
evernessince
chaosmassive
that's what you got if you try to rebrand same things, especially twice !
Only the 7970 was rebranded twice. This is far from abnormal in the graphics card world. The R9 390 is a single rebrand and the R9 380 is a newer Tonga based GPU.
Posted on Reply
#31
cdawall
where the hell are my stars
evernessince
Only the 7970 was rebranded twice. This is far from abnormal in the graphics card world. The R9 390 is a single rebrand and the R9 380 is a newer Tonga based GPU.
To be fair the Pitcairn silicone was used in three generations. 7850/7870->270/270x->370/370x
Posted on Reply
#32
sliderider
If AMD doesn't do something about the 460's lack of performance relative to GTX 1050 and 1050 Ti, that gap is going to get even bigger. nVidia's partners are going to be peddling those cards to the low end of the market as fast as they can make them and 460 is going to get squashed.
Posted on Reply
#33
FreedomEclipse
~Technological Technocrat~
john_
I expected that with HD 4870 and HD 4850 AMD would have gained more market share back then, but it seems that people where hypnotized with the Nvidia logo much earlier than what I thought.
I had 4870s in CF. They ran hot but i utterly loved them. Got about 830mhz on the core when it came to overclocking. Not sure about the ram speed though, bit too long ago for me to remember
Posted on Reply
#34
ShurikN
sliderider
If AMD doesn't do something about the 460's lack of performance relative to GTX 1050 and 1050 Ti, that gap is going to get even bigger. nVidia's partners are going to be peddling those cards to the low end of the market as fast as they can make them and 460 is going to get squashed.
Isn't there like a 470LE or smth in the works
Posted on Reply
#35
cdawall
where the hell are my stars
ShurikN
Isn't there like a 470LE or smth in the works
470D
Posted on Reply
#36
Jurassic1024
So much for the RX 480 dominating the mainstream market.
Posted on Reply
#37
Jurassic1024
evernessince
Only the 7970 was rebranded twice. This is far from abnormal in AMD's world.
*fixed*

nVIDIA doesn't rebrand flagships.
Posted on Reply
#38
cdawall
where the hell are my stars
Jurassic1024
*fixed*

nVIDIA doesn't rebrand flagships.
9800GTX and GTX+ would have been just as flagship as the 7970 was to AMD. That card from nvidia was already just a rebranded 8800GTS 512 and rebranded into the GTS250 after. SO what were you saying? Let me guess the 9800GX2 was the flagship? Wait that was just two underclocked 8800GTS 512's.
Posted on Reply
#39
Jurassic1024
cdawall
9800GTX and GTX+ would have been just as flagship as the 7970 was to AMD. That card from nvidia was already just a rebranded 8800GTS 512 and rebranded into the GTS250 after. SO what were you saying? Let me guess the 9800GX2 was the flagship? Wait that was just two underclocked 8800GTS 512's.
Interesting, because...

A very smart man at Intel once told me that when designing a microprocessor you can either build a new architecture, or move to a smaller manufacturing process, but you don't do both at the same time. The reason you don't do both is because it significantly complicates the design, validation and manufacturing processes - you want to instead limit the number of variables you're changing in order to guarantee a quick ramp up and good yields of your silicon.

NVIDIA followed this rule of thumb with the GT200, building its "brand new" (or at least significantly evolved) architecture on a tried-and-true 65nm process instead of starting at 55nm. Despite AMD building both RV670 and the new RV770 GPU on TSMC's 55nm process, NVIDIA hadn't built anything on a smaller than 65nm process, including the 1.4 billion transistor GT200.
-Anandtech
Posted on Reply
#40
cdawall
where the hell are my stars
Jurassic1024
Interesting, because...

A very smart man at Intel once told me that when designing a microprocessor you can either build a new architecture, or move to a smaller manufacturing process, but you don't do both at the same time. The reason you don't do both is because it significantly complicates the design, validation and manufacturing processes - you want to instead limit the number of variables you're changing in order to guarantee a quick ramp up and good yields of your silicon.

NVIDIA followed this rule of thumb with the GT200, building its "brand new" (or at least significantly evolved) architecture on a tried-and-true 65nm process instead of starting at 55nm. Despite AMD building both RV670 and the new RV770 GPU on TSMC's 55nm process, NVIDIA hadn't built anything on a smaller than 65nm process, including the 1.4 billion transistor GT200.
-Anandtech
That quote is literally worthless in this context. It has zilch to do with nvidia doing the exact same thing as amd, actually worse I don't remember amd rebranding a mid range high end as a flagship card in their entire history as a company.

Also for note a wise man once said the sun revolved around the earth.

Not to mention amd and nvidia just moved new gpus to a brand new process.
Posted on Reply
#41
synapse
Everyone should pray for a 50/50~ market, if AMD goes bankrupt we are f******, Nvidia & Intel can do w/e they want with their pricing.

You should thank AMD instead of bashing.
Posted on Reply
#42
cdawall
where the hell are my stars
synapse
Everyone should pray for a 50/50~ market, if AMD go bankrupt we are basically fucked, Nvidia & Intel can do w/e they want with their pricing.

You should thank AMD instead of bashing.
AMD can go bankrupt and it wont change things. People forget how important it is to have a competitor the term monopoly gets dropped and then Intel is split into 6-8 companies. Intel doesn't want that and would bail AMD out to save themselves.
Posted on Reply
#43
john_
cdawall
AMD can go bankrupt and it wont change things. People forget how important it is to have a competitor the term monopoly gets dropped and then Intel is split into 6-8 companies. Intel doesn't want that and would bail AMD out to save themselves.
Maybe 10 years ago, but not today, or tomorrow.

Intel will not split for the same reason Nvidia is not going to split. We could also say that Nvidia should split because it will be the only remaining company selling DirectX compatible discrete GPUs into the x86 market. But no one says anything about Nvidia. Why? Because none of those companies will have any problems with, for example, FTC. None will split. The reason why we think that Intel would split, is because we are remembering this argument from a long time ago, when the market was completely different.

Intel will just point the finger to the ARM platform.
Nvidia will just point the finger to Intel's 70% market share when counting integrated GPUs.


Also 50%-50% is not the best percentage we should hope for. No, the best is 40% AMD - 60% Nvidia. Because AMD, if it doesn't build another bulldozer again with Zen, will gain GPU market share anyway, thanks to the APUs. So to have a balance, we need Nvidia to remain the stronger GPU company in the discrete GPU market. Just not in the extend it is now. We need a more balanced market where AMD has at least 25% of the CPU market, sells plenty of APUs and also it is as competitive to Nvidia as it was a few years ago, before the Maxwell series.
Posted on Reply
#44
cdawall
where the hell are my stars
Intel will not be able to point at arm. They are not in the desktop market, this works in the same way Microsoft had to save apple since it couldn't point at linux.
Posted on Reply
#45
bug
synapse
Everyone should pray for a 50/50~ market, if AMD goes bankrupt we are f******, Nvidia & Intel can do w/e they want with their pricing.

You should thank AMD instead of bashing.
Nvidia isn't stupid. Why do you think they left the whole console market to AMD for an entire generation? ;)
Posted on Reply
Add your own comment