What went so horribly wrong?

the marketing department wanted bigger gigahertz so they made it reach 5GHz at the expense of everything else, also it had really bad branch prediction and inappropriately sized GIGANTIC cache lines, about 4x bigger than anything available today.
Naturally, this led to lots of cache misses, forcing dumps and rereads from main memory nonstop which made the CPU work harder for no reason.

they fucked up so bad they had to go back to pentium 3 and all their modern chips including all the core derivatives are tangibly still based on P3.

It was marketed as a CPU instead of a space heater.

Intel let the Marketing Department take over the design.

File: trumpbubble.png (243KB, 462x309px) Image search: [Google]

243KB, 462x309px

Nothing, just some hyper shilling from AMD to make lies about their great products

Plus you can't just wait for coffeelake and have access to a great line of processors that nobody will be able to compete with.
Any rational person would do that.

I remember like 2 would burn out each year in highscool. One caught fire. Then again dell was really shit back in the mid 2000s

>>61037618
Yeah, Dell learned the hard way that saving 25 cents on a PSU isn't worth it.

File: 1496931676496.png (257KB, 709x765px) Image search: [Google]

257KB, 709x765px

>>61037465
this

They left tualatin architecture

>>61037257
Bibelines.

>>61038026
someone post the picture, I cant seem to find it :(

>>61038032
I only have edits. Someone post the OG one.

It was last decade's Kaby Lake X

File: 1486572572292.jpg (314KB, 801x1500px) Image search: [Google]

314KB, 801x1500px

>>61038032

File: hyberbipeline.png (973KB, 801x1500px) Image search: [Google]

973KB, 801x1500px

>>61038070

>>61037465
Yes, everyone stop waiting for better processors, the best choice lies in the past. Sell your iCore and Ryzen systems to buy Pentium 4s.

>>61037257
What? I'm using a PC with a Pentium right now. Have had no issues and it's quick. Of course not optimal for gaming but I prefer console anyways

>>61038090
>>61038111
Ebyc

File: Happy birthday dick matt.jpg (94KB, 960x590px) Image search: [Google]

94KB, 960x590px

>>61037439
Soooooooo much win

File: download.jpg (6KB, 271x186px) Image search: [Google]

6KB, 271x186px

keller decided it was amd time again

>>61037257
They had a 4GHz part ready to ship (and probably even faster ones lined up) which they canceled to make Core 2's clock speed regression seem not as steep.

>>61037257
Literal housefire.

You could use it in summer this much time, it would shutdown itself from the heat unless you open and clean the case every fucking year

>>61037465
>>61037987
>>61038145
>>61038152
What the fuck are you guys talking about? Pentium 4 was laid to rest together with its core architecture.

The Intel chips you use today are delivered for P6 and not NetBurst.

>>61040249
Hey, Tejas was taped out. It still exists somewhere.

>>61040259
>It still exists somewhere
Nope, discounted.

>>61037418
>it had really bad branch prediction
No, branch prediction was much like anything else on the market, it's just that the mispredict penalty was so bad due to the long pipeline. If anything, that caused Intel to improve their prediction algorithms vastly, which Core could then inherit and benefit from.
>GIGANTIC cache lines, about 4x bigger than anything available today
That's just untrue. It used the same 32-byte (later 64-byte) lines that everything else at the time used and still use today.
However, the D-cache was pretty small, only 16 kB even in later models (8 kB at first) at a time when AMD had 64 kB L1D. Probably done in order to get the clock up.

What really sucked were, rather, mostly the following:
The trace-cache wasn't too great, often being very poorly utilized due to its internal organization. Even worse, for code that wasn't in the trace cache, decoding was only one single instruction per clock cycle, seriously hampering performance on anything that is not inner loops.
The replay mechanism sucked very badly, wasting tons of cycles and resources whenever a cache-miss occurred. On a related note, the OoO logic couldn't cancel already queued instructions after a branch mispredict, they had to be executed and reach retirement in order to be cancelled, worsening the mispredict penalty to far worse than just a pipeline flush.
Add to that that many instructions that weren't very simple ALU instructions had almost weirdly long latencies. Agner Fog even speculates that the MMX units ran at half clock speed.

Ultimately, everything about the processor was focused at reaching abnormally high clock speeds. It's often said that they believed they could reach 10 GHz. I don't know if they actually believed this or if it was just a marketing stunt, but had they been able to, it would probably have sucked a lot less than it actually did.

>>61037257
longer bibeline :DDDDDDDDD

>>61040581
ebin :-------DDDDDDDDDDD

>>61040515
>everything about the processor was focused at reaching abnormally high clock speeds
Why did they design it so that parts of it had to be clocked twice as high? Wouldn't that limit potential increases?

>>61040684
it did
but intels marketing wanted moah niggahurtz

>>61040684
Not necessarily, it was just the simple ALUs that were double-pumped, and since they have rather simple datapaths, it never was a limiting factor, even late in the design. The critical paths are elsewhere.

>>61040710
>but intels marketing wanted moah niggahurtz
The double-pumping of the ALUs was very seldom advertised, so I actually don't think that was the reason.

Nothing went wrong. AMD is still a newbie in the industry, and that actually works in their favor, because 'gamers' are conditioned to believe 'more is better' : more FPS, more resolution, more refresh rate. And ultimately that mentality is what AMD uses to (unintentionally) take advantage of you. Most people looking to buy a new processor for gaming are usually coming from the dual core era of Intel. It's more than common knowledge that the newer quad core Intel chips perform better, and now that they see Ryzen 7, the same cost of the Intel flagship, with EIGHT CORES, they fall into a trap of believing that it's somehow 2x better than the Intel because it has 2x the cores.

One thing I'd also like to point out is that most modern games are more GPU intensive than anything, and if you're main reason for buying a PC is to play games, buy a cheaper CPU and use the extra money for a better graphics card, you only need a good CPU for productivity.

>>61040829
>Intel fanboy can't into context
No surprises there.

>bendium 4 had HT
>later only the lowest end and highest end intel processors had it

wath went wrong

>>61040870
Nothing went wrong. AMD is still a newbie in the industry, and that actually works in their favor, because 'gamers' are conditioned to believe 'more is better' : more FPS, more resolution, more refresh rate. And ultimately that mentality is what AMD uses to (unintentionally) take advantage of you. Most people looking to buy a new processor for gaming are usually coming from the dual core era of Intel. It's more than common knowledge that the newer quad core Intel chips perform better, and now that they see Ryzen 7, the same cost of the Intel flagship, with EIGHT CORES, they fall into a trap of believing that it's somehow 2x better than the Intel because it has 2x the cores.

One thing I'd also like to point out is that most modern games are more GPU intensive than anything, and if you're main reason for buying a PC is to play games, buy a cheaper CPU and use the extra money for a better graphics card, you only need a good CPU for productivity.

>>61040915
>AMD is still a newbie in the industry
Oh, I see. I thought it was AMD that manufactured Intel's x86 chips when they couldn't fulfill the demand back when Intel was just a few year old company. Also couldn't be that AMD made the 64-bit extensions we are using today, I must be remembering shit wrong.

You must be right, no way a dumbass kid/hipster on /g/ could be wrong.

>>61040829
>>61040915
Why the fuck are you spamming the thread with this garbage

>>61040849
>>>61040829 (You)
>>Intel fanboy can't into context
>No surprises there.

Easy to throw labels around when you read something that doesn't line up with your childish worldview.

I use a Ryzen 7 rig at home for gaming, I'm happy with it because it does what I bought it for : gaming.

The Ryzen is pathetic for work. Software compatibility issues, RAM issues, freezes and shutdowns under stress.... which is why I have an Intel rig for that, and I could write it off as a business expense, so no money out of my pocket there.

>>61040956
Except everything that you said is not only wrong, but completely irrelevant to the topic at hand. Just kill yourself retard.

>>61040952
4chans been doing this for a while for me

halp

File: 3452.png (75KB, 450x650px) Image search: [Google]

75KB, 450x650px

I HURT MYSELF TODAY

>>61040982
Daily reminder that Pentium 4 was never as far behind Athlon 64 as Athlon 64 was behind Core 2

>>61038026
>>61040581
>bibelines
Daily reminder that Wilamette was 20 stages, which is about as long as both Skylake and Zen.

>>61041297
So was Northwood, but then they decided to go for super-ultra-mega-ebin-deep bibeline.

File: 1491911677612.png (162KB, 633x900px) Image search: [Google]

162KB, 633x900px

>>61040982
DELET

>>61040249
I was being sarcastic anon. You can tell by the obvious sarcasm.

>>61041297
Skylake is 14-16 stages. Zen is 19

>>61040982
>MOAR GIGAHERTZ

>>61041581
>Skylake is 14-16 stages.
Only when running via the µop cache.

>>61037257
Pentium 4 was fine, pentium D, not so much.

>>61043030
multi core processors were a mistake

File: celery x.png (123KB, 679x214px) Image search: [Google]

123KB, 679x214px

>>61044338
Coming soon ..

File: 2017-02-26 07.25.31.jpg (1MB, 2592x1456px) Image search: [Google]

1MB, 2592x1456px

>>61040982
I got an OG FX-51 sitting in my junk parts box. I want a board for it but have had absolutely 0 luck in finding a compatible one.

>>61044372
if memory was dual channel i'd unironically buy this for dwarf fortress

>>61045400
Isnt Dorf Fortress cache, memory access, and latency limited more than anything else?

>>61045442
technically yes, but single core performance helps a lot

If there was something dual core at crazy clockspeeds with a lot of L3 cache and quad channel RAM i'd buy it, even if it needs phase change cooling

File: 8568585.jpg (1MB, 4996x2436px) Image search: [Google]

1MB, 4996x2436px

good old times, sitting on the shelf for nostalgia

>>61040515
The simplest ALUs ran at twice the clock speed, but they couldn't do anything but the simplest instructions

>>61045515
Well, there was that one time intel put out 4.6ghz dual core westmere chips for financial shit. You can find them on ebay but be prepared to shell out a pretty penny because they are not cheap (at least $300 a chip)

my desktop currently runs on a pentium 4, it's really nice in the winter

>>61045583
but how's the single threaded performance compared to modern cpus? DF just makes use of cache and ram speeds more if it's not bottlenecked by the processor itself

>>61045710
Well, I can tell you that my 4ghz hex core Xeon (westmere) approaches single thread parity +/- a few % with my 3.3ghz Octo Xeon (Sandy Bridge E). Make what you will of that.

I'm trying to find cinebench screenshots of 32nm westmeres clocked to those speeds that show the single thread score, but not having much luck with that, and my own will not clock that high and remain stable.

>>61045980
don't worry anon, if it's too hard to find you don't have to bother, this is just speculation anyways seeing that I can't afford shit right now

>>61046385
Heh, especially since you'd likely have better results shelling out the money for a 3960x and board, then overclocking those to the moon vs getting a "pre overclocked" dual core and having to deal with trying to find boards that will handle it. X58 board be expensive as hell man.

File: 1481915065259.jpg (52KB, 932x576px) Image search: [Google]

52KB, 932x576px

>>61040915
>AMD is still a newbie in the industry,

Advanced Micro Devices, Inc.
Founded May 1, 1969; 48 years ago

Intel Corporation
Founded July 18, 1968; 48 years

Really makes you think.