Why was the original Phenom so bad, again?

[Phynaz]

Back in 1990 , an Intel based 486 DX2 66 PC was some 5000$...

Your point is?..

 

[Abwx]

Your point is?..

That some concurrence allowed worldwide democratisation of PCs..

Is such a point questionnable , or did it happen that it was simply not
obvious for you.?..

 

[Phynaz]

That some concurrence allowed worldwide democratisation of PCs..

Is such a point questionnable , or did it happen that it was simply not
obvious for you.?..

Not obvious to me. What was this occurrence?

 

[LOL_Wut_Axel]

High power consumption coupled with low clock speeds and somewhat lower IPC than Intel equaled horrible efficiency. That's what killed the original Phenom.

 

[Abwx]

Not obvious to me. What was this occurrence?

At a time where the CPU alone was at least 1000$ or so,
numerous manufacturers jumped in the X86 band waggon
such as the CYRIX , IBM , ST Micro, Texas Inst. and a few
others , leading to a competition that saw CPU price
drastically reduced in a few years.

As the CPUs were cheaper, the rest of the plateform
did follow the movement thanks to the subsequent
high volume induced scale savings.

 

[Phynaz]

At a time where the CPU alone was at least 1000$ or so,
numerous manufacturers jumped in the X86 band waggon
such as the CYRIX , IBM , ST Micro, Texas Inst. and a few
others , leading to a competition that saw CPU price
drastically reduced in a few years.

Hmmm.....Why did it occur at this time when they were all producing CPUs long before the 486dx2?

 

[dac7nco]

Back in 1990 , an Intel based 486 DX2 66 PC was some 5000$...

Back in 1990, the DX2 wasn't retailing. I was on a Mexicana air flight back to SF in late 1990 when I was pricing-out my i486DX/50 system, which was the fastest x86 money could buy. Clock doubling was late to the game, and it was AMD who charged more for a doubled or trippled x86 core on the same 33MHz bus... Intel's ran @ a solid 50MHz.

Daimon

Edit: My 486/50 was ~$2200, about half of my SparcStation-II. I've been in this game a long time.

 

[AnandThenMan]

You might laugh at this, but there once was a time when Britain ruled the sea and the sun never set on her imperial lands. Those people died, the people that replaced them weren't necessarily cut from the same cloth. Andy Grove and Paul Otellini won't live forever. Intel will hire their own Hector Ruiz for CEO someday, as DEC did and as Fairchild Semi did, and history will repeat itself.

I don't laugh at this at all, in fact it's a very valid point. However, just as it is very unlikely that one nation will again impose their will on much of the free world, it is very unlikely that a company will rise from the basement of some mad genius and topple the likes of Intel.

Things change.

 

[BlueBlazer]

Hmmm.....Why did it occur at this time when they were all producing CPUs long before the 486dx2?

I had a Cyrix 486DLC for a while, before jumping to a real 486.

Back in 1990, the DX2 wasn't retailing. I was on a Mexicana air flight back to SF in late 1990 when I was pricing-out my i486DX/50 system, which was the fastest x86 money could buy. Clock doubling was late to the game, and it was AMD who charged more for a doubled or trippled x86 core on the same 33MHz bus... Intel's ran @ a solid 50MHz.

That DX-50 was certainly faster then DX2-50 (had these chips before). Those DX2s and DX4s were more appealing to upgraders (upgrading from SX or DX). Intel also had their own DX2s, DX4s and OverdriveDX CPUs (Intel's own DX4 and Overdrive CPUs were shortlived). Cyrix also had DX2 and DX4s by the way. However by the time DX4-100MHz was making rounds, the first Pentiums appeared. Despite the lower clock speed, the Pentium was much faster. That took the gloss off 486s and the age of the Pentium had arrived.

 

[nenforcer]

Errata #298

 

[dac7nco]

However by the time DX4-100MHz was making rounds, the first Pentiums appeared. Despite the lower clock speed, the Pentium was much faster. That took the gloss off 486s and the age of the Pentium had arrived.

Yeah, I had a buddy who kept insisting his 486/100 was 10% faster than my Pentium 90, which is why I humor certain people here. Sometimes pity is the worst insult off all.

 

[formulav8]

Get back on track people. Stop it with the fanny stuff already.

 

[86waterpumper]

High power consumption coupled with low clock speeds and somewhat lower IPC than Intel equaled horrible efficiency. That's what killed the original Phenom.

so nothing has changed Let's hope bulldozer is good!

 

[Gundark]

There is a change from PhI to PhII, altough i don't know what is different. Using console emulators ( mostly noticeable in PCSX2 and to a less extent in Dolphin ). On the same speed PhII is generally 5-20% fatser in emulation ( depending on game ). Athlon x2 to Athlon II x2, difference is even higher, but that's due to the bigger cash. Anyway, PhII bring AMD performance in PCSX2 very close (or somewhat the same) to C2Q. I was wondering also what is different, but i couldn't find nothing meaningfull.

 

[Idontcare]

There is a change from PhI to PhII, altough i don't know what is different. Using console emulators ( mostly noticeable in PCSX2 and to a less extent in Dolphin ). On the same speed PhII is generally 5-20% fatser in emulation ( depending on game ). Athlon x2 to Athlon II x2, difference is even higher, but that's due to the bigger cash. Anyway, PhII bring AMD performance in PCSX2 very close (or somewhat the same) to C2Q. I was wondering also what is different, but i couldn't find nothing meaningfull.

IIRC the way in which the hardware-based TLB fix was implemented on Phenom/Barcelona is such that if your app was limited by L1$ read/write speeds but not dependent/limited by thread-to-thread data sharing (L2$ and L3$ speed wasn't a factor) then your code was likely to see a significant speedup in going to the PhenomII cache hierarchy because it eliminated the mandatory cache flush/eviction that was required in Phenom/Barcelona so as to mitigate the TLB bug that was present in that cache hierarchy.

I remember reading an article or two about this when the new steppings of Barcelona/Phenom came out, my memory of the details may be wrong but I'm sure you can ferret out the details with a few quick searches through Anandtech's earlier CPU reviews of the B3 stepping Phenom.

 

[Gundark]

Thanks. That makes sense. Hmm, I always tought that tlb bug was in Phenom I, and that Athlon X2 was not affected. Guess I'm wrong.

 

[Idontcare]

90nm or 65nm Athlon X2? The latency timings for the 65nm Athlon X2 were significantly relaxed compared to the 90nm chips because AMD wanted to build-in the option of expanding the size of the cache itself (an option it then never leveraged because Core 2 Duo came out and decimated the Athlon X2's performance regardless of cache sizes).

If it was the 65nm Athlon X2 -> 45nm Athlon II X2 (a 45nm stars core based chip) then the same sort of "cache enabled speedups" would have materialized as well.

Basically both the 65nm Athlon X2 as well as the 65nm Phenom/Barcelona chips had their cache hobbled for different reasons, all of which were addressed in the 45nm iteration of products.

 

[Gundark]

Yes, in my case, it was 65nm X2 to 45nm x2.

 

[nismotigerwvu]

Yeah, I had a buddy who kept insisting his 486/100 was 10% faster than my Pentium 90, which is why I humor certain people here. Sometimes pity is the worst insult off all.

Yikes, anyone remember the break even point between the DX4 (which oddly enough doesn't quadruple clock, it just triples from 33.33 to 99.99) 100 and the pentium? I know the DX4 is faster than the early 60 and 66 mhz P5's but something tells me the 75 mhz P5 was much faster. I want to say AMD slapped a "p rating" of their 133mhz 486 as p-75 but this is going really far into my way back machine.

 

[Idontcare]

Yikes, anyone remember the break even point between the DX4 (which oddly enough doesn't quadruple clock, it just triples from 33.33 to 99.99) 100 and the pentium? I know the DX4 is faster than the early 60 and 66 mhz P5's but something tells me the 75 mhz P5 was much faster. I want to say AMD slapped a "p rating" of their 133mhz 486 as p-75 but this is going really far into my way back machine.

That's correct. An AMD 486 DX4-133 was given a "P rating" of P-75 to imply it had the same performance as a 75MHz Pentium.

The P-rating was mostly true for integer dominated apps, but it lagged sorely for floating point dominated apps. The same held true of the K5, the Cyrix chip, and the NexGen chip (in addition to its successor which was turned into the AMD K6).

It wasn't until the K7 that anyone had an x86 microarchitecture that could go toe-to-toe with Intel's P6 (and PPro) FPU.

Also the DX4-133 was easy to OC to 150MHz. Cheap as chips they were, too.

 

[JimmiG]

so nothing has changed Let's hope bulldozer is good!

For a short while, the Phenom II was quite competitive with mid-range C2Q CPUs. It's not a bad architecture, it's just outdated and overdue for replacement.

The original Phenom failed for several reasons:
-Delayed

-Too little L3 cache

-Low clock speeds (originally it was supposed to launch at 3 GHz, when it finally came out, the fastest model was running at 2.3 GHz)

-The TLB bug, a minor bug which very few people would actually have noticed, which took on huge proportions in the media and on forums. What's worse, the workaround caused a big performance hit.

Without those flaws, it would still not have been a complete success, but it would have been a viable alternative to Intel Quads of the time.

 

[Rvenger]

I didn't have an issue with my Phenom 9600B chip. Yeah it was slower than Intel's Core 2 Quad but I don't think by a landslide. I disabled the TLB bug with a program someone made back in the day and I never seen a crash. One time maybe, the machine just shut off without warning. I restarted and it said "Thermal shutdown" in the event viewer. My temps were in the 30c range... so it could of been the bug but who knows.

 

[grkM3]

At the other end, competition allowed us to have really versatile
machines with relatively open standards that allow for affordable prices.

Back in 1990 , an Intel based 486 DX2 66 PC was some 5000$...

5000? Um no

I bought a brand new compaq prolinia 486 sx 25 for 1049$ and I clearly remember begging my parents to get me the dx 60 or 66mhz with the better super vga monitor for 1400