A64 Performance

[Beerknurd]

I have absolutely no idea what anyone is talking about.... I don't even have an amd processor, therefor I have no idea what an A64 is. All I know is Pentium. I would like to have an amd processor, because I hear yall talk about how good they are. I have one of the fastest processors on the market, 3.4 ghtz with HT technology. There are only 3 pentium processors on the market that are faster than mine, 3.2 extreme, 3.4 extreme, and they just released a 3.6. I enjoy helping out because I have been on a couple of teams and you guys really care about what yall are doing and have alot of teamwork. So even though I don't know half... who am I kidding 3/4 of what yall are saying I still am honored to be on the team.

[BC]

I have absolutely no idea what anyone is talking about.... I don't even have an amd processor, therefor I have no idea what an A64 is. All I know is Pentium. I would like to have an amd processor, because I hear yall talk about how good they are. I have one of the fastest processors on the market, 3.4 ghtz with HT technology. There are only 3 pentium processors on the market that are faster than mine, 3.2 extreme, 3.4 extreme, and they just released a 3.6. I enjoy helping out because I have been on a couple of teams and you guys really care about what yall are doing and have alot of teamwork. So even though I don't know half... who am I kidding 3/4 of what yall are saying I still am honored to be on the team.

Beer,
To put things in perspective... here's the basics..

a) Intel chips take 6 or 8 'clock cycles' per instruction to execute (that's an actual 'add' or 'multiply'). AMD only needs 4. Both companies add 1 cycle for the processor to reset & fetch & do internal housekeeping. The math results in Intel needs multiple of 9 (or 3 since 3 is the prime #) and AMD runs on multiples of 5 (already a prime #). So simply forget Megahertz or Gigahertz in figuring performance. Use the number to help do the proper math for buying parts.

b)
AMD started with a quad bus architecture in the athlon 1200.. It also had 3 8086 instruction decoders and 128K L1 cache (64k inst / 64k data) and 512K of L2 cache and you could run TWO of them (the MP version) on DDR 266 memory with no problems.... By comparison, at the same time Intel had the P4 1.7 with 1 8086 instruction decoder, 1 internal bus, 8k of data cache and a 'fancy' 20k of 'micro instruction' cache. It also required a 400Mhz RAMBUS to keep the processor from starving for data or instructions.

FYI 20k Microinstruction cache is meaningless because it deals with the internal execution of an instruction... it has NOTHING to do with your program UNLESS you execute the exact same 'add' or 'multiply' or whatever over and over.


So, you ask........ Why the reason for the vast difference? AMD knows how to make more efficient circuits... there are fewer transistors internally and the logic needed is much simpler. Intel had to give up the cache because it would not fit on the chip... period. The Intel P4 power (heat) rating was originally "79 watts (typical)"... notice it didn't say "Max" ??? The AMD Athlon as "54-55 (i forget the #) MAX"... BIG difference.

RAMBUS is fast ***IF*** you are doing sequential memory access ONLY. A normal computer, running multiple programs (like windows) still starves on Rambus. I will share the internals of Rambus if you like.


b)
Intel finally figured out how to get away from Rambus and use a different chipset (north & south bridge). They can now use cheaper, more efficient memory by going back to their Pentium III design and combining its technology into the P4. This is now the P4A and P4B series.

Again, FYI: The P-III at 1.0 Ghz could outrun a P4 at 1.7. When I benchmarked two P3's at 1.0 Ghz (dual cpu mode), they both ran fine... dual P4s had NO performance gain.... Rambus and even the new memory chipset could not feed it fast enough.


c)
Now, skip forward into the Pentium HT's and Athlon XPs....


The HT is a dual bus (they tried to mimmick AMDs design of quad) but could only get 1 more bus on the chip. The XP was an outright improvement over the original Athlon design... The circuits were smaller AND engineering figured out how to keep the cache from buring up.. (Yes, guys... that is what happened... the first athlons were supposed to be 128/128 L1, 512k L2 cache, but the heat was too much and it melted down)

So, now we have HT's... trying to keep the CPU 100% busy, but still missing essential pieces (8086 decoders and an output data bus *AND* cache.. the processor sits and 'spins', looking busy 100% busy, waiting for memory or something else). The XP internally got smaller, allowing it to go faster. There are other changes in the XP, but not worth mentioning right now.


d)
Where do we stand today? Intel bought the DEC Alpha chip and created the IA-32/64... A design that is 15+ years old at least. It is entirely different and not compatible in any way with Pentium. You know it as Xeon. zero cross compatibility.

AMD pulled the same trick as another company did 20 years ago when going from 16 to 32 bits... they put BOTH on the chip because there was room. This is why the Athlon64 runs in both 32 and 64 bit worlds.

What about clock speeds & all that? the math is simple... the math is 5/3. that is 1.6... this is telling you that AMD is 1.6x better at doing the same thing as Intel in the 32 bit world.

The Athlon64 3200+..... runs at 2.0Ghz.... 2.0 x 1.6 = 3200... Presto!

e)
Now, there is one fundamental thing left... Memory.... and that is based totally on this 5 vs 9 (or 5 vs 3) ratio.

AMD decided to go with DDR-400... so 5 became 10.... 200FSB uses DDR 400 (5x2 and 200x2)..... 200 FSB x 10 (AMDs magic number) = 2.0 Ghz.

Intel does the exact same thing... take a DDR 333 on a Pentium 4b with a 533 FSB... watch the math and you will laugh. The only difference is intel has to play with both the numbers 9 AND 3 because they still think of 133 in their heads (that's why the P3 w/ DDR-266 is so fast). NOW, take 266*2 and you get 533... PRESTO again!

In *MY* opinion, Intel has not yet learned how to make it better, just emphasized on making it smaller & faster. The original design is still there with more *stuff* piled on.

(( side note: The first 80286, 80386 and 80486 burned up because of heat! ))


back to AMD. AMD stays at 10... but we have all found that we can use '5' when we want to. But that is where DDR-400 came from...

The Athlon64s simply cut 400FSB back to 200 (and grab twice as much).. so they grab the same amount as an FSB 400. They do it as 2x2x200.

If you have heard of 'dual channel'... now you can take out one of those '2x' and that is why they are so fast... 2 seperate channels at the same time.

I will stop here... Hope this helps. I'm sure it'll lead to questions about why DDR-333 works , etc etc.... (answer about DDR-333... usually CL2.5, it works because the PCI bus is 33mhz or AGP is 66... 333/5 = 66. 333/10 = 33.. presto) Quiz: what XP processors like DDR-333, what is their REAL frequency and what is the math. LOL


F)
Last, BUT BY FAR LEAST... Beerknurd, you want to learn... you've got my support... Learn and grow with us.. and as your Intels need replacing, you know what to buy.


BC

[Beerknurd]

:shock: That was the most informative answer I have ever heard. I'm not worthy!!!! You are a computer genius. You have information ooozing out of your pores. Very impressive.

:notworthy: I bow to you.

[Keith75]

Wow BC, you must love to type. Anyway, maybe you can answer a question I have. If you did in the previous post forgive me cause I missed it or didn't understand it. LOL Why are the Intels memory running Quad Pumped while the AMDs run just Double speed. If Quad pumping works so well why wouldn't AMD use it?

Keith

[BC]

Beer,
Please don't bow... I'm not worthy of such..

I'm just sharing 25+ years of experience. I used to wire backplanes
together to make computers bigger and used to change the microcode
chips to change the processor type... (yes we used to do that stuff).

BC.

[em99010pepe]

I think the administrators should create a doubts room called:

"Take your doubts with BC"

Your are amazing.

[BC]

Wow BC, you must love to type. Anyway, maybe you can answer a question I have. If you did in the previous post forgive me cause I missed it or didn't understand it. LOL Why are the Intels memory running Quad Pumped while the AMDs run just Double speed. If Quad pumping works so well why wouldn't AMD use it?

Keith

I dont' realy love to type... it takes forever to fix the typos (spinal injury here that affects arms), and voice synth isn't reliable yet. As far as technology and architectures go it's my job... I am a computer engineer... i can hook up a logic analyzer or pull up the software debugger. That is how I was trained.

( Per the Intel site, and the P4 with HT Extreme with 2MB L3 cache )
800 MHz, 533 MHz or 400 MHz System Bus
In the Pentium 4 processor with 800 MHz system bus, the bus supports high performance by delivering up to 6.4 GB of data-per-second into and out of the processor. This is accomplished through a physical signaling scheme of quad-pumping the data transfers over a 200 MHz clocked system bus and a buffering scheme allowing for sustained 800 MHz data transfers. The same quad-pumping of data is true for Pentium 4 processors with 533 MHz system bus (4.2 GB data-per-second with a 133 MHz clocked system bus) and Pentium 4 processors with 400 MHz system bus (3.2 GB data-per-second with a 100 MHz clocked system bus). This compares to 1.06 GB/s delivered on the Pentium® III processor's 133 MHz system bus.

Also note,,, from the same page:

Level 1 Execution Trace Cache
The 90 nm Pentium 4 processor features 16-KB data cache compared to 8-KB on the 0.13 micron

Also on the same page... the L2 cache is:
Non-Blocking, full speed, on-die level 2 cache
8-way set associativity
256-bit data bus to the level 2 cache
Data clocked into and out of the cache every clock cycle

(( intel had to unveil its .090 technology now... AMD is waiting until .065 is ready))

Now, the simple AMD answer....

AMD64 939pin 130nm (which *can* be as memory thirsty as a pentium HT Extreme and up) but uses industry standard Hypertransport technology ( http://www.hypertransport.org ) instead of L3 cache and Quad pumping at the moment.


The specs for Hypertransport are:
Feature/Function
HyperTransport Technology
Bus Type Dual, unidirectional, point-to-point links
Link Width 2, 4, 8, 16, or 32 bits
Protocol Packet-based, with packets in multiples of four bytes (32 bits)
Bandwidth
(Each Direction) 100 MBytes to 11.2 GB/sec.
Data Throughput Up to 22.4 Gigabytes/sec.
Signaling 1.2-V Low-Voltage Differential Signaling (LVDS) with a 100-ohm differential impedance
Multiprocessing Support Yes
Memory Model Coherent and non-coherent

1.2 LVDS is your quad pump
(( Athlon64's and up aleady have his capability ))


Notice that AMD is not talking 256 bits yet? but my previous posts said 512/1024??? , nor am I talking L3 cache which CAN be predictive and access memory every cycle *If* it wants to starve the rest.

Now, who's faster? Intel by forcing "Up to 6.4GB/sec" just to keep it running or AMD by having the architecture that is industry standard that goes above that (up to 22.4GB) and not needing all that for the processor.... ALSO, AMD isn't using a L3 cache yet either..... Intel is pretty desperate if they must resort to quad channel equivalent (400x2x4) to keep up.

One last little piece of info.... Intel is running Quad at 32 bits.... AMD is running Dual at 64 bits.... Notice I didn't say anythng about the new AMD processors either (next month)...... Yet the Intel P4 HT Extreme is their best....

Please forgive what looks like anti-Intel bias... but this P4 Ht Extreme is their latest, you have not seen AMD's matching processor yet. I have also commented on the fact that Intel appears to build more on top of what they already have instead of fixing the core....

AMD is not like that.... Please, just sit back, watch and keep that Hypertransport in mind please. Keep 939 pin packages in mind (the 3800+ is just the initial step. Sempron (and immediate successors) in mind...... AMD is VERY VERY deliberate in what they are doing right now. The steps are very logical and there is a deliberate migration of technology.... not a 'more on top of the old same stuff'.

I will make you a bet.... I bet we see 1024bit, quad core, 3.2 - 4.0 Ghz processors and busses on AMD before we see such animals on Intel. Deal??? LOL

BC.


PS: apologize for typos... had a tough day.

[BC]

Ok.. Here they are... on a web page.. (thumbnails included)

You can see the construction technique.....

Air flows in the bottom... over the HDs, and onto the cards.
There is a fan there for extra cooling as needed.

The lower rear fan catches the PCI and AGP heat.

The upper rear fan is dedicated to the CPU and you can see the
thermal sensors into the CPU heatsink fins.

On the AGP card, that white 'connector', is the other temp sensor.


The NewCastle, when finished, had an Antec TruePower550 with
exhaust fan as well to pull CPU heat if needed (another safety).

I am not good at mechanical, but the air flow is clean, the board is
not cluttered... Everything is pulled off to the sides of the chassis
(why it looks so bad) to keep the center open for best air flow.

Someday I will pull out ALL the 12v connectors and get a tie-wrap
gun from work and do it nice and neat.

Notice the lower left vent and upper vents which allow that 'extra' air
pressure I talked about to escape. Extra heat (in case wires are in
the way) gets out there.

(( I tried to embed the HTML but can't get it right.... ANYONE is welcome to fix it as needed))... I am *NOT* a web weenie...lol


http://www.tumoeng.com/amd/pc1and2.htm



BC.

[em99010pepe]

BC wrote this,

Ok, Nothing fancy.... All pics are thumbnail hyperlinks to higher quality/resolution images.

Here are the PC's.... The first is the bare NewCastle as it was being built, you can see the T-Take fans
and hard disk bay cooler mounted. The AGP is there, as are the thermal sensors in the heatsinks of the
CPU and AGP card.

(FYI: The NewCastle had to leave today.... Dad's P3 is in bad shape, so he has the NewCastle.)


NewCastle being built.

Now, Here is the older, but FULLY loaded Clawhammer that I chat with you all on.
Count the HD's, Cards, etc. Notice the front air inlet is taped as I have suggested.
and the design / cooling as I have described it. The one T-take is all black.


Side view....


Front View
(there is an 80GB HD behind that temp gauge,
and the reading at that moment is chassis ambient in F)

Last, but not least, THE INSIDE VIEW!!!! LOL


Yes, it needs 'dressing up', but it works as I want it to.
Airflow is NOT restricted. ALL PCI slots are full.
The Red AGP is that 100C heater 9800XT Radeon.
Those other cables are both Audigy and disk controller #2.
If you look carefully, you can see the temp sensors running around.



(Let me know via the boards if I should identify all the parts)

Source

[BC]

Gents,
I made a mistake and wish to ammend my comments.

I was correct that the A64's and above all have the quad
capability....

What I failed to mention was that the FX-53 has it available
and it runs by default. It runs at 1600 .

Just wanted to get that straightened out before I forgot.

My apologies for the oversight, I have been a bit busy and
as such, I was thinking one product and writing of another...
knowing I am not allowed to put names, details, or dates together.



BC