nd22 wrote on Yesterday, 17:17:

marxveix wrote on Yesterday, 14:11:

At the moment i have these Socket 754 cpu-s, maybe i have more: AMD Mobile K8 Athlon XP-M 3000+ – AHN3000BIX3AY, Socket 754 […]
Show full quote

At the moment i have these Socket 754 cpu-s, maybe i have more:
AMD Mobile K8 Athlon XP-M 3000+ – AHN3000BIX3AY, Socket 754

AMD Athlon 64 2800+ – ADA2800AEP4AX NewCastle DH-CG Socket 754
AMD Athlon 64 3000+ – ADA3000AEP4AX NewCastle DH-CG Socket 754

AMD Sempron 64 2500+ – SDA2500AIO3BX Socket754
AMD Sempron 64 2600+ – SDA2600AIO2BX Socket754
AMD Sempron 64 2600+ – SDA2600AIO2BO Socket754
AMD Sempron 64 2800+ – SDA2800AIO3BX Socket754
AMD Sempron 64 2800+ – SDA2800AIO3BA Socket754
AMD Sempron 64 3000+ – SDA3000AIO2BX, SDA3000AIO2BA Socket754
AMD Sempron 64 3100+ – SDA3100AIO3BX Socket754

AMD Mobile Sempron 2800+ – SMS2800BQX3LF L2 256kb Roma Socket 754
AMD Mobile Sempron 2800+ – SMS2800BOX3LB L2 256kb Sonora Socket 754
AMD Mobile Sempron 3000+ – SMS3000BQX2LF Roma DH-E6 Socket 754
AMD Mobile Sempron 3100+ – SMS3100BQX3LFx2 Socet 754
AMD Mobile Sempron 3300+ – SMS3300BQX2LFx4 E6 Roma 128kb, Socet 754
AMD Mobile Sempron 3300+ – SMN3300BKX2BX Albany E6, 62W, Socket 754

AMD Turion 64 Mobile ML-28 – TMDML28BKX4LD Socket 754
AMD Turion 64 Mobile MT-32 – TMSMT32BQX4LD Socket 754
AMD Turion 64 Mobile ML-32 – TMDML32BKX4LD Socket 754
AMD Turion 64 Mobile ML-34 – TMDML34BKX5LD Socket 754
AMD Turion 64 Mobile ML-37 – TMDML37BKX5LD Socket 754

AMD Mobile Athlon 64 3000+ 1800Mhz (AMA3400BEX5AR ClawHammer CG) Socket 754
AMD Mobile Athlon 64 3200+ 2000MHz (AMN3200BKX5BU Newark E5) Socket754
AMD Mobile Athlon 64 3400+ 2200Mhz (AMA3400BEX5AR ClawHammer CG) Socket 754

SMS3300BQX2LF i have 4pcs maybe its better to take out ML-37 and use Sempron 3300+?
It has lower graphics card anyway, still 3400+ is missing from Mobile Sempron + 25w cpu!
Does the Sempron 3300+ and Radeon 9600xt work better together, less bottlenecking???
Of course this Mobile Sempron 3300+ is slower and it has only 128kb cache vs 1mb@turion
3400+ has 256kb and that would be the maximum Mobile Semrpon for Socket754 system

2.4GHz i dont have yet, max is 2.2GHz @ my 754cpu collection.

You got quite a nice collection there! You will see some high end CPU's, all desktop versions but exotic ones.

Thank you! 2.4 and 2.6GHz still missing, but yes, its good option of 754 CPU-s.
https://www.cpumedics.com/amd-amn4000bkx5bu-2 … -cpu-processor/

Now the systems details - all the parts meet the following criteria: easy to find! Be aware that I did not say cheap!
1. CPU: the most powerful processor for socket 754 is the Athlon 64 3700 based on the first revision: Clawhammer! However that CPU is very rare and expensive. Very few people have it and very few can afford to pay for one!
I had one but it was a disaster!

nd22 wrote on Yesterday, 17:32:

Now the systems details - all the parts meet the following criteria: easy to find! Be aware that I did not say cheap!
1. CPU: the most powerful processor for socket 754 is the Athlon 64 3700 based on the first revision: Clawhammer! However that CPU is very rare and expensive. Very few people have it and very few can afford to pay for one!
I had one but it was a disaster!

This is fastest Socket754 CPU, its named Mobile, but it takes 62W: AMD Mobile Athlon 64 4000+ - AMN4000BKX5BU
https://www.cpu-world.com/CPUs/K8/AMD-Mobile% … 4000BKX5BU.html

That would not be on the official sanctioned AMD list because it is a mobile chip! Also good luck finding one!

The story of my 3700: some 10 years ago I bought one off eBay. After I installed it in one of the boards the system started, I heard the standard beep, saw the screen with the CPU correctly recognized but it locked up immediately. I could not understand why so I changed the board from one Abit to another. The same thing happened; finally I put it in my one and only KV8 – MAX3 and the system froze on start up again with a strong smell of burned electronics and afterwards no matter what board I used it would not start at all!
Not only that but my KV8 – MAX3 died and I could not resurrect that board. I trashed that 3700 and never bought another one again! So I did not use that CPU simply because I do not have it.

However I choose the next in line: Athlon 64 3400.
There are several generations of Athlon 64: the very first one is the Clawhammer; armed with 1024kb of L2 cache it is one that proved that Pentium 4 can be beaten. It has only 2200 MHz and it is built on the old 130 nm process so it is going to be interesting to see how it is standing up performance wise.
The second generation is Newcastle: with half the cache – 512kb – it has a higher frequency to make up for the lost cache – 2400 MHz. Also built on the 130 nm process we shall se what matters most: cache or frequency?
The third generation is Winchester but it is not included because it is present only on socket 939.
The fourth generation is Venice. It has SSE3 instruction set, it is manufactured on the 90 nm process and has much lower TDP but otherwise it is identical to the Newcastle. Unlike Clawhammer and Newcastle which are plentiful on eBay it is very rare. After I put up the results we will discuss socket 754 Venice.
Because I tested both the Clawhammer and the Newcastle we are going to see a battle between cache and frequency!

nd22 wrote on Yesterday, 17:20:

Today I do now want to repeat the same experience! I want the dream experience, when I play Farcry I want to play with the settings turned all the way up including AA and AF. That’s why the target period for socket 754 is 2000 – 2004. Going beyond 2004 is beyond the capabilities of the CPU, the graphics card or both; of course all will become clear when you will see the screenshots.

Because this is a comparative - that means identical graphics cards on all the systems - you will not see anything later than geforce 7000 series and radeon x1k series.
And no, a radeon 3850 will not help you play 2005 games at max details! In fact one of the games in the testing suite has been tested with AA disabled!

If you would like to use target period 2000-2004 for the purpose of the test, to be able to compare them with the same GPU that's fair. But then the statement Going beyond 2004 is beyond the capabilities of the CPU, the graphics card or both is misleading readers. I never mentioned radeon 3850 as it isn't sufficiently powerful. I do have Clawhammer 3400+ and tested it with various games so I can stand by my statement. Most s754 motherboards are AGP, but there is also a plenty of micro-ATX PCIe ones.

nd22 wrote on Yesterday, 18:00:

The fourth generation is Venice. It has SSE3 instruction set, it is manufactured on the 90 nm process and has much lower TDP but otherwise it is identical to the Newcastle. Unlike Clawhammer and Newcastle which are plentiful on eBay it is very rare.

See https://www.ebay.de/itm/236030895621 . 2.2Ghz Venice sold for 14 EUR from Austria. It's a good price. Seller has 2 items.

nd22 wrote on Yesterday, 18:00:

The second generation is Newcastle: with half the cache – 512kb – it has a higher frequency to make up for the lost cache – 2400 MHz. Also built on the 130 nm process we shall se what matters most: cache or frequency?

Because I tested both the Clawhammer and the Newcastle we are going to see a battle between cache and frequency!

Regarding your statement: cache or frequency

These are the CPU's of interest in this test review:
Athlon 64 3400+ (2,2 GHz, Single Channel, 1024 KB) - Sockel 754
Athlon 64 3200+ (2,0 GHz, Single Channel, 1024 KB) - Sockel 754
Athlon 64 3000+ (2,0 GHz, Single Channel, 512 KB) - Sockel 754

Games performance rating.
3DMark2001SE
3DMark03
Aquamark 3
Comanche 4
Gunmetal
Quake 3 Arena
Serious Sam 2
Splinter Cell
Unreal Tournament 2003
Warcraft 3
X-2 The Threat

With the same clockspeed (@2Ghz), the 3200+ is 3% faster than the 3000+ (difference in cache).
With 200Mhz more clockspeed (same cache), the 3400+ is 4% faster than the 3200+ (difference in clockspeed).

melbar wrote on Today, 07:55:

With the same clockspeed (@2Ghz), the 3200+ is 3% faster than the 3000+ (difference in cache).
With 200Mhz more clockspeed (same cache), the 3400+ is 4% faster than the 3200+ (difference in clockspeed).

AMD calculated roughly the same thing.
200 MHz = half the cache.
This is also evident from the markings.
But at the same frequency, Venice is often faster than Clowhammer with the same cache size.

Venice should also OC better. 2.2Ghz should run at 2.4Ghz without any tweaks. Someone needs to buy it and test it.

AlexZ wrote on Today, 08:36:

Venice should also OC better. 2.2Ghz should run at 2.4Ghz without any tweaks. Someone needs to buy it and test it.

Well, i recently got a box of s939 single core CPUs including some Venice ones, since i wanted to finally be able to test the motherboards i have without disassembling existing systems. I wonder how close s939 is to s754 if only one memory channel is used...

Archer57 wrote on Today, 11:06:

AlexZ wrote on Today, 08:36:

Venice should also OC better. 2.2Ghz should run at 2.4Ghz without any tweaks. Someone needs to buy it and test it.

Well, i recently got a box of s939 single core CPUs including some Venice ones, since i wanted to finally be able to test the motherboards i have without disassembling existing systems. I wonder how close s939 is to s754 if only one memory channel is used...

That would be nice with same mb chipsets at s754 and s939, single channel and venice vs venice.

something like that
Sorry, I don't have any Clawhammer/Newcastle CPU.

This is my Athlon 64 3400+ (Clawhammer). Note very high L2 cache latency and low memory latency.

First, I aligned the memory timings 3-3-3-8 CR2 and CPU frequency - 2GHz.
My Turion usual work at 275MHz HTT base clock, has a 2.2GHz core, 1.1GHz Hypertransport, and 220MHz memory.
With CR1, write and copy become equal to read – about 3000.
Second, that's not it.
Look at the gigaflops on the Lynx stability test/benchmark.
https://github.com/sanekgusev/LinX-old
512K = 2.68 GFlops
1024K = 2.83 GFlops.
5% difference at the same frequency.
We've been discussing L2 cache for two days now.

AlexZ wrote on Yesterday, 18:33:

If you would like to use target period 2000-2004 for the purpose of the test, to be able to compare them with the same GPU that's fair. But then the statement Going beyond 2004 is beyond the capabilities of the CPU, the graphics card or both is misleading readers. I never mentioned radeon 3850 as it isn't sufficiently powerful. I do have Clawhammer 3400+ and tested it with various games so I can stand by my statement. Most s754 motherboards are AGP, but there is also a plenty of micro-ATX PCIe ones.

See https://www.ebay.de/itm/236030895621 . 2.2Ghz Venice sold for 14 EUR from Austria. It's a good price. Seller has 2 items.

Sorry if you misunderstood but that is exactly the goal: testing all the systems with the same GPU in part 3! So the most powerful card used must be both formats: AGP and PCI-express!

I got that CPU and - spoiler - in the "target"period which is 2000 - 2004 there is absolutely no difference between Venice and Newcastle.

melbar wrote on Today, 07:55:

Regarding your statement: cache or frequency […]
Show full quote

nd22 wrote on Yesterday, 18:00:

The second generation is Newcastle: with half the cache – 512kb – it has a higher frequency to make up for the lost cache – 2400 MHz. Also built on the 130 nm process we shall se what matters most: cache or frequency?

Because I tested both the Clawhammer and the Newcastle we are going to see a battle between cache and frequency!

Regarding your statement: cache or frequency

These are the CPU's of interest in this test review:
Athlon 64 3400+ (2,2 GHz, Single Channel, 1024 KB) - Sockel 754
Athlon 64 3200+ (2,0 GHz, Single Channel, 1024 KB) - Sockel 754
Athlon 64 3000+ (2,0 GHz, Single Channel, 512 KB) - Sockel 754

Games performance rating.
3DMark2001SE
3DMark03
Aquamark 3
Comanche 4
Gunmetal
Quake 3 Arena
Serious Sam 2
Splinter Cell
Unreal Tournament 2003
Warcraft 3
X-2 The Threat

The attachment Athlon64_performance-rating_games.jpeg is no longer available

With the same clockspeed (@2Ghz), the 3200+ is 3% faster than the 3000+ (difference in cache).
With 200Mhz more clockspeed (same cache), the 3400+ is 4% faster than the 3200+ (difference in clockspeed).

Can you give a link please? I want to see what parts they used and testing methodology!
Thank you.

marxveix wrote on Today, 11:18:

Archer57 wrote on Today, 11:06:

AlexZ wrote on Today, 08:36:

Venice should also OC better. 2.2Ghz should run at 2.4Ghz without any tweaks. Someone needs to buy it and test it.

Well, i recently got a box of s939 single core CPUs including some Venice ones, since i wanted to finally be able to test the motherboards i have without disassembling existing systems. I wonder how close s939 is to s754 if only one memory channel is used...

That would be nice with same mb chipsets at s754 and s939, single channel and venice vs venice.

Hold that thought! For the distant future 😀 !

2. RAM: all the boards that you will see have only 2 DIMM slots so can have a maximum of 2 GB of memory.
The question that will spring immediately to everyone’s mind is: are 2 GB enough?
All the tests have been done with the page file deactivated so everything is loaded into the system RAM. The highest amount of memory used is in games: Farcry and Doom 3 are the biggest hogs and take around 1.3 – 1.4 GB of memory. So 2 sticks of 1 GB are more than enough for any game up to 2004!
Despite using various kits from different manufacturers all timings are the same: 3-3-3-8 with command rate 1T. Because one of the boards refused to run with anything but AUTO timings I used your average memory modules and not some high end DDR ones – I should mention that aggressive timings make no difference whatsoever at the settings I used. And no, I am not going to test at 640*480 or 800*600!

Here its visible 128kb cache vs 256kb cache with Sempron 754 cpus / sometimes bigger cache is better than 200Mhz higher clocks.
https://www.youtube.com/watch?v=gKMWlezaUeI