VOGONS


Battle of the platforms: socket 939!

Topic actions

Reply 140 of 284, by AlexZ

User metadata
Rank Oldbie
Rank
Oldbie

I have read that nforce4 standard supports max HT 800, while ultra supports HT 1000. This probably means just the 5x multiplier. On my s754 Gigabyte GA-K8NE the system didn't boot with Clawhammer and 5x HT multiplier. Maybe you could verify that with Venice on both s754 and s939.

Pentium III 900E,ECS P6BXT-A+,384MB,GeForce FX 5600, Voodoo 2,Yamaha SM718
Turion 64 MT-40@2.4Ghz,Gigabyte GA-K8NE,2GB,GeForce GTX 275,Audigy 2ZS
Phenom II X4 965,Gigabyte GA-MA770-UD3,8GB,GeForce GTX 780
Vishera FX-8370,Asus 990FX,32GB,GeForce GTX 980 Ti

Reply 141 of 284, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

Just FYI : HT speed (or Hyper Transport frequency), does absolutely nothing to actual performance on K8 CPUs (unless you have multi CPU server board), you can have it at 600/800/1000 and it shouldn't change anything for either CPU or memory (RAM).
In single CPU systems, it can have impact on AGP/PCIe bus transfer speeds to/from CPU - but that usually doesn't happen in any meaningful way (or can't be separated from other variables that also affect performance). Sticking to ~600 is what I would recommend if you want to OC nForce3 for example.

Reply 142 of 284, by AlexZ

User metadata
Rank Oldbie
Rank
Oldbie

HT 1.0 at 1000Mhz is supposed to have 4GB/s bandwidth in each direction, same as PCIe 1.0 x16 (250MB/s per lane). HT running at 800Mhz cannot operate PCIe at full speed. Probably not relevant for practical purposes as that data has to come from somewhere and the only "device" fast enough is RAM, which is typically not big enough in s939. It matters mostly from factual perspective, whether the claim of HT 800 limit in nforce4 standard is true. A more severe limit is support of SATA-1 only.

Pentium III 900E,ECS P6BXT-A+,384MB,GeForce FX 5600, Voodoo 2,Yamaha SM718
Turion 64 MT-40@2.4Ghz,Gigabyte GA-K8NE,2GB,GeForce GTX 275,Audigy 2ZS
Phenom II X4 965,Gigabyte GA-MA770-UD3,8GB,GeForce GTX 780
Vishera FX-8370,Asus 990FX,32GB,GeForce GTX 980 Ti

Reply 143 of 284, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

Extra OC (HTT base clock) headroom you can get from less HT link is usually more impactful on performance than a bit extra bandwidth on PCIe bus.

Reply 144 of 284, by FullYes

User metadata
Rank Member
Rank
Member

I owned an abit UL8 board back in about 2006 paired with an x1950GT and a San Diego 4000+. I ended up changing it due to death of AGP to an nforce4-based A8N-SLI Deluxe with a HD3850 and an opteron 185, but the UL8 was a lot better at overclocking, particularly the FSB. Seeing your review makes me regret getting rid of it, i probably gave it away!

System 1: K6-3 400, MS-5158 (430TX), 256MB SDRAM, Voodoo3 2000 PCI, AWE64 (Dos/95-early 98)
System 2: Athlon XP 2000+, SL-75KAV (KT133A), 512MB SDRAM, FX5900XT, Audigy 2 ZS (late 98)
System 3: i7-2600k, P8P67-M, 2x4GB DDR3-1866, GTX 460 (XP)

Reply 145 of 284, by nd22

User metadata
Rank l33t
Rank
l33t
AlexZ wrote on 2026-04-18, 20:09:

I have read that nforce4 standard supports max HT 800, while ultra supports HT 1000. This probably means just the 5x multiplier. On my s754 Gigabyte GA-K8NE the system didn't boot with Clawhammer and 5x HT multiplier. Maybe you could verify that with Venice on both s754 and s939.

Sorry for the delay but I got hold up at work with a new project!
Standard nforce4 supports 1000 MHz HTT out of the box; there are limitations - only SATA 150, no hardware firewall.

Reply 146 of 284, by nd22

User metadata
Rank l33t
Rank
l33t
agent_x007 wrote on 2026-04-19, 17:46:

Just FYI : HT speed (or Hyper Transport frequency), does absolutely nothing to actual performance on K8 CPUs (unless you have multi CPU server board), you can have it at 600/800/1000 and it shouldn't change anything for either CPU or memory (RAM).
In single CPU systems, it can have impact on AGP/PCIe bus transfer speeds to/from CPU - but that usually doesn't happen in any meaningful way (or can't be separated from other variables that also affect performance). Sticking to ~600 is what I would recommend if you want to OC nForce3 for example.

That is correct; I tested Abit NF8 with HTT set at 600 MHz to simulate a nforce3 150 and the difference between 800 HTT and 600 HHT is negligible! However poor performance relative to its AGP counterparts remains!

Reply 147 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

Abit KN8 is a standard ATX board with plenty of room around the CPU socket for large coolers. Everything I said about Abit NV8 applies to KN8 with the exception of the SATA connectors; yopu can read it here: Re: Battle of the platforms: socket 754!

Reply 148 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

You can see the similarities between the 2 boards in this picture. It is mandatory to connect the molex on both boards regardless of the graphic card used! You can see clearly the SATA connectors on KN8 are above the pci-express 16X slot so you can use double slot cards without any issues!

Reply 149 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

I used a SATA SSD bought before the price hikes with a Raptor as slave and a SATA DVD-RW for a full SATA system so no more ribbons!

Reply 150 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

The system after the tests (please note the "exceptionally good"cable management) :

Reply 151 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

HTT running at full speed on standard nforce4:

Reply 152 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

To keep things comparable with the AGP tests I used the same Radeon X1950 PRO 512 mb videocard; on KN8 I did install a radeon 4890 1gb without any issues, however on NV8 this is not possible because this dual slot card will block off all SATA connectors - a common design flaw on many socket 754 boards with nforce4-4x! AlexZ has one of the very few socket 754 boards where SATA connectors are placed above the pci-express connector: Gigabyte K8NE!

Reply 153 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

In the screenshot above you can also see that I used the latest driver for the radeon: 10.2!
Other drivers used: nforce4 platform driver 15.23 (latest); realtek AC97 driver version 4.06.
Windows is XP SP3 with net framework 2.0 installed which is required by the catalyst control center; msi afterburner 4.30; fraps 3.4.7; Everest 5.50 running in the background at all times.

Reply 154 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

I come back to resume testing after 1 month of non-stop work with a new project, so let's continue!
In order to have the 2 systems as close as possible in specs I decided to use for part 2 another pair of processors, one for each socket: on Abit NV8 I tested with Athlon 64 3400 Newcastle and on Abit KN8 I used Athlon 64 3800 Newcastle.
That means that we got 2 processors with the same revision, same frequency and same amount of Level 2 cache: on socket 754 we got the Athlon with part number ADA3400AEP4AX working at 2400 MHz with 512 KB of cache and on socket 939 we got Athlon 64 part number ADA3800DEP4AW also at 2400 MHz with the same 512 KB of cache.

Reply 155 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

So we are going to see if the 11.76% difference in the PR ratings of the 2 processors is justified; is dual channel is "strong" enough to provide this much more performance or, as in the case of Venice 3500 socket 939 versus Venice 3200 socket 754 there is only an extra 200 PR points courtesy of dual channel?
Also we are going to see how much difference is between the Athlon 64 3800 Newcastle and 3500 Venice, in preparation for part 3 when we are going to test all 7 Athlon 64 revisions in a head-on battle.

Reply 156 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

I am not going to present the nforce4-4x based system again as all details about the socket 754 system can be read here:
- details of the parts and BIOS settings: Re: Battle of the platforms: socket 754!
- system details with the Clawhammer 3400: Re: Battle of the platforms: socket 754!
- system details with the Newcastle 3400: Re: Battle of the platforms: socket 754!

Reply 157 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

Socket 939 system is made up of the following parts:

Reply 158 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

You can see that Newcastle clocks down to 1.000 MHz if cool and quiet is enabled and the driver installed. During testing it was disabled at all times.

Reply 159 of 284, by nd22

User metadata
Rank l33t
Rank
l33t

The radeon x1950 pro has 512 mb of RAM and default clocks, the board has the same nforce4 chipset as Abit NV8 so no XP reinstall was necessary: