VIA K8T800 PRO: PCI/AGP Lock and 1000 HyperTransport for Athlon 64
by Wesley Fink on May 6, 2004 5:00 AM EST- Posted in
- CPUs
Overclocking and Stress Testing: VIA K8T800 PRO Reference Board
FSB Overclocking Results
Since the PRO chipset claims the first working AGP lock for a VIA chipset on an Athlon 64 board, we were anxious to see if it indeed worked. At first, we thought that this was all smoke and mirrors, since we could not achieve any decent overclocks on the VIA board. Finally, after lowering the multiplier to 8, we were able to hit the 255 limit of the board.| Front Side Bus Overclocking Testbed | |
| Default Voltage | |
| Processor: | FX51 (2.2GHz) |
| CPU Voltage: | 1.525V (default) |
| Cooling: | OCZ Eliminator 2 |
| Power Supply: | Antec TruePower 430W |
| Maximum OC: | 2310 231x10 |
| Maximum FSB: | 255 x 8 |
With no voltage adjustments at all for the processor or memory, overclocking was a challenge. We were also hampered by the fact that this early FX51 is not a particularly good overclocker. Our goal with the AGP lock was to find an overclock above the 216 to 220, which seems to be the working limit of our ATI 9800 PRO.
We reached 231x10, and then tried to lower the multiplier and max out the board.
We were able to reach the highest FSB available on the K8T800 PRO at a multiplier of 8. Both these overclocks certainly demonstrate that the AGP lock is working. Above about 222, we needed to lower the HT frequency to 800. Above 250, we had to drop further to 600 HT. The fact that we had to lower multipliers and HyperTransport speed at such relatively low levels suggests that VIA needs to get more voltage to the chipset. Board manufacturers that address the chipset voltage issue will likely do much better in overclocking the K8T800 PRO.
The OCZ PC3500 Registered ECC memory handled frequencies to 250 just fine, but asynchronous timings were required at 255 FSB. Timings could still be set at the excellent 2-2-3-6 that also worked well at DDR400.
The good news is that from the tests we ran, VIA finally has a working AGP/PCI lock on their Athlon 64 boards. With both VIA and nVidia offering AGP locks and 1000 HyperTransport, we will now have a lot more options for overclocking Athlon 64 chips.
Memory Stress Test Results:
The memory stress test is very basic, as it simply tests the ability of the K8T800 PRO Reference Board to operate at its officially supported memory frequency (400MHz DDR) at the lowest supported memory timings that our OCZ PC3500 Registered ECC will support. Memory stress testing was first conducted by running RAM at 400MHz with 2 DIMM slots filled, which is one Dual-Channel bank.| Stable DDR400 Timings - 2 DIMMs (2/4 DIMMs populated - DC mode) |
|
| Clock Speed: | 200MHz |
| Timing Mode: | N/A |
| CAS Latency: | 2.0 |
| Bank Interleave: | 4-bank |
| RAS to CAS Delay: | 3T |
| RAS Precharge: | 6T |
| Precharge Delay: | 2T |
| Command Rate: | Auto |
The K8T800 PRO was completely stable with the most aggressive 2-2-3-6 timings that our Registered ECC memory could support. This performance was achieved without any voltage options at all on the board.
Filling all four available memory slots in a 2-bank dual-channel configuration is more strenuous on the memory subsystem than testing 2 DIMMs as a single DC bank. We were very pleased to find 4 DS DIMMs (2GB) of memory worked fine at the same timings we used for 2 DIMMs . Performance with all 4 DIMMs installed was very stable.
| Stable DDR400 Timings - 4 DIMMs (4/4 DIMMs populated - 2 DC mode) |
|
| Clock Speed: | 200MHz |
| Timing Mode: | N/A |
| CAS Latency: | 2.0 |
| Bank Interleave: | 4-bank |
| RAS to CAS Delay: | 3T |
| RAS Precharge: | 6T |
| Precharge Delay: | 2T |
| Command Rate: | Auto |
Since we had the board for only a limited time, we were not able to run the complete stress testing that is normal for AnandTech memory testing. However, there were no indications of memory stability issues in any of our benchmarking.
Socket 939 will bring unbuffered Dual-Channel memory to Athlon 64 processors. With that, there will be many more options for high-speed memory on the A64. The OCZ 3500 Registered memory that we used for testing performed with great stability on the K8T800 PRO - even at the most aggressive timings that we could set.
Facebook
Twitter
RSS
15 Comments
View All Comments
Odeen - Saturday, May 8, 2004 - link
#4: The answer is pretty much a "yes"Reason: Firewire, in this design, sits on the PCI bus. Gigabit networking sits on the PCI bus. Any soundcard / chip better than chipset-provided onboard sound sits on the PCI bus. Any additional hard drive controllers sit on the PCI bus.
On non-server motherboards, the PCI bus runs at 33mhz and is 32 bits wide, for 133 megabytes/sec of bandwidth
Firewire is 400mbit/sec per port. The new design is 800mbit/sec. This means you will use a maximum of 50 megabytes/sec for a firewire/400 port (granted, the practical peak is less than 50mb/sec, but you'll never eke out 133mb/sec out of the PCI bus either). Add to that the 250mb/sec full-duplex Gigabit Ethernet can generate, and 150mb/sec for ever SATA port courtesy of those Silicon Image chips, and you can see that you're starved for bandwidth.
On the other hand, there's a tidy 533 megabytes/sec of bandwidth between the northbridge and the southbridge.. that means everything that runs straight off the southbridge can report back to the northbridge and, consequently, to the CPU at 533mb/sec. Subtract 133mb/sec for the PCI, and you still have 400 megabytes for LAN and hard drives and firewire, if it's implemented in the southbridge.
Unfortunately, the only true "onboard" firewire comes courtesy of the MCP-T southbridge on nForce2 boards. Everyone else (All Via, all SiS, all Intel chipsets, and all chipsets for Athlon64) has to use a PCI firewire chip.
Pumpkinierre - Friday, May 7, 2004 - link
Welsey #11, I would liked to have seen an 8or9 x 233and234. I still smell a rat. I've run some mobos at 41MHz PCI speed in the past while others (same brand and chipset but different review) would not function. The fact that other posts claim that ABIT have left out the AGP/PCI lock feature of this chipset adds to my suspicion.Klaasman - Friday, May 7, 2004 - link
Why oh why ABIT, would they leave this out?? Damn it!blup - Friday, May 7, 2004 - link
This board breaks my heart. I was hoping for a "BH7 for Athlon 64" i.e. cheap, fast, very overclockable. It is not to be.ABIT has NOT implemented the PCI/AGP lock on this board - maybe there will be a more expensive MAX model that does. Checkout one man's review at http://www.nickgoodall.org/kv8pro/review1.htm
Wesley Fink - Friday, May 7, 2004 - link
#10 -I did a quick check with PC Geiger and the bus was locked on the Reference Board. Since I had the board for only 2 days for testing, there was no time for much more. You will also see I reached 231 overclock on the FX51 at 10 ratio, in addition to the 255x8.
Checking my review notes, I also tested the 10 multiplier at 233 and 234 to determine if there was a ratio drop. I could boot at 233 but the system wasn't completely stable; the CPU simply would not do 234 at standard multiplier. This also suggests a working lock.
Please also keep in mind that VIA has a working pci/agp lock on the PT880 chipset for the P4, so this is not their first PCI/AGP lock. It is just their first lock on an Athlon 64 chipset.
Pumpkinierre - Thursday, May 6, 2004 - link
You got to 8x252 on the Aopen AK86-L (VIA chipset/Skt754) which didnt have PCI/AGP lock but ratio controlled. Given its an a64/FX chipset (and on top VIA) where AGP/PCI lock has been a matter of controversy, would'nt it have been prudent to have used your PCIgeiger to check the PCI frequencies at the different FSB settings-so that we wont be fooled once again!Jeff7181 - Thursday, May 6, 2004 - link
This is definately a good thing... finally overclockers have a choice between chipsets.Warder45 - Thursday, May 6, 2004 - link
I thought the Pro was going to have a HT speed of 1200?Wesley Fink - Thursday, May 6, 2004 - link
#6 -PCI/AGP lock is DEFINITELY working on the Reference Board, but I can not yet address whether Abit has implemented the lock or just dropped the PRO chipset in the older design with no changes. I did see the one review in the UK reporting this issue before the VIA review was finished.
We do have the Abit board coming to us for testing and I will report what I find.
Klaasman - Thursday, May 6, 2004 - link
Abit has a board out now but reports are that it does NOT have the AGP/PCI lock working.