nForce3-250 - Part 1: Taking Athlon 64 to the Next Level
by Wesley Fink on March 23, 2004 11:55 AM EST- Posted in
- CPUs
nForce3-250Gb: WORKING AGP/PCI Lock
It should not be a surprise that the first thing we confirmed is a working PCI/AGP lock. After discovering that none of the previous Athlon 64 chipsets had a working AGP lock, we went immediately to a test to verify that the AGP lock was indeed working. We used PC Geiger that was used for measuring PCI in PCI Speed and Overclocking: A Closer Look at A64 and P4 Chipsets. With FSB set at 249 and AGP frequency set to 67, we measured a PCI speed of 33.3MHz.
This is the expected results for a working AGP lock. We certainly can confirm that the AGP lock is working on the nForce3-250GB Reference Board, and we'll take another look when shipping nF3-250 boards start showing up in a couple of weeks.
FSB Overclocking Results
Reference Boards are not really designed for overclocking, and there are normally not any voltage adjustments available as we would see on production motherboards. However, with PCI/AGP lock available, we were anxious to see if the nForce3-250Gb did indeed overclock better. This was also an opportunity to verify a working PCI/AGP lock by a different method.| Front Side Bus Overclocking Testbed | |
| Default Voltage | |
| Processor: | Athlon 64 3400+ 2.2GHz |
| CPU Voltage: | 1.5V (default) |
| Cooling: | AMD Stock Athlon 64 Heatsink/Fan |
| Power Supply: | Powmax 350W |
| Maximum OC: | 2442MHz (+11%)@222x12 2375Mhz@250FSBx9.5 (+25% FSB) |
The above overclocking setup at default voltage allowed us to reach a stable Frequency of 250 at 800 HyperTransport with AGP/PCI fixed at 33/66. The limit of this CPU at default voltage is apparently somewhere around 2450MHz, since we could not reach 250FSB with a 10 multiplier. That would have represented 2.5GHz, had we been successful.
Important here is the fact that we reached the absolute highest FSB setting of the Reference Board, which is 250. This is further evidence that the AGP lock is indeed working on nforce3-250.
We asked nVidia about the issues with the nForce3-150 AGP/PCI lock, which are apparently fixed in nF3-250. nVidia assured us that the AGP lock was indeed working in the 150 Reference Board, but that the BIOS implementations by manufacturers with the 150 chipset were not correct. Therefore, the shipping boards for nF3-150 did not have a consistently working AGP lock. nVidia also assured us that they were working more closely with manufacturers on the nF3-250 launch to make sure manufacturers were delivering nF3-250 boards with working AGP lock.
We tested nVidia's claim and found the Reference nF3-150 board did have a working AGP lock. The AGP lock is also definitely working on the 250 Reference Board, and we sincerely hope we will find the same working lock on shipping nF3-250 motherboards. We will be looking closely at production nF3-250 boards to verify a working AGP lock.
Reference Boards are rarely a good indication of the true overclocking abilities of a chipset because they are designed to qualify and demonstrate a chipset at design parameters. The features are definitely here for the best overclocking in current Athlon 64 chipsets. It is also worth mentioning that the Athlon 64 Multiplier Utility, available as shareware from CPU-Z, worked fine on the nForce3-250. While it was designed for the 150 chipset, nVidia saw no reason why it should not work properly and we did confirm that the multiplier adjust utility works.
Facebook
Twitter
RSS
71 Comments
View All Comments
Visual - Friday, March 26, 2004 - link
This chipset looks promising, I like it. And a great article about it :)I'm a bit curious about the raid - do you guys think it may be possible to implement a hot-swappable raid array with integrated raid controllers anytime soon?
Maybe you can make an article testing the performance boost from using a 4-drive raid 0 array with this baby?
Another thing that interests me - are there any mobos with IGP for Athlon64? I know it won't be a performer, I'm just curious if it even exists. Also is anything being heared about some new DX9 IGP anytime soon(hopefully with this chipset)? It'd also be cool if having an AGP card doesn't disable the IGP, like the ati-intel chipsets... Well I guess I'm dreaming now, but I'd like to see your comments or any info you have on nVidia's IGP plans. I guess you AT folks could ask nVidia about this :)
Thanks,
Visual
Reflex - Thursday, March 25, 2004 - link
#59: Try measuring your bandwidth with a 4 drive RAID 0 array using fast drives on that setup and then put the same array in an Intel or AMD chipset system. nVidia's PCI implementation is not very good at all.MichaelD - Thursday, March 25, 2004 - link
[q] Actually, to date nVidia has had a *very* troublesome PCI implementation, anyone with a PCI RAID controller and a 4 disk RAID 0 array can tell you that. It is so bad, in fact, that prototype NF3-150 boards for Opteron used AMD PCI chips just to avoid using the nForce3 integrated PCI bus. I am not certain if these boards ever reached production status however.[/q]Uh, no. Not in my experience. On my 8RDA+, I've used:
Highpoint ATA133 Contoller Card
3Ware7000-2 Two-channel IDE RAID card
LSI Megaraid 1600 SCSI RAID card
I've had zero problems. Wha'chu talkin' bout, Willis?
Reflex - Thursday, March 25, 2004 - link
#55: I did not say DDR2 was needed right now, its not and AMD is making the right decision. I was just pointing out that the latency penalty should not be a real issue since it moves more data. But time will tell.#54: I have not checked out the Catalina yet, however if it does not have a coax output, it will not find a home in my setup. SPDIF is a consumer level technology, championed by Sony, but it is not as high quality as coax simply due to the fact that the signal must be converted twice(to and from optical) which is never a good thing. Furthermore, the cables are frail and expensive. Professional level equipment never has SPDIF, it uses coax exclusively.
Wesley: Glad they are dropping SoundStorm. Waste of time and effort in my opinion.
BikeDude - Thursday, March 25, 2004 - link
Thanks Wesley; a single chip implementation makes sense. Now show us the benchmarks! :)Wesley Fink - Thursday, March 25, 2004 - link
#54 and others regarding Sound Storm -1 - nVidia is committed to the one-chip chipset for Athlon 64. They are firmly convinced that the one-chip eliminates the potential bottlenecks of a north-south bridge communications bus. Even with the the memory controller on the chip there is only so much real estate practically available on a single-chip chipset.
2 - Customer surveys by nVidia found that most buyers did not use Sound Storm, and that Sound Storm did not enter heavily into the decision to buy nForce. So the decision was made to choose the on-chip LAN, firewall, and much-expanded RAID capabilities which benefit greatly from being moved off the bus.
3 - There are new sound solutions in the works for nVidia. You may see them in a future chipset or on a sound card. Final decisions have not been made.
Pumpkinierre - Thursday, March 25, 2004 - link
#53, I'll believe it when I see the tests. It sounds like RAMBUS- that was supposed to be better at latency but turned out the opposite at over twice the cost at the time. Read the last paragraph of Wesley's post(#50)- he's closer to the industry and there are others expressing similar concerns. All these are things that Intel with its resources should iron out and AMD come in when its sorted, If AMD get to a third of the market and in the black then it can show leadership in these areas. Meanwhile stick to what they are best at cpus.BikeDude - Thursday, March 25, 2004 - link
#48: Turtle Beach Catalina which I suspect is a newer card (it's more expensive :) ) than SC, seem to tout optical SPDIF output as a feature (doesn't mention coax at all) and it's merely pass-through SPDIF at that (no hardware Dolby encoding -- thus I'll end up with the additional three audio cables again). Are you sure you have all your facts straight?If you're a professional musician -- I agree, the SS isn't for you, but I thought nForce was primarily a chipset targetted at gamers?
Reflex - Thursday, March 25, 2004 - link
#52: Latency ends up about the same due to the fact that twice the operations per clock are happening in the same span as regular DDR. It does not, however, give you any real benefit, just higher scalability. The lack of DDR2 support also really has nothing to do with the chipset, its a CPU feature on Athlon64/FX architecture's, not a chipset one, so people bemoaning the lack of DDR2 need to look at AMD, not nVidia.Like I said before, the only feature needed from my point of view is PCI Express. I refuse to buy anymore PCI or AGP devices at this point knowing that in a year or two they will be useless. Unlike my CPU, I don't often change out my sound card, motherboard, SCSI card, or other such devices, so when it comes time to upgrade my system, PCI Express will be the order of the day for me.
Pumpkinierre - Thursday, March 25, 2004 - link
Good to see your real opinions, wesley #50. I too am worried about this slow latency DDR2 particularly with the a64 where I see system memory latency as being the bottleneck for improved gaming speed. AMD have got themselves a winner with a64/newcastle but still have mainboard issues as well as heavy debt. In these conditions, good poker dictates that you play conservatively. So I'm quite happy to see only DDR1 and PCI on the nF3-250 for the moment.