Great info! What bus does the 9550 you tested have? I've got one with more rare 128-bit bus from ASUS, A9550GE they call it. I know the more common variant has 64-bit bus.

.

obobskivich wrote:

The high-poly was picked because it's the only place where the X1600 seemed to pull ahead (I don't have an explanation, but it was interesting).

Five high frequency vertex shaders running the TnL.

3dmark01 is obvious very dependent on cpu, a pentium4-2.0 is way too slow to show the difference between these cards.
and why not post a graph of game1 high detail FPS?

.

Other notes: The X1600 dropped to <10 FPS for 1-2 seconds during Game 1 High Detail and Game 3 High Detail. I don't know why, it just did. It didn't have any sort of artefacting, corruption, etc throughout and completed the benchmark just fine. Just felt it was worth mentioning though.

My AGP Radeon HD4670 does this also. No idea why.

obobskivich wrote:

As far as the part selection, it was dictated primarily by availability. However I think the Willamette is a good analog for Pentium 3 or Athlon builds, which seem fairly popular these days. You aren't incorrect that it does preclude "highest possible scores" (which isn't my goal/purpose), but otoh the theoretical/synthetic results tend not to move very much even as CPU gets faster.

yeah that surely makes sense, but for such a slow platform, i would say that using video cards faster than gf5700/gf6200/r9600 is a complete waste.

BSA Starfire wrote:

My AGP Radeon HD4670 does this also. No idea why.

Out of curiosity, what CPU and drivers are you using for the 4670?

noshutdown wrote:

yeah that surely makes sense, but for such a slow platform, i would say that using video cards faster than gf5700/gf6200/r9600 is a complete waste.

I'm not saying you're wrong - I think the only big advantage to something like the 5800/9800 is support for higher levels of AA/resolution (I know, from past usage, the 5800U on this system has no trouble running 4x AA pretty much across the board at 1280x960, which the 9550/9600 may not be able to achieve). Power consumption/noise-wise I'd completely agree - the 5800 and 9800 are far from light-weights in that regard. 😊

Added two new cards:

- Chaintech GeForce FX 5700 Ultra ('Apogee GeForce FX 5700U'), fairly sure this is a DDR-II card.
- Chaintech GeForce FX 5900XT ('SA5900XT')

Also re-ran the 5800U with this PSU for power #s, and to see 4X AA in action. All new graphs to replace what was originally posted. I used the "new" VP 880 and 5800U numbers as well; the 5800U's 3D01 score changes by around 10 points (oh wowie!), the VP 880 by more like 1000 (see my second post in the thread about this).

Out of curiosity, what CPU and drivers are you using for the 4670?

CPU was a northwood 2.4B, 845E chipset, XP SP3, Drivers were Catalyst 14.4.

obobskivich wrote:

Was also going to post a reply about the VP880: I officially give up trying to understand this card. 🤣 It was bugging me not knowing if there were CuMine results in the charts or not, so I re-installed the VP880 in the P4 and fired up 3D01. Most of the scores didn't budge (all fill rates and game 4 are the same), but the overall score went up almost 1000 pts (to 6131), and some of the other game tests went up slightly (game 2 LD went up to 101.6 FPS, from the 89.6 in the chart). None of this changes its ordering in the charts (only the raw score and Game 2 LD actually changed), nor does it make sense to me: after posting I got to thinking about it and realized there's no way the XP results I used for this could have come from the CuMine, because the 3DLabs XP driver refuses to work with the i815 XP INF (Windows will not start), so it had to be on the P4 with the i845 to be running XP. So somehow by doing absolutely nothing but re-installing the card I gained almost 1000 pts in 3D01, and ~10 fps in a few game tests, while not improving at all on any of the fill-rate/etc metrics using the same drivers/configuration. Makes complete sense, doesn't it?

The 3dlabs drivers have some quirks. The vp990 drivers will not work on 2000 beyond sp1 iirc. The drivers also default to vertical sync ON [which 3dmark2001se does not override] and don't always ask you to reboot when a setting has changed which requires a reboot to take effect.

I've been looking at my results with a p4 northwood [vp870], with 2gb ram, windows 2000, HT off; 845 motherboard. With the cpu at 2ghz I get a score of 7360, with 2.8ghz I get 7840. Yours should be the same vpu speed [just more ram on the card] so I suppose the willamette holds it back quite a lot. In fact my athlon 1600 was faster [6800]. I wonder actually if my northwood at 2ghz, with HT on, would give a score very close to the dual mp1600s.

Northwood 2.0 is going to be faster than Willamette 2.0 (IIRC they all have 512k cache, for example); consider the Willamette 2.0 to be more like a 1.2-1.4GHz Athlon (not XP) or similar Pentium 3. Not at all surprising the "3DMarks" score is higher - but what's the FPS actually look like for tests that the Wildcat struggles with (like Nature)? I also think the RAM on the 880Pro is slower than the 870/970/etc - those cards use BGA chips, 880Pro just has normal DDR modules (of course they could use slow BGA chips, but I'm guessing instead that 3DL sacrificed memory speed for memory capacity to keep costs around the same). IOW: I'm not at all surprised to see an 870 with a faster CPU scoring somewhat higher.

As far as driver quirks go - I've never had the drivers not notify me of a reboot, but whenever I install them they default Max Geometry which hurts performance for gaming pretty seriously. Vsync always appears to be off though. I don't know if they're separate drivers per GPU or not, but the readme for what I installed said 2000 SP2 (or higher) or XP as the minimum; I've only ever tried on 2k SP4 and XP SP2. It worked fine with both on the i845, but with the i815 it refuses to work in XP if the 815 loads its INF driver (works great in 2k SP4 though).

.