VOGONS


First post, by MattRocks

User metadata
Rank Member
Rank
Member

I sense I am one of a handful who values the Ensoniq AudioPCI (ES1370), and I value it because it was philosophically the Voodoo of sound cards - a simple FIFO bitstream with minimal CPU overhead.

My understanding is the original ES1370 was brutally optimised for 44.1KHz (CD audio native) bitrates and its integer divisions (22KHz, 11KHz), such that anything else needed resampling by the CPU to match. And that was okay because 22KHz was for gaming the standard that mattered. Less CPU work meant in some games I'd notice an extra 5FPS by swapping out an SBLive for an ES1370.

SB Live was doing more on the DSP, but ordering that stuff and loading the supporting files needed orchestrating by the CPU. So, when all that mattered was maximum FPS and critical in-game emotional feedback like "Humiliation!" there was nothing better than an ES1370.

After acquiring Ensoniq and revising the chips, the biggest thing Creative changed was the sample rate to 48KHz (DVD standard). That meant the CPU needed to resample all 22KHz game streams to match DVD audio. For me, that change marked the end of an era.

When looking at the earlier PCBs I note subtle differences between the 44.1KHz Ensoniq AudioPCI ES1370 and subsequent 44.1KHz Creative PCI64 ES1370 - do these hardware tweaks reflect actual improvements, or cost cutting regressions? Did Creative make other changes to the Ensoniq design, and were those changes good or bad?

Reply 1 of 12, by leileilol

User metadata
Rank l33t++
Rank
l33t++

No. The 3dfx analogy would be better applied to Aureal hardware:

- Own proprietary API early (like Glide)
- Diamond Monster branding
- There's A3D wrappers
- Could not move beyond DirectX 7
- Died in 2000 after an aggressive acquisition and legal issues

Creative is an nVidia like with big arrogance to match, and AudioPCI is the Vanta.

The 48khz change in ES1371/PCI128 was more in revising for AC'97 than anything else.

apsosig.png
long live PCem
FUCK "AI". It is a tool of fascism. We do not need it. We do not use it.

Reply 2 of 12, by MattRocks

User metadata
Rank Member
Rank
Member
leileilol wrote on Today, 14:14:
No. The 3dfx analogy would be better applied to Aureal hardware: […]
Show full quote

No. The 3dfx analogy would be better applied to Aureal hardware:

- Own proprietary API early (like Glide)
- Diamond Monster branding
- There's A3D wrappers
- Could not move beyond DirectX 7
- Died in 2000 after an aggressive acquisition and legal issues

Creative is an nVidia like with big arrogance to match, and AudioPCI is the Vanta.

The 48khz change in ES1371/PCI128 was more in revising for AC'97 than anything else.

Ensoniq AudioPCI and 3Dfx Voodoo share one design philosophy: minimalism in the hot path. They both pump data from CPU to expansion card using a simple command FIFO and minimal driver intervention. Ensoniq AudioPCI and 3Dfx Voodoo both run fine on a Pentium 100.

The Aureal Vortex represents the complete opposite of that! Aureal Vortex is philosophically more aligned to GeForce 256: far more capable, far more CPU offloading, but with significantly higher driver and control overhead due to its DSP-driven spatial pipeline.

Reply 3 of 12, by ZanQuance

User metadata
Rank Member
Rank
Member
MattRocks wrote on Today, 14:33:

Ensoniq AudioPCI and 3Dfx Voodoo share one design philosophy: minimalism in the hot path. They both pump data from CPU to expansion card using a simple command FIFO and minimal driver intervention. Ensoniq AudioPCI and 3Dfx Voodoo both run fine on a Pentium 100.

The Aureal Vortex represents the complete opposite of that! Aureal Vortex is philosophically more aligned to GeForce 256: far more capable, far more CPU offloading, but with significantly higher driver and control overhead due to its DSP-driven spatial pipeline.

It's been a long while since I looked at how the Voodoo 1 and 2 chips operated, but I do know that Aureal even called themselves the 3DFX of Audio.
The Aureal Vortex 1 and 2 ASIC's aren't all that complicated in their design and there seem to still be many misunderstandings on what they can and cannot do.
They are not like GPU's at all, they do not contain internal geometry processing units for offloading or calculating room reflections for wavetracing or reverb. They contain a set of FIFOs which are then linked into different fixed function processing units via the Vortex Data Bus. Those fixed function units do allow uploading of certain parameters or coefficients for filtering or panning of the audio streams. But aside from that they are very much CPU/Software dependent for preparing and processing all 3D Audio functions, which is why A3D has cpu overhead whenever it is turned on.

The diagrams which Aureal shared in their chip product fliers do accurately represent what you get in the Vortex 1 and 2 ASIC's.
Here is the one for the Vortex 1 Au8820
and here is the AU8830 Vortex 2
The A3D Engine seemed mystical without further investigation, however it turns out to be a collection of FIR filters attached to delay lines for HRTF processing on 16 audio streams. It only allowed uploading coefficients and not any world geometry.
Inside the Wavetable engine where the reflections were "processed" we see a wavetable engine which also allowed being placed in a direct sound mode for 64 direct sound streams to be played through it, also attached to delay lines and panning/filters, nothing GPU accelerated here either.
They were very cost effective ASIC's which focused mostly on fixed function units piped through a routable bus.

And all this isn't said to dismiss that Ensoniq may have also followed this simplicity in design, however I do not believe the "crown" of being 3DFX like should go to them.
It belongs rightfully to Aureal for all the reasons leileilol listed and more.
[Edit: fixed typos]

Last edited by ZanQuance on 2026-04-14, 20:25. Edited 1 time in total.

Reply 4 of 12, by MattRocks

User metadata
Rank Member
Rank
Member
ZanQuance wrote on Today, 15:29:
It's been a long while since I looked at how the Voodoo 1 and 2 chips operated, but I do know that Aureal even called themselves […]
Show full quote
MattRocks wrote on Today, 14:33:

Ensoniq AudioPCI and 3Dfx Voodoo share one design philosophy: minimalism in the hot path. They both pump data from CPU to expansion card using a simple command FIFO and minimal driver intervention. Ensoniq AudioPCI and 3Dfx Voodoo both run fine on a Pentium 100.

The Aureal Vortex represents the complete opposite of that! Aureal Vortex is philosophically more aligned to GeForce 256: far more capable, far more CPU offloading, but with significantly higher driver and control overhead due to its DSP-driven spatial pipeline.

It's been a long while since I looked at how the Voodoo 1 and 2 chips operated, but I do know that Aureal even called themselves the 3DFX of Audio.
The Aureal Vortex 1 and 2 ASIC's aren't all that complicated in their design and there seem to still be many misunderstandings on what they can and cannot do.
They are not like GPU's at all, they do not contain internal geometry processing units for offloading or calculating room reflections for wavetracing or reverb. They contain a set of FIFOs which are then linked into different fixed function processing units via the Vortex Data Bus. Those fixed function units do allow uploading of certain parameters or coefficients for filtering or panning of the audio streams. But aside from that they are very much CPU/Software dependent for preparing and processing all 3D Audio functions, which is why A3D has cpu overhead whenever it is turned on.

The diagrams which Aureal shared in their chip product fliers do accurately represent what you get in the Vortex 1 and 2 ASIC's.
Here is the one for the Vortex 1 Au8820
and here is the AU8830 Vortex 2
The A3D Engine seemed mystical without further investigation, however it turns out to be a collection of FIR filters attached to delay lines for HRTF processing on 16 audio streams. It only allowed uploading coefficients and not any world geometry.
Inside the Wavetable engine where the reflections were "processed" we see a wavetable engine which also allowed being placed in a direct sound mode for 64 direct sound streams to be played through it, also attached to delay lines and panning/filters, nothing GPU accelerated here either.
They were very cost effective ASIC's which focused mostly on fixed function units piped through a routable bus.

And all this isn't said to dismiss that Ensoniq may have also followed this simplicity in design, however I do not believe the "crown" of being 3DFX like should go to them.
It belongs to rightfully to Aureal for all the reasons leileilol listed and more.

🤣 - this is not about crowing anyone!

I'll give you one point: 3Dfx and Aureal both went bust and their parts discarded, so they have that much in common - but I'm not talking about business leadership or financial management, I'm talking about how their products behaved when installed into a PC.

3Dfx approach did not produce accurate renderings - they just focussed on using less CPU time than more design accurate renderers like FireGL.
3Dfx approach produced higher FPS.

Ensoniq approach did not produce a more authentic audio - they just focussed on using less CPU time when playing the 22KHz audio produced by games.
Ensoniq approach produced higher FPS.

Aureal cards produced spatially accurate audio positioning - FireGL territory, slow but correct.
Aureal cards produced lower FPS.

In 90s PC games, all audio mixing was done by the game engine itself before audio ever reached the sound card. Under that constraint, sound cards that added effects or mixing could only increase CPU load and lower FPS. Ensoniq leaned into that constraint by making a sound processing pipeline that did little as possible. And that materially mattered in competitive gaming because FPS chasing was real at every tier in the 90s.

Where Ensoniq sucks is in accuracy, immersion, sensations. But if accuracy, immersion, sensations matter then 16bit renderings of low detail 8 bit textures sucks too. So, let's be consistent: Are we praising each solution for its accurate precision or its fast approximation?

P.S. Ensoniq, Aureal, and Sensaura each presented a threat to Creative - so Creative acquired all three, but only Ensoniq provided a synergy that Creative could immediately profit from.

Reply 5 of 12, by ZanQuance

User metadata
Rank Member
Rank
Member
MattRocks wrote on Today, 18:59:
LOL - this is not about crowing anyone! […]
Show full quote

🤣 - this is not about crowing anyone!

I'll give you one point: 3Dfx and Aureal both went bust and their parts discarded, so they have that much in common - but I'm not talking about business leadership or financial management, I'm talking about how their products behaved when installed into a PC.

3Dfx approach did not produce accurate renderings - they just focussed on using less CPU time than more design accurate renderers like FireGL.
3Dfx approach produced higher FPS.

Ensoniq approach did not produce a more authentic audio - they just focussed on using less CPU time when playing the 22KHz audio produced by games.
Ensoniq approach produced higher FPS.

Aureal cards produced spatially accurate audio positioning - FireGL territory, slow but correct.
Aureal cards produced lower FPS.

In 90s PC games, all audio mixing was done by the game engine itself before audio ever reached the sound card. Under that constraint, sound cards that added effects or mixing could only increase CPU load and lower FPS. Ensoniq leaned into that constraint by making a sound processing pipeline that did little as possible. And that materially mattered in competitive gaming because FPS chasing was real at every tier in the 90s.

Where Ensoniq sucks is in accuracy, immersion, sensations. But if accuracy, immersion, sensations matter then 16bit renderings of low detail 8 bit textures sucks too. So, let's be consistent: Are we praising each solution for its accurate precision or its fast approximation?

P.S. Ensoniq, Aureal, and Sensaura each presented a threat to Creative - so Creative acquired all three, but only Ensoniq provided a synergy that Creative could immediately profit from.

Actual testing says otherwise later models are slow (ES1373 model, not ES1370) more testing required.
It's okay, I don't think we are trying to incite any anecdotal based debates here.
I am sure the Ensoniq Audio PCI has some good things about it.
I can't think of any off the top of my head right now, but I am sure there are plenty.

A sound cards purpose is to faithfully convert digital audio information into a "clean" analog signal. If a sound card can do this, then it's served its purpose.
Anything else is icing on the cake.

Last edited by ZanQuance on 2026-04-14, 21:00. Edited 1 time in total.

Reply 6 of 12, by onethirdxcubed

User metadata
Rank Newbie
Rank
Newbie

The ES1370 was briefly prized by audiophiles because it ran at 44.1khz avoiding the software resampling, could be used with an ASIO driver for exclusive hardware buffered output as well as with Digital Audio Extraction from a CD, and had a lower noise floor than onboard audio. But for games from the late 90s and onward the CPU overhead from software mixing and positional audio was more expensive than the resampling from 44.1 to 48khz. Also the OPL emulation was laughably bad which made it a non starter for DOS gaming (though now we would either use the General MIDI or run the game in DOSBox).

Last edited by onethirdxcubed on 2026-04-14, 20:23. Edited 1 time in total.

Reply 7 of 12, by MattRocks

User metadata
Rank Member
Rank
Member
ZanQuance wrote on Today, 20:20:
Actual testing says otherwise It's okay, I don't think we are trying to incite any anecdotal based debates here. I am sure the […]
Show full quote
MattRocks wrote on Today, 18:59:
LOL - this is not about crowing anyone! […]
Show full quote

🤣 - this is not about crowing anyone!

I'll give you one point: 3Dfx and Aureal both went bust and their parts discarded, so they have that much in common - but I'm not talking about business leadership or financial management, I'm talking about how their products behaved when installed into a PC.

3Dfx approach did not produce accurate renderings - they just focussed on using less CPU time than more design accurate renderers like FireGL.
3Dfx approach produced higher FPS.

Ensoniq approach did not produce a more authentic audio - they just focussed on using less CPU time when playing the 22KHz audio produced by games.
Ensoniq approach produced higher FPS.

Aureal cards produced spatially accurate audio positioning - FireGL territory, slow but correct.
Aureal cards produced lower FPS.

In 90s PC games, all audio mixing was done by the game engine itself before audio ever reached the sound card. Under that constraint, sound cards that added effects or mixing could only increase CPU load and lower FPS. Ensoniq leaned into that constraint by making a sound processing pipeline that did little as possible. And that materially mattered in competitive gaming because FPS chasing was real at every tier in the 90s.

Where Ensoniq sucks is in accuracy, immersion, sensations. But if accuracy, immersion, sensations matter then 16bit renderings of low detail 8 bit textures sucks too. So, let's be consistent: Are we praising each solution for its accurate precision or its fast approximation?

P.S. Ensoniq, Aureal, and Sensaura each presented a threat to Creative - so Creative acquired all three, but only Ensoniq provided a synergy that Creative could immediately profit from.

Actual testing says otherwise
It's okay, I don't think we are trying to incite any anecdotal based debates here.
I am sure the Ensoniq Audio PCI has some good things about it.
I can't think of any off the top of my head right now, but I am sure there are plenty.

A sound cards purpose is to faithfully convert digital audio information into a "clean" analog signal. If a sound card can do this, then it's served its purpose.
Anything else is icing on the cake.

He did not test an Ensoniq ES1370!

What he tested is a later Creative derivative optimised for DVD audio, such as ES1371 or ES1373. It also matters which drivers you install: the original lightweight Ensoniq driver, or a later Creative driver with software positional audio.

One, I know he could not have tested an Ensoniq ES1370 because his FPS figures show its something else. Two, I checked his website - he absolutely states it is a Creative PCI128 in his reviews, and holds it up in his video, and a PCI128 is absolutely not an ES1370 - it probably cannot even be used with Ensoniq drivers.

Last edited by MattRocks on 2026-04-14, 20:38. Edited 5 times in total.

Reply 8 of 12, by MattRocks

User metadata
Rank Member
Rank
Member
onethirdxcubed wrote on Today, 20:21:

The ES1370 was briefly prized by audiophiles because it ran at 44.1khz avoiding the software resampling, could be used with an ASIO driver for exclusive hardware buffered output as well as with Digital Audio Extraction from a CD, and had a lower noise floor than onboard audio. But for games from the late 90s and onward the CPU overhead from software mixing and positional audio was more expensive than the resampling from 44.1 to 48khz. Also the OPL emulation was laughably bad which made it a non starter for DOS gaming (though now we would either use the General MIDI or run the game in DOSBox).

You guys are falling into Creative Labs marketing for EMU10K1 trap.

You are mixing up the original Ensoniq ES1370 and the post-Ensoniq ES1371. They are different chips. There is no resampling when playing 22KHz game audio on ES1370. There is CPU resampling when playing 22KHz game audio on ES1371 or newer.

Reply 9 of 12, by ZanQuance

User metadata
Rank Member
Rank
Member
MattRocks wrote on Today, 20:22:

He did not test an Ensoniq ES1370!

I'm pretty sure he tested a later Creative derivative optimised for DVD audio, such as ES1371 or ES1373. It also matters which drivers you install: the original lightweight Ensoniq driver, or a later Creative driver with software positional audio.

Very possible and good catch.
After reviewing his youtube video I see he didn't actually mention which chip version his card used, so I stand corrected if it's of the later versions.
Let me reach out to Phil and ask him which one his card used.

Reply 10 of 12, by MattRocks

User metadata
Rank Member
Rank
Member
ZanQuance wrote on Today, 20:36:
Very possible and good catch. After reviewing his youtube video I see he didn't actually mention which chip version his card us […]
Show full quote
MattRocks wrote on Today, 20:22:

He did not test an Ensoniq ES1370!

I'm pretty sure he tested a later Creative derivative optimised for DVD audio, such as ES1371 or ES1373. It also matters which drivers you install: the original lightweight Ensoniq driver, or a later Creative driver with software positional audio.

Very possible and good catch.
After reviewing his youtube video I see he didn't actually mention which chip version his card used, so I stand corrected if it's of the later versions.
Let me reach out to Phil and ask him which one his card used.

But he holds the card up in the video and it's clearly a later Creative package - too blurry to read the text, but its brown on black and that is characteristic of later variants.

Reply 11 of 12, by ZanQuance

User metadata
Rank Member
Rank
Member

Pausing the video at the right moment, I can read it's the ES1373 version.
So I stand corrected.
Now this begs the question how much more performant is the ES1370 version in comparison?

Reply 12 of 12, by onethirdxcubed

User metadata
Rank Newbie
Rank
Newbie
MattRocks wrote on Today, 20:24:
onethirdxcubed wrote on Today, 20:21:

The ES1370 was briefly prized by audiophiles because it ran at 44.1khz avoiding the software resampling, could be used with an ASIO driver for exclusive hardware buffered output as well as with Digital Audio Extraction from a CD, and had a lower noise floor than onboard audio. But for games from the late 90s and onward the CPU overhead from software mixing and positional audio was more expensive than the resampling from 44.1 to 48khz. Also the OPL emulation was laughably bad which made it a non starter for DOS gaming (though now we would either use the General MIDI or run the game in DOSBox).

You guys are falling into Creative Labs marketing for EMU10K1 trap.

You are mixing up the original Ensoniq ES1370 and the post-Ensoniq ES1371. They are different chips. There is no resampling when playing 22KHz game audio on ES1370.

No, I'm perfectly aware of the difference between the ES1370 and the ES1371/1373, as well as that all of the ES1371/1373 and EMU10k1 operated at multiples of 48khz only.
I also know that (at least for Windows 9x with WDM audio drivers and software mixing, I can't guarantee it's true for VxDs but it might still be if using DirectSound) if the sound card supports 48khz Windows will always interpolate 22khz to 48khz (even if 22khz is supported directly!), but if the sound card only supports up to 44.1khz then the output rate will be properly left at 22khz.
Also I know that with fixed rate AC97 the software resampling of 44.1khz to 48khz or vice versa causes very audible artifacts, but later AC97 codecs support 44.1 khz directly and Windows won't resample in this case.
It may be resampling only if the maximum supported rate is >2x the requested rate?
I have no idea why this behavior is the way it is though and only discovered this while writing my own HD Audio drivers so it's not widely known. For this reason I won't expose any supported sample rate higher than 48khz as that causes KMixer to resample absolutely everything.

So I am acknowledging there are valid reasons why you might want an ES1370 specifically to avoid resampling, I just don't think it makes a huge difference for gaming but maybe that's because I mostly played later FPSes which supported 48khz and where positional audio gave you a huge advantage at tracking down your opponents.