Work in progress... TCI K2

[The Stilt]

Any chance you are planning to make this work on win8?

The problem is not the program itself but the driver it uses to access all of the hardware registers.
The driver is not written by me so I have no control over adding Windows 8 support to it.

In my opinion: Windows 8 offers no improvements over Windows 7 (on the contrary) for 'advanced' users and it should not be installed on any high performance desktop systems. It seems to be more targeted to 'challenged users' and tablet computers.

But I?ll see if there is anything that could be done.

[Khenglish]

So I threw win7 and V1.1 on a spare HDD to see if it would work. When running as admin it closes quickly after launching it, with no text ever displayed. The only thing I see is RW-everything's startup splash image prior to everything closing.

I figured that I'd share my live CAS changing method for you to add to the next TCIK2 version if you want:

1. write 0x1F5F8 to D0F0xB8
2. write 0xFF to lowest byte of D0F0xDC
3. write desired CAS latency to D0F2x200 bits 0-4
4. set bit 0 of D0F1x10C high
5. write desired CAS latency to D0F2x200 bits 0-4
6. write 0x00 to lowest byte of D0F0xDC

The idea is to switch the memory to the low speed memory P-state so that when altering the high memory speed P-state, there is never a point where both DCTs have different latency settings. For me, only the lowest performance NBP-state uses this memory speed, so in the first two lines I force the northbridge into its lowest P-state. Changing the NBP-state definitions in D0F4x160 to call for M1 instead of M0 would result in a crash, so I do not know of alternative methods for systems that do not already use M1.

I've found that this method will result in a crash if the other memory timings are already tight. When I change the memory from 11-11-12-28 to 9-9-8-24, I need to change and set the CAS first, then I can do the other timings. This method does not force NBP0 if the IGP is under load, but it will cause NBP0 to be maintained if set during low IGP activity, then IGP activity goes high. AOD will sometimes cause a system crash with the northbridge forced to P0. When it doesn't crash, it resets the northbridge P-state control.

sidenote:
Trying to disable SCLK DPM by setting D0F0xB8 to 0x1F100, then clearing bit 0 of D0F0xBC does not work. The bit is cleared, but there is no effect. Trying to change SCLK through afterburner or AOD does something, but I don't know what. Setting the clock to anything default or higher results in a performance increase. I know DPM is not disabled because the system will still have random performance swings (with a corresponding stutter) during 3D rendering. I know it's not actually overclocked because I've "set" the SCLK to be over 2GHz once and the system did not crash, and just setting the default clock speed has the same performance impact as setting over 1GHz. I checked D0F0xB8x1F428, and BAPM and current limiting were not disabled. Disabling these anyway has no effect. I think all it does it cause the DPM to be more performance aggressive. Lowering clocks has the expected results, so it looks like the 7660G has a frequency cap bitlock somewhere. Even without real overclocking, just fully disabling DPM alone would net a large performance increase and reduce microstutter.

[The Stilt]

Work in progress...

For Trinity Comal (mobile) and non-K (desktop) APUs:

TCI GPU L.C.E (Lock Cracker Edition)

The GPU (SCLK) frequency on mobile and non-K series APUs is locked, meaning the GPU cannot be overclocked without raising the BCLK frequency. The software will bypass the lock and allow the user to overclock the GPU without touching the BCLK frequency

Theoretical limitations:

A10-4600M (7660G): 1066MHz
A10-4655M (7620G): 772MHz
A8-4500M (7640G): 1018MHz
A6-4400M (7520G): 1066MHz
A6-4455M (7500G): 658MHz

Tested on a retail HP Pavilion G6 laptop (A8-4500M), the GPU hits 900MHz with a minor (+0.0125V) voltage boost
Unfortunately increasing the voltage further does not make any difference as the temperatures get too high.
Also the Insyde bios on HP does not support memory beyond 1333MHz even when the DRAM officially supports it...

[FlanK3r]

Stilt, will be TCI working with Richland APU too?

[The Stilt]

Richland works fine however I have not tested with Bolton FCH yet.
Should not be too much of trouble.

[The Stilt]

Alpha...

[FlanK3r]

wow! Do you have some web with Paypal for free? This siftware need little support

[The Stilt]

Stock:



Unlocked:

[FlanK3r]

Only we need this version of TCIK :-(...Where will be public?

[The Stilt]

It is still cooking.

Trying to get Orochi Control Interface and TCI M^2 out at the same time.

[Opteron146]

It would require a full rewrite as without the GPU Trinity is a whole different chip.

Missing DPMs, dividers, voltage control, different external Pll outputs, etc.
No going to happen.

Ah sorry to hear that, thought of buying a cheap Athlon 750K and oc it till the end
But well, seems you'll have your reasons.

[VulgarHandle]

The Stilt,

Most important test of all:
Can it hold a steady 125fps on Quakelive on the map 'Cold Cathode'? (quakelive.com, free)

[FlanK3r]

nice, Quaker here ...If you find me, my profil:
http://www.quakelive.com/#!profile/s...lankeeeeeeeeer

Im playing TDM most time (elo around 1900), but I wanna begin with duels too

PS:srry for OT, I can try it later (in April) with Trinity

[The Stilt]

At the moment it seems that the first public version of the new TCI should be ready within a week.

New features:

- APU/CPU support increased from 3 to 70 models (i.e ALL Richland & Trinity based A-series APUs, Athlon CPUs, FirePro APUs, R-series APUs and Sempron CPUs are supported), both desktop and mobile.

- GPU frequency unlock for all locked APUs.

- Turbo Core recalibration

- New silicon leakage scale based on real production parameters (vs. estimation)

- LN2 Tune

- 'Node' version detection (minor stepping version).

- Support for a new ext. Pll model (Realtek RTM-880)

- A new dedicated driver (should be Win8 compatible, currently untested however)

- A GUI

The OCI will get all of the new and existing features, excluding the GPU related features of course.
In addition features such as CTC ('Chimeric Turbo Core' ) and TCF ('Turbo Core Fixer') will be available.

[FlanK3r]

Excelent work Stilt. You are the good adept to working in AMD team I mean.

[harpek]

This is a huge contribution to the community. Thank you!

Also - Updating BIOS to revision F4 on Gigabyte F2A85X-UP4 ended up with TCI causing reboots for me. Any info on that?

[The Stilt]

This is a huge contribution to the community. Thank you!

Also - Updating BIOS to revision F4 on Gigabyte F2A85X-UP4 ended up with TCI causing reboots for me. Any info on that?

The F4 bios version changes atleast programming of the internal Pll and CPU Core OCP limit.
Should not cause any issues thou.

There won't be any updates for the TCI K2 as the "APU Control Interface" will be released soon.

[harpek]

The F4 bios version changes atleast programming of the internal Pll and CPU Core OCP limit.
Should not cause any issues thou.

There won't be any updates for the TCI K2 as the "APU Control Interface" will be released soon.

Hey again! Apparently, the same issue hits Gigabytes own EasyTune too, so it's very likely something on their end.

APU Control Interface? Is that your "rebranding", or someone elses design?

[The Stilt]

APU Control Interface is basically a completely rewritten TCI K2.
It supports ALL 15h based desktop, mobile & embedded APUs (Richland, Trinity).
No matter if they are unlocked or not.

It also features a GUI and separate drivers instead of an assisting program (R&W).

It is still not ready and there is plenty to do, but currently it looks like this:





That is not likely to be the final look thou.

[FlanK3r]

this si awesome work Stilt!

[harpek]

That... Is... So... Cool! Love it.

An update on the BIOS/TCI/EasyTune issue. It seems to have resolved by itself after reflashing. Now both TCI and ET work as intended.

Also - Been fiddling around a bit with different parameters. I can't seem to get RAM stable at 2400MHz, no matter the NB frequency or voltage(constrained by temp), though tighter timings and 2133MHz work just fine. Rock solid. I don't want to feed the NB too much voltage either as temperatures skyrocket when i do. It's interesting how far the NB will go frequencywise without further changes, though. And how much this chip loves NB frequency for memory bandwidth. I wonder how far that goes, if there's a sweetspot, or if scaling will only be constrained by memory/clockability. I haven't toyed too much with the timings made available by TCI yet, though, as i'm not sure how they behave. Everything is a learning experience

Lovely chip! Haven't had this much fun with a processor since i was a kid playing around with a Duron 1600 MHz on Socket A.

And again - Thank you, The Stilt, for this great contribution.

[FlanK3r]

I got question from few Czechs APU fans, if The great man The STilt can add to APU Control Interface something as "profile saves/load"

[The Stilt]

I've been thinking of adding the export / import feature, however the more I have looked into it the harder it seems.

The level of integration on 15h APUs is so high that basically every single setting has a relation to another two or three settings atleast.

If one setting is changed the software might need to recalibrate two other settings or perform a specific operation just to get the first one changed reliably.

In case the imported profile would change even ten settings... You'll do the math

There are also many settings which are product specific (APU model, motherboard, DRAM, GPU) and some which cannot be changed in flight (prior AGESA init only).

This is why TCI K2 was restricted to K-series APUs (three models) only.
ACI supports over 150 different APU/CPU models.

If the program would be a simple memory or MSR tweaker only, it could be easily done.
However in this extent it is quite impossible.

[Khenglish]

APU Control Interface is basically a completely rewritten TCI K2.
It supports ALL 15h based desktop, mobile & embedded APUs (Richland, Trinity).
No matter if they are unlocked or not.

It also features a GUI and separate drivers instead of an assisting program (R&W).

It is still not ready and there is plenty to do, but currently it looks like this:





That is not likely to be the final look thou.

Let me know if you want a tester. This toshiba has surprisingly good cooling despite how toshiba cheaped out on the rest of the system. I do have thermal room to raise voltages.

I suppose you never found a workaround for the boost lock? This laptop's BIOS is a disaster so I can't find the pci config space write. 3.2ghz would be a hell of a lot better than 2.7.

[zsx47]

I have integrated gpu enabled and everything ,but for some reason every time i launch the program it crashes. is there something i'm missing.
System specs
OS: windows 8 (might be the cause)
CPU: A10-5800K@4.5GHz
RAM: 2x4GB dual channel DDR3 10-10-11-28@1866Mhz
GPU: ASUS AMD Radeon HD 7770 2GB GDDR5
MOBO: ASUS F2A85-v PRO bios ver. 6104