News NVIDIA and MediaTek want to bring RTX graphics to ARM laptops

NTMBK

Lifer
Nov 14, 2011
10,240
5,027
136
What would an ARM-based laptop look like with an RTX graphics card? That's something NVIDIA is exploring together with MediaTek, a company best known for building ARM-based chips. Together, they're building a reference laptop platform that'll support Chromium, Linux and NVIDIA SDKs (software development kits). While it's unclear what, if anything, this partnership will lead to, it's not hard to get excited about the idea of a next-generation Chromebook that's light, energy efficient and equipped with NVIDIA's ray-tracing RTX hardware, even if they inevitably end up being stripped down.


I don't really get why Mediatek is involved here? NVidia can clearly produce their own ARM SoCs. And I'm not taking it seriously until it runs Windows.

BUT... I can definitely see a future where an ARM laptop chip can emulate x86 games faster than an x86 chip in the same power envelope. And at that point, this could be viable.
 
  • Like
Reactions: SarahKerrigan

Shivansps

Diamond Member
Sep 11, 2013
3,855
1,518
136
If you switching alignment checks off in the CPU for EL0, then you will not get alignment errors - i am not talking about some OS level fixing but switching it off in HW. Did you check in the kernel, that HW alignment checks were switched off in SCTLR_EL1 for the CPUs?

I had enabled unaligned support both on boot asm code and in /proc/cpu/alignment, and i was still getting a SIGBUS, no idea why. But, this was under armhf, since there was no 64 bit kernel avalible at the time, i never actually tested it under AArch64. ARM64 makes any difference here?

Not sure the discussion led anywhere last time we discussed it: https://forums.anandtech.com/threads/arm-apple-high-end-cpu-intel-replacement.2571738/post-40007359
I provided code and evidence of mis alignment support. But no code was provided in answer to show what instruction led to the fault.

Because i was getting a SIGBUS trying a game on a RPI, since ive been playing this game since i was a kid and it is open source i took it upon myself to find the cause and fix it, i did just that and i moved on.

But if you insist, this is one of the functions were the game was crashing.

C++:
bool mc_check_sldc(int offset)
{
    if (offset > Mc_pm->sldc_size-5) //no way is this big enough
        return false;
    char *type_p = (char *)(Mc_pm->shield_collision_tree+offset);
 
    // split and polygons
    vec3d *minbox_p = (vec3d*)(Mc_pm->shield_collision_tree+offset+5);
    vec3d *maxbox_p = (vec3d*)(Mc_pm->shield_collision_tree+offset+17);

    // split
    unsigned int *front_offset_p = (unsigned int*)(Mc_pm->shield_collision_tree+offset+29);
    unsigned int *back_offset_p = (unsigned int*)(Mc_pm->shield_collision_tree+offset+33);

    // polygons
    unsigned int *num_polygons_p = (unsigned int*)(Mc_pm->shield_collision_tree+offset+29);

    unsigned int *shld_polys = (unsigned int*)(Mc_pm->shield_collision_tree+offset+33);



    // see if it fits inside our bbox
    if (!mc_ray_boundingbox( minbox_p, maxbox_p, &Mc_p0, &Mc_direction, NULL ))    {
        return false;
    }

    if (*type_p == 0) // SPLIT
    {
            return mc_check_sldc(offset+*front_offset_p) || mc_check_sldc(offset+*back_offset_p);
    }
    else
    {
        // poly list
        shield_tri    * tri;
        for (unsigned int i = 0; i < *num_polygons_p; i++)
        {
            tri = &Mc_pm->shield.tris[shld_polys[i]];
                     
            mc_shield_check_common(tri);

        } // for (unsigned int i = 0; i < leaf->num_polygons; i++)
    }

    // shouldn't be reached
    return false;
}

But not sure were you are trying to get to, i was not seeing illusions, that was happening.
 
Last edited:

Thala

Golden Member
Nov 12, 2014
1,355
653
136
I had enabled unaligned support both on boot asm code and in /proc/cpu/alignment, and i was still getting a SIGBUS, no idea why. But, this was under armhf, since there was no 64 bit kernel avalible at the time, i never actually tested it under AArch64. ARM64 makes any difference here?



Because i was getting a SIGBUS trying a game on a RPI, since ive been playing this game since i was a kid and it is open source i took it upon myself to find the cause and fix it, i did just that and i moved on.

But if you insist, this is one of the functions were the game was crashing.

C++:
bool mc_check_sldc(int offset)
{
    if (offset > Mc_pm->sldc_size-5) //no way is this big enough
        return false;
    char *type_p = (char *)(Mc_pm->shield_collision_tree+offset);

    // split and polygons
    vec3d *minbox_p = (vec3d*)(Mc_pm->shield_collision_tree+offset+5);
    vec3d *maxbox_p = (vec3d*)(Mc_pm->shield_collision_tree+offset+17);

    // split
    unsigned int *front_offset_p = (unsigned int*)(Mc_pm->shield_collision_tree+offset+29);
    unsigned int *back_offset_p = (unsigned int*)(Mc_pm->shield_collision_tree+offset+33);

    // polygons
    unsigned int *num_polygons_p = (unsigned int*)(Mc_pm->shield_collision_tree+offset+29);

    unsigned int *shld_polys = (unsigned int*)(Mc_pm->shield_collision_tree+offset+33);



    // see if it fits inside our bbox
    if (!mc_ray_boundingbox( minbox_p, maxbox_p, &Mc_p0, &Mc_direction, NULL ))    {
        return false;
    }

    if (*type_p == 0) // SPLIT
    {
            return mc_check_sldc(offset+*front_offset_p) || mc_check_sldc(offset+*back_offset_p);
    }
    else
    {
        // poly list
        shield_tri    * tri;
        for (unsigned int i = 0; i < *num_polygons_p; i++)
        {
            tri = &Mc_pm->shield.tris[shld_polys[i]];
                
            mc_shield_check_common(tri);

        } // for (unsigned int i = 0; i < leaf->num_polygons; i++)
    }

    // shouldn't be reached
    return false;
}

But not sure were you are trying to get to, i was not seeing illusions, that was happening.

I just compiled FS2_open from git master branch for both Windows x64 and Windows ARM64 with Visual Studio 2019 - and both variants (first variant under x64 emulation and the second native) are running without any crashes. In short, i cannot verify your alignment issue at all - how could I even - given that unaligned access is supported with ARMv8?
 
Last edited:

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
It doesn't run windows so it's fairly irrelevant. Chromium is virtually fine with just about any other graphics vendor and Linux is absolutely irrelevant for client ...

Nvidia is practically paperweight without Windows so it this not newsworthy in the slightest ...
 

SarahKerrigan

Senior member
Oct 12, 2014
373
539
136
I would not be surprised if this collaboration has a long-term eye toward Windows, even though it's not being mentioned right now. Especially if the rumors about Samsung preparing to launch Exynos/Radeon Windows laptops are true.
 
  • Like
Reactions: Tlh97 and NTMBK

Gideon

Golden Member
Nov 27, 2007
1,646
3,711
136
Especially if the rumors about Samsung preparing to launch Exynos/Radeon Windows laptops are true.
I wonder if they can do it. AFAIK AMD's license mentioned Samsung can only use their GPUs in devices that are not in direct competiton with AMD's own offerings. Laptops to me would seem to breach that.

Then again maybe it's a more intimate partnership rather than pure licencing? If AMD want's to make ARM chips, samsung could offer quite a few IP blocks AMD does not have
 

SarahKerrigan

Senior member
Oct 12, 2014
373
539
136
I wonder if they can do it. AFAIK AMD's license mentioned Samsung can only use their GPUs in devices that are not in direct competiton with AMD's own offerings. Laptops to me would seem to breach that.

Then again maybe it's a more intimate partnership rather than pure licencing? If AMD want's to make ARM chips, samsung could offer quite a few IP blocks AMD does not have

Interesting point, and one I hadn't thought of! Maybe "direct competition" is sufficiently broadly defined that ARM machines, even ones in a laptop form factor and running NT, don't count?
 
  • Like
Reactions: Tlh97

Thala

Golden Member
Nov 12, 2014
1,355
653
136
BUT... I can definitely see a future where an ARM laptop chip can emulate x86 games faster than an x86 chip in the same power envelope. And at that point, this could be viable.

Speaking of emulation i did compile 7zip for x86, x64, arm64 from the same source with same compiler setting using MSVC16 (Visual Studio 2019). And then i ran the benchmark:
"7zip b -md20" of all 3 version on the Surface Pro X. It is using 8 threads in all 3 versions. Results below:

Microsoft SQ1 @3GHz.
arm64: 19660 (100%)
x86: 11211 (57%)
x64: 13837 (70%)

As reference my desktop machine:

Intel Skylake 6700k @4.2GHz:
x64: 24400
x86: 21200
 

Shivansps

Diamond Member
Sep 11, 2013
3,855
1,518
136
So 4 A76 cores + 4 A55 cores still cant beat a 6 year old 6700K quad core +HT native vs native? The result is not that impressive when you realise there is a 5 year gap there. Thats around the score of a 3200G.

The 4300U (4/4) gets around 15K and the 4500U (6/6) around 22K, and thats the low end, the 8CX is a premium chip as result the fair thing would be to compare it to a 4800U that gets well over 40k.
 
Last edited:

podspi

Golden Member
Jan 11, 2011
1,965
71
91
I wonder if they can do it. AFAIK AMD's license mentioned Samsung can only use their GPUs in devices that are not in direct competiton with AMD's own offerings. Laptops to me would seem to breach that.

Then again maybe it's a more intimate partnership rather than pure licencing? If AMD want's to make ARM chips, samsung could offer quite a few IP blocks AMD does not have

Do we know for a fact that the Exynos chipset is going to be ARM based? This could be x86+radeon+Samsung IP blocks.

Alternatively, as you say, maybe they're working on a jointly branded ARM SoC as a way for AMD to hedge against ARM.

Exciting times.
 

jpiniero

Lifer
Oct 1, 2010
14,620
5,231
136
Do we know for a fact that the Exynos chipset is going to be ARM based? This could be x86+radeon+Samsung IP blocks.

Alternatively, as you say, maybe they're working on a jointly branded ARM SoC as a way for AMD to hedge against ARM.

I think it's more a hedge against Mali dying and/or the need for Graphics IP if Samsung decides/is forced to move to Risc-V because of the deal. Which is smart.
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
So 4 A76 cores + 4 A55 cores still cant beat a 6 year old 6700K quad core +HT native vs native? The result is not that impressive when you realise there is a 5 year gap there. Thats around the score of a 3200G.

The 4300U (4/4) gets around 15K and the 4500U (6/6) around 22K, and thats the low end, the 8CX is a premium chip as result the fair thing would be to compare it to a 4800U that gets well over 40k.

You might wan't to check the power the different SoCs your are quoting here are drawing while running this benchmark. The only competition in the 7W TDP range is last years Lakefield - but for obvious reasons Lakefield is nowhere to be seen in your stupid rant...

Besides i did not make the post to impress you, but give some idea of the relative emulation performance on ARM64.
 
Last edited:

Shivansps

Diamond Member
Sep 11, 2013
3,855
1,518
136
You might wan't to check the power the different SoCs your are quoting here are drawing while running this benchmark. The only competition in the 7W TDP range is last years Lakefield - but for obvious reasons Lakefield is nowhere to be seen in your stupid rant...

Dont try to move the goalposts here, you didnt mention anything about power efficiency, you did a comparison based on raw perf and i did another comparison based on raw perf using more modern cpus.

In fact this thread is about laptops, thats the market for Renoir-U 15W chips, i dont see whats the problem in bringing Renoir-U perf and there is no rant there either.

I more than welcome ARM CPUs in the laptop market, but you cant forget about pricing and the proper x86 counterparts, the 8CX is a ultra premium SKU that has lower emulation MIPS than the entry-level 4300U and lower native MIPS than the 4500U at 50% TDP, but 7W vs 15W is not that much of a issue for a laptop, it is for ultra-portables, like that Surface.
Offcourse ill like to see a proper comparison.
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
I think it's more a hedge against Mali dying and/or the need for Graphics IP if Samsung decides/is forced to move to Risc-V because of the deal. Which is smart.

Well you can use Mali with any other CPU architecture. It is not even unprecedented as Intel combined Mali and Airmont cores in their Sofia-3G SoC. In any case looks like Samsung is dropping Mali anyway going forward.
 

SarahKerrigan

Senior member
Oct 12, 2014
373
539
136
Well you can use Mali with any other CPU architecture. It is not even unprecedented as Intel combined Mali and Airmont cores in their Sofia-3G SoC. In any case looks like Samsung is dropping Mali anyway going forward.

It's been used with SH (ST Micro set-top box parts) and MIPS (Alchemy) too.
 

Shivansps

Diamond Member
Sep 11, 2013
3,855
1,518
136
As a tie-in to my other thread in the GPU section, what are the odds of a PC Gaming GPU based on scaled-up Mali IP? Any chance at all?

Mali GPUs right now at least, dosent support Feature Level 12, not sure if is a driver thing or a missing hardware feature, so Windows gaming on Mali should work around the same as Kepler does with DX12 games as issues are concerned.

Most of the issues with these gpu are due to not target Windows as a platform. Something i dont understand, Windows 10 can run on Raspberry Pi 4 whiout a working GPU driver... it would be a totally valid platform if the driver stack. were OK.
 
Last edited:

ThatBuzzkiller

Golden Member
Nov 14, 2014
1,120
260
136
Mali GPUs right now at least, dosent support Feature Level 12, not sure if is a driver thing or a missing hardware feature, so Windows gaming on Mali should work around the same as Kepler does with DX12 games as issues are concerned.

Most of the issues with these gpu are due to not target Windows as a platform. Something i dont understand, Windows 10 can run on Raspberry Pi 4 whiout a working GPU driver... it would be a totally valid platform if the driver stack. were OK.

The biggest problem with Mali GPUs isn't the features. The biggest problem with Mali GPUs is that they're a bad match for the D3D12 API in general like all the other tile based GPUs. Scary stuff happens in their drivers like emulating geometry shaders and enabling bounds checking is very expensive but this feature can't be turned off in D3D so performance will tank automatically over there ...
 

Shivansps

Diamond Member
Sep 11, 2013
3,855
1,518
136
Yeah ive been looking at gaming videos on the Surface Pro X and the Mali GPU is a complete trash, i dont even think it is FL 11.1 there is a shitload of games that dont run, have issues or just crash, not sure if the crashing is due to the GPU or unaligned memory access from the game itself, i saw that problem in two open source games that run on linux when i tryied to run then on the RPI (never tested on ARM64).
 
Last edited:

NTMBK

Lifer
Nov 14, 2011
10,240
5,027
136
Yeah ive been looking at gaming videos on the Surface Pro X and the Mali GPU is a complete trash, i dont even think it is FL 11.1 there is a shitload of games that dont run, have issues or just crash, not sure if the crashing is due to the GPU or unaligned memory access from the game itself, i saw that problem in two open source games that run on linux when i tryied to run then on the RPI (never tested on ARM64).

Surface Pro X isn't Mali. It's a Qualcomm Adreno GPU, completely different architecture.
 

Kryohi

Junior Member
Nov 12, 2019
16
17
81

I don't really get why Mediatek is involved here? NVidia can clearly produce their own ARM SoCs. And I'm not taking it seriously until it runs Windows.

BUT... I can definitely see a future where an ARM laptop chip can emulate x86 games faster than an x86 chip in the same power envelope. And at that point, this could be viable.

Can Nvidia make a SoC with an integrated 5G modem though?
For some reason Microsoft and Qualcomm are really pushing for 5G always-connected devices, so Nvidia and AMD may be disadvantaged in that regard?

Same reasoning for what Samsung will be allowed to do with RDNA: will AMD see a low-power, 5G enabled SoC as direct competition in the laptop market?
 

NTMBK

Lifer
Nov 14, 2011
10,240
5,027
136
Can Nvidia make a SoC with an integrated 5G modem though?
For some reason Microsoft and Qualcomm are really pushing for 5G always-connected devices, so Nvidia and AMD may be disadvantaged in that regard?

Same reasoning for what Samsung will be allowed to do with RDNA: will AMD see a low-power, 5G enabled SoC as direct competition in the laptop market?

I don't see the point of 5G in the laptop market. 99% of users will use the laptop on WiFi, and you can tether your laptop to your phone if you really, really need to use it without WiFi. It just doesn't make sense to pay hundreds of dollars a year for a 5G contract on a laptop.
 

Kryohi

Junior Member
Nov 12, 2019
16
17
81
I don't see the point of 5G in the laptop market. 99% of users will use the laptop on WiFi, and you can tether your laptop to your phone if you really, really need to use it without WiFi. It just doesn't make sense to pay hundreds of dollars a year for a 5G contract on a laptop.
I fully agree, but 5G and AI are buzzwords that probably increase sales
 
  • Like
Reactions: Tlh97 and NTMBK

DrMrLordX

Lifer
Apr 27, 2000
21,637
10,856
136
I would not be surprised if this collaboration has a long-term eye toward Windows, even though it's not being mentioned right now. Especially if the rumors about Samsung preparing to launch Exynos/Radeon Windows laptops are true.

Interestingly enough, Mediatek licensed Radeon GPU tech from AMD six years ago.
 
  • Wow
Reactions: Thibsie