Current Generation Hardware Speculation with a Technical Spin [post GDC 2020] [XBSX, PS5]

[BRiT]

Agreed. Keep retail costs low and maybe persuade consumers to buy more and swap them in and out like carts instead.

 

[Esrever]

* APU = $149
* 16GB GDDR6 = $80-$90
* 1TB NVME SSD = $55
* Motherboard = $35
* 250w PSU = $20
* Case = $20
* Misc = $110

BOM = $474-$484
Retail = $499 or $399 w/ $75-$85 BoM loss.

I think you are underestimating the memory and SSD costs but probably overestimating some of the others like the case. I think it will likely have a BOM slightly north of 500.

 

[RobertR1]

One thing for sure, MS has delivered and will be purely focused on core gamers. Seems like x360 gen part 2 coming up!

 

[RDGoodla]

Xsx ray tracing performance is equivalent to a 13 TF Xsx GPU (RDNA2). Is it possible to estimate the performance compared to RTX 2000 series?

 

[Shortbread]

I think you are underestimating the memory and SSD costs but probably overestimating some of the others like the case. I think it will likely have a BOM slightly north of 500.

Not really. I'm using our vendor system along with various volume / discount pricing on current available parts. Of course, MS is using custom parts, however, my BOM estimate for Series-X wasn't necessarily about pure accuracy, but possible BOM costs.

 

[RobertR1]

Xsx ray tracing performance is equivalent to a 13 TF Xsx GPU (RDNA2). Is it possible to estimate the performance compared to RTX 2000 series?

In the DF article, they more or less equated performed to a RTX 2080.

" We'll be covering more on this soon, but there was one startling takeaway - we were shown benchmark results that, on this two-week-old, unoptimised port, already deliver very, very similar performance to an RTX 2080."

If real, I'd expect 2nd gen games built from the ground up to be at a base 2080ti level.

 

[Deleted member 13524]

Have they commented on whether or not the BVH accelerators are in the TMUs like we've seen on AMD's patents?

I bet they'll have a proprietary solution just like MS and will use it to make back money offset by the loss on the console itself.

Sony hasn't used proprietary formats on their home consoles since the PS2 memory cards.
I think it'd be better if they used standard M.2 drives that passed a speed check. Sony could have e.g. an updateable list of M.2 drives (PCI IDs?) that the console would accept.

 

[Remij]

Have they commented on whether or not the BVH accelerators are in the TMUs like we've seen on AMD's patents?


Sony hasn't used proprietary formats on their home consoles since the PS2 memory cards.
I think it'd be better if they used standard M.2 drives that passed a speed check. Sony could have e.g. an updateable list of M.2 drives (PCI IDs?) that the console would accept.

I think that's asking for trouble tbh. The list also wouldn't be very big... considering Sony's solution is apparently more robust with higher bandwidth. (according to rumors)

But we'll see.

 

[Dictator]

In the DF article, they more or less equated performed to a RTX 2080.

" We'll be covering more on this soon, but there was one startling takeaway - we were shown benchmark results that, on this two-week-old, unoptimised port, already deliver very, very similar performance to an RTX 2080."

If real, I'd expect 2nd gen games built from the ground up to be at a base 2080ti level.

That is what we equivalate it to for rasterised performance there - Gears 5 does not use hardware RT in that demo. RT performance can be a whole other beast and should not be said to be equivalent and we purposefully did not say that for RT.

 

[Deleted member 13524]

I think that's asking for trouble tbh. The list also wouldn't be very big... considering Sony's solution is apparently more robust with higher bandwidth. (according to rumors)

But we'll see.

Isn't the expansion storage a secondary unit for storage? I thought the PS5's storage system is supposed to have several layers of storage speeds, from DDR4 through some very fast NVRAM and then the large slower storage SSD. That way, the last level of storage can afford to be slower.

Regardless, we should expect most PCIe 4.0 4x drives to quite faster. Faster than the 2.5GB/s on the SeriesX on peak bandwidth at least.

 

[tunafish]

Isn't the expansion storage a secondary unit for storage? I thought the PS5's storage system is supposed to have several layers of storage speeds, from DDR4 through some very fast NVRAM and then the large slower storage SSD.

No. This does not work, because very fast flash is very expensive. The way you make a cost-effective very fast drive is not by making it small, it's by using cheap, commodity flash of reasonable speed and making it big and wide. The cost of "some very fast NVRAM + a 1TB slower storage SSD" cannot be made less than just putting a good controller in front of 1TB of normal flash.

All the tiered storage bullshit is cooked up by people who do not understand the economics. There is simply no place for such a thing in a console.

 

[BRiT]

This was something I didn't catch earlier from skimming, but now taking the time to watch and pay attention to the DF video.

RPM FP16 is in Xbox Series X. It's guaranteed to be in Sony PS5 since 4Pro had it. It should be in PC RDNA2 graphic cards too, right?

 

[Kaotik]

This was something I didn't catch earlier from skimming, but now taking the time to watch and pay attention to the DF video.

RPM FP16 is in Xbox Series X. It's guaranteed to be in Sony PS5 since 4Pro had it. It should be in PC RDNA2 graphic cards too, right?

Of course, just like it's in all the different Vega models and RDNA1 models

 

[BRiT]

Of course, just like it's in all the different Vega models and RDNA1 models

Ooops. Forgot it was Vega/RDNA1 feature.

 

[tunafish]

Yeah, it seems like it's here to stay. I for one applaud this.

An interesting figure from the DF video that I didn't notice in the article: The raytracing hardware is capable of 380 billion intersection tests per second. Anyone know how this compares to nVidia rtx performance? I have no idea what the RX Ops they quote are, or how many intersections a typical ray needs.

 

[Deleted member 13524]

All the tiered storage bullshit is cooked up by people who do not understand the economics. There is simply no place for such a thing in a console.

Tiered storage already exists in common cheap SSDs without DRAM that use fast SLC for cache + slower MLC / TLC / QLC, so I don't know why you'd come to a conclusion like this.

Of course, just like it's in all the different Vega models and RDNA1 models

Does RDNA1 have quad rate INT8 / octo rate INT4 throughput?
I remember that being present in Vega 20, but I don't recall it being mentioned when Navi 10 launched.

 

[tunafish]

Tiered storage already exists in common cheap SSDs without DRAM that use fast SLC for cache + slower MLC / TLC / QLC, so I don't know why you'd come to a conclusion like this.

For writes, not reads. Show me a "common cheap SSD" with a SLC read cache.

 

[Jay]

Still surprised at no usb c.
Understand for BC, but going forward?
I still hoped would have both types

 

[anexanhume]

Still surprised at no usb c.
Understand for BC, but going forward?
I still hoped would have both types

Why is USB C important?

 

[Deleted member 90741]

Does RDNA1 have quad rate INT8 / octo rate INT4 throughput?
I remember that being present in Vega 20, but I don't recall it being mentioned when Navi 10 launched.

RDNA1 (Navi12) seems to have them

https://github.com/llvm/llvm-project/blob/master/llvm/lib/Target/AMDGPU/AMDGPU.td#L867

def FeatureISAVersion10_1_1 : FeatureSet<
!listconcat(FeatureGroup.GFX10_1_Bugs,
[FeatureGFX10,
FeatureLDSBankCount32,
FeatureDLInsts,
FeatureDot1Insts,
FeatureDot2Insts,
FeatureDot5Insts,
FeatureDot6Insts,

To see which kind of operands and matrices are supported for those you can check out the generated code from the compiler tests.
https://github.com/llvm/llvm-project/tree/master/llvm/test/CodeGen/AMDGPU