Couldn't you create machines for TRC testing that manipulated clocks in such a way as to identify these bugs?
Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]
- Status
Couldn't you create machines for TRC testing that manipulated clocks in such a way as to identify these bugs?
most likely bullshit but fun speculation to read none the less:
http://www.neogaf.com/forum/showpost.php?p=229799600&postcount=4593
http://www.neogaf.com/forum/showpost.php?p=229799600&postcount=4593
Most likely?
That was straight up bat sh|t crazy talk.
But then maybe it's me, I could only skim it, when I get chance.... I'll still not read it though haha
The bits I read sounded mad
Not really. Witcher 3 running at 4K60 PC Ultra settings - Google what PC config it takes to achieve this.
6TF GPU ain't it
Most outrageous - albeit interesting - claim is that Microsoft/AMD collaborated to come up with a (i'm paraphrasing) "unique communication method between CPU and GPU to reduce latency to almost zero"...and that Microsoft has patented it.
Also that Microsoft "funded" nearly 40% of Zen/Vega R&D costs just to be able to patent this tech...
Metal_Spirit
Regular
Hardly likely specially because of cost... AMD 8 Core SR5 CPUs are expected to start at 250 euros. A high price for the CPU alone!
Place a Vega GPU with 50 CU for 6.2 Tflops at 977 Mhz, and the price would explode.
Not impossible, but certainly not a cheap product!
Remember the S was launched in 2016. This hardware would be a new generation 1 year after.
Metal_Spirit
Regular
I would like a more substantial proof that this happened than just a claim of a user in the comments section of a website!
This one:
http://digiworthy.com/2016/09/11/amd-zen-custom-socs-project-scorpio/
I still think it could be possible that we see a $500 Scorpio...especially if Micrososft sees this as a "premium product" alongside Xbox One...not to mention price could drop over time as Xbox One fades out...
Personally I would be willing to pay an extra $100-150 over $400 if it means they went "all out" on this thing.
lol on the bright side, there's only < 4 months of this type of noise until we actually get something to talk about.
I know people think $500 is doa, but I don't think so.
For a totally new gen maybe, but x1s will still be current gen.
Early adopters, and hardcore people will buy it at that price, so I think it could be that price for a year before having to drop down.
But as you say, it would have to prove it self both hardware and software wise.
This is my my belief as well. The fact that this is part of the Xbox One product family...I don't think "driving the install base" is as necessary for Scorpio...at least not initially.
So essentially they can release a premium price Xbox that they can sell to a certain market segment at first and over time reduce the price and shift Scorpio into a more mainstream/baseline console.
A lot of Microsoft's language around Scorpio indicates this...
What has the PC CPU/GPU list price to do with the costs for a custom SoC I don't get. PC List price != cost.
When telling a joke, always save the punchline for the end:
"...AS MICROSOFT HAS FULLY PATENTED THIS NEW ARCHITECTURE AS XBOX SCORPIO CONSOLE SPECIFIC which means neither Pc nor PlayStation will be able to use and or develop graphics chipsets using this new architecture as it's fully custom and designed specifically for Xbox Scorpio."
Hurr hurr hurr.
Pricing for hardware on consoles is significantly lower in margins than pricing the CPU for direct to consumers; this is because console market is extremely price sensitive and all parties are aware of this when forming the contracts for the pricing of the silicon. So the offset is that their console customers pay the R&D costs (for their projects). I believe the R&D can be leveraged by AMD if they choose to. (ie 7790)
TLDR; you'd never compare the list price of a desktop class CPU to a console APU as the price differential would be enormous for 2 reasons: 1 the margin difference, 2 the performance points
You don't need special hardware to change code timings. We for example had a #define to enable random stalls on all synchronization primitives. Helped to find bugs. If you have a custom task/job scheduler, you can add randomization to task execution order in debug mode. You can also add critical section guards that break execution if threads are found executing the same piece of code (usually with additional checks such as same index/pointer). However in general case it is not possible to (efficiently) detect that the same data locations are not modified/read from multiple threads concurrently. You could almost do this with Intel TSX... but TSX can give false positives if transaction cache line is evicted because of false aliasing.
Breakdowns for the X1S were estimating $100 for the SoC shortly after launch. ~$100 for 240 mm^2 (accounting for probable yields).
Assuming no effect on yields (unlikely), an NX to Zen from Jaguar would be rather less than $100 worth of silicon. A good portion of Zen is interfaces, and the chip will also have other stuff (northbridge, PCI, SATA) that will already be needed for a console or not be needed anyway. A console might also use a smaller L3 cache than the the large 2 x 8MB arrangement on the server focused module designs.
Depending on how yields were affected, you might be looking at as little as a few tens of dollars on the cost. Which depending on how you look at it, is either a very small price to pay, or a huge and unjustifiable jump in BOM.
Perhaps Microsoft found that there weren't too many situations where the ESRAM was being used in a DRAM-unfriendly way, or they're counting on the larger number of channels and the depth of their buffers to massage it sufficiently. It would be a complex undertaking without some kind of reordering of commands or patch to dynamically make a particularly bad sequence into a not so bad one if the hardware isn't provisioned to do so.
If a resource were marked as read-only, perhaps an optimization at the system level would be to elide copy operations and just alias?
What interests me in the case of Scorpio is whether there was a strong distinction in Durango for accesses to the ESRAM from the CPU. There was a disparity, but what hoops the CPU had to go through wasn't made clear. It wouldn't matter for most software if for Scorpio the hardware-created distinction went away. However, when assumptions at a platform level about "this hardware surely can't access X in this way" become challenged, I am curious about buggy or perhaps malicious code having some fun.
One consideration is that even if the games are coded against race/timing conditions, the systems below them are also subject to them. That could be kernel functions, firmware, or at a hardware level. Assumptions the games don't make might be wrapped up in the functions they are assuming work just fine.
Perhaps one area that wouldn't be publicly discussed for the Pro is what system methods or functions are deprecated or de-prioritized in Pro mode. Those might be OS hooks or system blocks that were only validated for the BC mode. Boost mode might be the result of them finally being validated, but it could also put a layer of hacks/hangs in place for when they don't work.
Naughty Dog's job system mostly uses atomic counters, except for when it falls back to OS lock routines for when things get potentially too complex or the OS can do things like invert priorities. It's a leap of faith that system functions or the hardware they plug into behaves when the design changes.
An old example I can't find the link for was Id's experience developing a game on the PS3 (Rage?) that had HDD performance tank on specific console revisions.
For a more modern example of foundational assumptions breaking, we can look to the PS4 Pro apparently having problems withstanding the assault of performance bruisers like Stardew Valley...
http://www.playstationlifestyle.net...pring-says-stardew-valley-developer/#/slide/1
It may very well be coincidence that the supposed fix for that problem is in the firmware that introduces boost mode, although I would expect getting the PS4 Pro to not fail at being a PS4 Pro would come first before having the hybrid case handled.
- Status
