|
It's not that surprising that AMD has trouble building a compiler that outperforms LLVM; LLVM has decent code generation, Ryzen's scheduling requirements can't be that different from Core's or they'd suck wind on existing software, and AMD doesn't afaik have a serious compiler team. What's surprising is that they'd build a wet fart of a compiler and then actually release it. Why bother? How do they gain?
|
# ¿ May 20, 2017 21:59 |
|
|
# ¿ Apr 25, 2024 20:03 |
|
Boiled Water posted:Weird strategy to release the mid tier first. Works for NVIDIA with GPUs.
|
# ¿ May 20, 2017 22:18 |
|
Boiled Water posted:Ah yes the mid tier card the Titan Z, X, XP. They release their middle cards first, then later release the x50, Ti, Titan.
|
# ¿ May 22, 2017 19:38 |
|
So if AMD can only build good architectures when Jim Keller is around, what's their plan for after Zen?
|
# ¿ May 26, 2017 23:45 |
|
FaustianQ posted:It's two months, I highly doubt drivers aren't "ready" at this point. I was getting drops of drivers with "critical fixes" (including follow-up email to make sure we'd deployed it to all the test/demo machines) 3 weeks before Fiji street date. They also likely locked the demo config a fair while before the demo date. Nobody gets to check in anything meaningful to the demo branch the week before Lisa goes on stage. Subjunctive fucked around with this message at 20:51 on May 31, 2017 |
# ¿ May 31, 2017 20:45 |
|
FaustianQ posted:So drivers used in demo like this aren't even likely good, rather "stable". Well, they believe they show off the product well enough to put the CEO in front of it, so that tells you something. They're very likely not the drivers that will be available on day 1 of the glorious Vega future. Whether they are now ready, or will be then, or will be ever, is more philosophical than technical.
|
# ¿ May 31, 2017 21:47 |
|
wargames posted:This was the hope with zen smaller dies on an interposer for mondo yields. I have an ex-AMD friend who says that yield-friendly design was one of the priorities for Zen. Grain of salt, but he's damned smart and worked on Zen.
|
# ¿ Jun 2, 2017 01:31 |
|
ufarn posted:Ah, that's great to hear. Puts the ball in devs' court. That's a lovely place for a ball to be on this stuff.
|
# ¿ Jun 5, 2017 00:48 |
|
I have literally owned an entire (terrible) laptop smaller than that.
|
# ¿ Jun 6, 2017 17:26 |
|
The other thing is that people infer quality from price. If something is 10% cheaper it's a good deal. If it's 30% cheaper it's a lower-tier product.
|
# ¿ Jun 6, 2017 23:30 |
|
Intel needs to be careful to not piss off the customers who've been buying at $2K per. "You could have sold this to me at $750? WTF?" is a tough sales context for future conversations.
|
# ¿ Jun 8, 2017 16:50 |
|
Paul MaudDib posted:Oh yeah, on this note, the DOE just awarded a $258 million contract to AMD, Cray, HPE, IBM, Intel and Nvidia to build the first exascale computer with a target date of 2021 and a second computer online in 2023. Dunno if they're going with AMD's design or what, that's quite a mix of companies. Presumably AMD is doing the iGP.
|
# ¿ Jun 15, 2017 21:46 |
|
drat.
|
# ¿ Jun 20, 2017 21:59 |
|
Paul MaudDib posted:Yeah, ICC is good at extracting performance from Intel chips. But that's real performance that you would have in real-world usage. It's AMD's responsibility to push patches into LLVM or GCC or write their own compiler if they think they're leaving performance on the table. Didn't AMD release a compiler that sucked?
|
# ¿ Jun 21, 2017 03:39 |
|
Kazinsal posted:It was basically just LLVM with some patches. LLVM with some patches could probably be pretty good, given the right patches, but no.
|
# ¿ Jun 21, 2017 13:29 |
|
At unspecified scale, it seems so.
|
# ¿ Jun 21, 2017 19:01 |
|
Risky Bisquick posted:This is not true in all cases. Sometimes the OS can be totally frozen and the keyboard lights will cycle correctly. No, because the OS signals the light change. If no part of the OS code is executing, the lights don't change. Sometimes big pieces can be locked, though, and have that portion still function.
|
# ¿ Jul 4, 2017 16:42 |
|
Risky Bisquick posted:mouse frozen, keyboard lights work, os doesn't respond. Machine is frozen. You'll need to explain to me what is triggering the light change then, if not the handler in the OS that's executing on time.
|
# ¿ Jul 4, 2017 16:50 |
|
Munkeymon posted:Holy crap - I'd have given up and bought spray paint after a week or so Velvet curtains, gold rod, maybe a little valance.
|
# ¿ Jul 5, 2017 17:43 |
|
priznat posted:I hear wallpaper is back in. That could be fabulous.
|
# ¿ Jul 5, 2017 17:53 |
|
wargames posted:It doesn't need to, remember AM4 is suppose to be supported for 5 years. What does "supported" mean? Warranty? BIOS updates? Some of their chips will use it? All of their chips will use it?
|
# ¿ Jul 10, 2017 18:16 |
|
Measly Twerp posted:Feels good. What are you running?
|
# ¿ Jul 14, 2017 21:14 |
|
Yeah, OK.
|
# ¿ Jul 14, 2017 21:24 |
|
AMD needs to acquire Tesla to get Keller back and land a design win for their AI ambitions. Leverage, baby.
|
# ¿ Jul 24, 2017 23:26 |
|
Rastor posted:Jim Keller is a ronin, when AMD needs him again he will be there. From mutual friends, I've heard that Tesla is paying him much, much more than AMD or Apple did. If they want a solid Zen 3/4, the last 15 years' history suggests that they need him back ASAP.
|
# ¿ Jul 25, 2017 01:43 |
|
Arivia posted:Has it always been platfrom discussion in the title? Relatively recent I think.
|
# ¿ Jul 25, 2017 23:27 |
|
Combat Pretzel posted:"PCIe" is a registered trademark? wat? http://pcisig.com/sites/default/files/newsroom_attachments/Trademark_and_Logo_Usage_Guidelines_updated_112206.pdf Pretty much all standards are trademarked. That's how they control compatibility claims.
|
# ¿ Jul 27, 2017 22:21 |
|
How is it even possible that it won't be streamed? What are they thinking?
|
# ¿ Jul 29, 2017 19:20 |
|
It might be capped at 144, so Intel *could* have a big lead. That said,
|
# ¿ Aug 7, 2017 17:18 |
|
Kernel work is much less tightly coupled than game engine work.
|
# ¿ Aug 9, 2017 20:45 |
|
Consistent (but not universal) with my small number of observations, which is surprising because you'd think that low pay and lovely working conditions would attract the best.
|
# ¿ Aug 9, 2017 20:52 |
|
Why does that graph show values for prices less than the MSRP of the cards?
|
# ¿ Aug 15, 2017 02:25 |
|
Don't torque-shame.
|
# ¿ Aug 16, 2017 17:18 |
|
The Chromium (really Blink) embedding APIs are much richer than those for Edge, which are basically enough for a help viewer and not an application. There's a reason nobody has done the work to rehost Electron on top of Edge, and it's not because nobody has whined about using 0.2% of the disk. I expect that the Westminster-hosted app will be totally new code, and much less capable. (Like all WinRT ports of popular applications.)
|
# ¿ Aug 22, 2017 14:54 |
|
WinRT apps based on HTML/JS have much reduced capability compared to non-RT apps based on Blink. Just as with web-tech apps trying to build on top of Edge.
|
# ¿ Aug 22, 2017 15:56 |
|
Scheduling in the delay slots of PA-RISC was annoying, at least in assembly.
|
# ¿ Aug 27, 2017 16:37 |
|
I remember Intel sending dev support people to my work in the late 90s to tell us to change x<<4 to x*=16 all over our code so it would run faster, because of some architectural change in P4. Ridiculous in several ways, but that's Netburst for you.
|
# ¿ Aug 29, 2017 23:52 |
|
Combat Pretzel posted:IPv6 addresses have 16 bytes / 8 half-words (32-bit) though. 128 bits.
|
# ¿ Aug 30, 2017 12:42 |
|
Combat Pretzel posted:That's what I said. The 32-bit was to indicate half of which word width. Ah, ok. Sorry.
|
# ¿ Aug 30, 2017 16:00 |
|
|
# ¿ Apr 25, 2024 20:03 |
|
Looks like 8700K undercuts on price ($40 CAD cheaper with it at launch pricing), and is only 13% slower on perfectly parallel cinebench while 21% faster on single-threaded. Apparently someone has a 5.2GHz OC on a delidded 8700K, which should be another 20% (minus what Ryzen can get from OC, but that’s not much by my understanding). Maybe that was an early lottery win, I guess. Even if that’s a wash, is the Ryzen honeymoon over? Maybe in mid-spectrum with the 1600X they’re still solid? It’s a much stronger performance from Intel than I expected, tbh. https://www.extremetech.com/gaming/257006-intels-core-i7-8700k-reviewed-six-core-chip-punches-far-weight-class E: actually, looks like the 8700K OC situation might be overstated above. Subjunctive fucked around with this message at 18:47 on Oct 5, 2017 |
# ¿ Oct 5, 2017 18:30 |