|
Mofabio posted:Such a dumb waste of energy, what does anybody actually need calculated that quickly. Weather. I can see some supercomputers putting on a squeeze like that but usually they make that up with core count.
|
# ? Jun 19, 2019 22:41 |
|
|
# ? Apr 24, 2024 04:33 |
|
Which is gonna need even more computing power for error-correction, because loving weather forecasting, telecom uber alles, right Ajit Pai?
|
# ? Jun 19, 2019 23:06 |
|
SwissArmyDruid posted:Which is gonna need even more computing power for error-correction, because loving weather forecasting, telecom uber alles, right Ajit Pai? Literal jamming of weather sensors to own the libs
|
# ? Jun 20, 2019 03:16 |
|
We’ve entered the twilight zone with Paul making GBS threads on Intel in the AMD thread. A sight to behold
|
# ? Jun 20, 2019 05:06 |
|
Sub Rosa posted:stockfish benchmark stuff Ah, cool, thanks! Done, for the machines I have now. I'll update when I get my hands on a 3900X: https://firepear.net/grid/ryzen3900/#stockfish-chess-benchmark Edit: this app exhibits some of the best SMT scaling I've ever seen, btw. Whatever it's doing, it's highly amenable to having extra integer and scheduling units available. Also, in doing these tests I discovered that my 2700's motherboard was clocking its memory at 2133MHz instead of 3000MHz. Fixing that resulted in a 5% performance gain. mdxi fucked around with this message at 05:34 on Jun 20, 2019 |
# ? Jun 20, 2019 05:29 |
|
SwissArmyDruid posted:Which is gonna need even more computing power for error-correction, because loving weather forecasting, telecom uber alles, right Ajit Pai? he's such a hollow mouthed piece of poo poo shill
|
# ? Jun 20, 2019 05:36 |
|
Dramicus posted:Do they really though? I don't think there are many people who are inherently loyal to Intel, not like the guys who doggedly stuck with bulldozer until Ryzen. Disregarding fanboys, the thing that matters is that the guys who regularly buy a new high-end K every generation just aren't going to buy new systems at all. Even if they *would* buy whichever has the best gaming performance, Zen 2 isn't a drastic improvement on a 9700k. The most profitable segment of the market for Intel will be a blank this year. All the people posting in the thread about buying new Ryzens to replace their old Bridge/Haswell/Broadwell systems are the money Intel is missing out on. Risky Bisquick posted:We’ve entered the twilight zone with Paul making GBS threads on Intel in the AMD thread. A sight to behold And he's suddenly got a goatee too, what's up with that?
|
# ? Jun 20, 2019 13:01 |
|
I've seen plenty of build help requests (less here, but here too sometimes) that say "Intel/Nvidia only please". They're not weird about it like the AMD people but the notion that real gamers use Intel is going to be difficult to dislodge, especially if the 9900k holds any ST advantage at all.
|
# ? Jun 20, 2019 13:53 |
|
I would expect Intel and Nvidia to still have some advantages from how games are optimised simply because developers know there's a much bigger install base for that hardware, regardless of whether Zen 2 is better on paper.
|
# ? Jun 20, 2019 13:58 |
|
ItBreathes posted:I've seen plenty of build help requests (less here, but here too sometimes) that say "Intel/Nvidia only please". They're not weird about it like the AMD people but the notion that real gamers use Intel is going to be difficult to dislodge, especially if the 9900k holds any ST advantage at all. It will also depend on whether people disable security fixes or not, and Intel is gaming release season by deliberately releasing the latest security fix after Zen2 releases. Khorne fucked around with this message at 14:12 on Jun 20, 2019 |
# ? Jun 20, 2019 14:06 |
|
I bought 32GB of 4000 MHz DDR4 RAM that's just been sitting in the box because my current motherboard wouldn't be able to do anything with it. I'm getting antsy on picking up an x570 Mobo and a 3900, but nobody seems to be taking pre-orders on either yet, despite them being some two and a half weeks away from release.
|
# ? Jun 20, 2019 15:53 |
|
Gamers are predominantly cheap. Sure there are those guys who make systems worth several thousand dollars, but the vast majority of gamers are in the i5/1060 category. If AMD can put out value at that price level, I'm sure people will flock to it. Word gets around quickly.
|
# ? Jun 20, 2019 17:28 |
|
I wonder what will happen when Intel gets its poo poo together properly, which seems inevitable.
|
# ? Jun 20, 2019 17:38 |
|
GRINDCORE MEGGIDO posted:I wonder what will happen when Intel gets its poo poo together properly, which seems inevitable. Their 10nm process is hosed and they are years away from their 7nm being viable. Their monolithic die architecture will make it increasingly harder and harder to get good yields with sufficient performance and thermals to compete against AMD dialing up the core count on low-end systems. Moving to a similar chiplet architecture on the current 14nm process would take at least 2 years, though maybe their 7nm process will be ready by then. It's going to be a while. My best guess is that they'll move to a similar chiplet architecture but with 8-core complexes instead of 4 in an attempt to edge AMD out on single thread performance.
|
# ? Jun 20, 2019 17:48 |
|
GRINDCORE MEGGIDO posted:I wonder what will happen when Intel gets its poo poo together properly, which seems inevitable. Foveros looks really good...
|
# ? Jun 20, 2019 17:58 |
|
iospace posted:Dude, I specced out an AMD based "gently caress all of you" computer for comparison, and it had a 2990WX, 2080Ti, 128 GB RAM, 2 TB M.2 SSD, and 4 TB HDD and it still came under 5 grand I did the same! The only wrinkle was that the Mac Pro maxes out at over a terabyte of RAM, which IDK maybe you can do on some EPYC board, but not on any of the TR boards I could find. The main use case for that much RAM that I know of is keeping a live DB in memory, but a vanishingly small number of people would need that in a workstation.
|
# ? Jun 20, 2019 17:59 |
|
GRINDCORE MEGGIDO posted:I wonder what will happen when Intel gets its poo poo together properly, which seems inevitable. I mean, sooner or later they will. They didn't hire Jim Keller to host their departmental tea parties. And they're one of the largest semiconductor companies on the planet, they're not going to just fold. There's nothing super special about Infinity Fabric, it's just the modern branding for AMD's front-side bus, and Intel has QPI. That's how Intel is doing those 56C Xeons, and how they used to do Pentium D in the past. If they start putting more QPI links on their consumer chips they could do threadripper-style NUMA designs pretty easily and there's no reason they couldn't start engineering for an IO die approach as well. They don't need to exactly replicate the CCX - their "CCX" could be one ringbus. Obviously if the link is going off-die it would be higher latency, but in principle it's not terribly dissimilar from the dual-ring Haswell-EX designs. They can also de-tangle their uarchs from their process nodes. Rocket Lake is supposedly a step in that direction, but if they intend to be on 14nm for another 2 years they really need to start porting Ice Lake CPU stuff back to 14nm as well. There needs to be a Cooper Lake for consumer platform. That would (supposedly) put them mildly ahead of Zen2 IPC, and perhaps on par with Zen3. If not they're probably at a ~15% IPC loss to AMD next year and that's simply not tenable especially as AMD pushes up the clocks on 7nm. As for "when", who knows. Zen has been public for 2.5 years now, Intel might have gotten wind of it ahead of time. Assuming a 5-year product cycle, that might put the response in a year or two. BK has been out for a year now, and I'd assume the new CEO has lit a fire under the rear end of manufacturing and architecture. They could respond with price cuts in the meantime, but they'd need to be really deep to stay competitive. AMD is offering 12C at $500, Intel would need to offer the upcoming Comet Lake 10C at like $400, $450 tops, to stay competitive. That doesn't fit neatly into the i7/i9 pricing scheme but roughly something like 10C = i7. I guess that leaves i3=6C12T and i5=8C16T. Sounds extreme but that's basically what AMD is offering and Intel either follows or gets left behind. Knowing Intel it'll be something dopey like the i7 is 10C but has hyperthreading disabled and then they screw up the rest of the lineup to follow (8C16T would beat 10C10T so it needs to be gone there too, etc etc). The only way I see the math working is if they just straight-up make the i7 10C20T, any other configuration they're just handing AMD this generation and probably next generation too. Paul MaudDib fucked around with this message at 01:34 on Jun 21, 2019 |
# ? Jun 20, 2019 18:26 |
|
ItBreathes posted:I've seen plenty of build help requests (less here, but here too sometimes) that say "Intel/Nvidia only please". One niche to keep in mind: people who want to run MacOS (or try it out, or at least have the option in their pocket) have had Intel drilled into their heads. You’ll know who they are now because they’ll say “Intel with AMD graphics please.”
|
# ? Jun 20, 2019 19:03 |
|
BangersInMyKnickers posted:Their 10nm process is hosed and they are years away from their 7nm being viable. Their monolithic die architecture will make it increasingly harder and harder to get good yields with sufficient performance and thermals to compete against AMD dialing up the core count on low-end systems. Moving to a similar chiplet architecture on the current 14nm process would take at least 2 years, though maybe their 7nm process will be ready by then. It's going to be a while. AMD is already on 8-core CCUs with Zen 2, so they will still be a step behind.
|
# ? Jun 20, 2019 19:28 |
|
SwissArmyDruid posted:AMD is already on 8-core CCUs with Zen 2. No, still 4-core complexes, 2 CCXs per chiplet.
|
# ? Jun 20, 2019 19:29 |
|
Wait, I thought that was the entire loving point of moving to 8-core dies, so that you could merge the L3 caches together for coherency! JFC, even when AMD is doing fine, they still find a way to disappoint.
|
# ? Jun 20, 2019 19:32 |
|
Craptacular! posted:One niche to keep in mind: people who want to run MacOS (or try it out, or at least have the option in their pocket) have had Intel drilled into their heads. You’ll know who they are now because they’ll say “Intel with AMD graphics please.” Fair, but I'm talking people building gaming systems. There are a nonzero number of people out there who, often for historically valid reason, won't considering building AMD.
|
# ? Jun 20, 2019 19:34 |
|
SwissArmyDruid posted:moving to 8-core dies What do you mean "moving"? Zen/Zen+ were an 8-core die too... two 4-core CCXs per die. Despite the CCX architecture, Zeppelin is a monolithic die. Only TR/Epyc were MCM products.
|
# ? Jun 20, 2019 19:36 |
|
SwissArmyDruid posted:Wait, I thought that was the entire loving point of moving to 8-core dies, so that you could merge the L3 caches together for coherency! JFC, even when AMD is doing fine, they still find a way to disappoint. I think Rome is for the Eypc chips but not the Zen2 Ryzen stuff. That was my impression anyhow.
|
# ? Jun 20, 2019 19:38 |
|
The 3700 (8c/16t) is 1 CPU+stuff die and 1 I/O die, right? Its just the 3900 and 3950 that use 2 CPU+stuff dies and an I/O die?
|
# ? Jun 20, 2019 20:09 |
|
What's the deal with Ryzen 3 APUs? All I could find was the 3400g, which despite the model number is 12nm. My company will be buying a batch of PCs soon and it would be great to get all dem sweet cores If there are no 7nm APUs coming within the next couple of months, what cheap options are there for adding basic (i.e. office level) graphics capability to an AMD system?
|
# ? Jun 20, 2019 21:16 |
|
There may be chiplet-based APUs coming at some point but Raven Ridge-style 7nm monolithic APUs probably won't be until next year and may be 4C again. The absolute cheapest display adapters are still Tesla-based (GT210/GT730/etc) or Terascale-based (Radeon 5450/6450). If you want something with actual modern driver support I'd look at a RX 460, GT 1030, RX 570, or GTX 1050. If you want lots and lots of displayport outputs, probably a Quadro P400 or P600. You can source the RX460 pretty cheap from sketchy ebay sellers (as low as $50) but both it and the 1030 have fairly limited display outputs. The other options are going to be a little more expensive, and by then you are starting to get up into the price of the P400 (DP can turn into HDMI 1.4 or DVI single-link with a passive adapter). The P400 is going to be easy to source from an actual hardware vendor like newegg or CDW and not sketchy ebay sellers so your purchasing dept may prefer that one. However it's the weakest processor as far as gaming goes (but this is office stuff). Paul MaudDib fucked around with this message at 21:35 on Jun 20, 2019 |
# ? Jun 20, 2019 21:22 |
|
ConanTheLibrarian posted:What's the deal with Ryzen 3 APUs? All I could find was the 3400g, which despite the model number is 12nm. My company will be buying a batch of PCs soon and it would be great to get all dem sweet cores The Ryzen vs. Zen naming is confusing enough, but the APUs add another layer of complexity. * There were no 1X00 APUs * The first generation Ryzen APUs were the 2X00 models, which came out after all the 1X00 products and before any other 2X00 series products * The 2X00 APUs were fabbed on a 14nm process, but with the architecture improvments that went into the 2X00 chips * It is expected that the 3X00 APUs (which are, remember, the 2nd gen Ryzen APUs) will follow this trend. They'll be on 12nm, but will include 3X00 arch improvements. And from what I've read, the expectation is that they'll be monolithic instead of chiplets. * Edit: And they are apparently NOT coming out before the rest of the 3X00 series. And then there's the Athlon 300 series, which has shown up in a laptop at Computex, but which AMD hasn't even mentioned. Who knows? mdxi fucked around with this message at 21:27 on Jun 20, 2019 |
# ? Jun 20, 2019 21:24 |
|
mdxi posted:* It is expected that the 3X00 APUs (which are, remember, the 2nd gen Ryzen APUs) will follow this trend. They'll be on 12nm, but will include 3X00 arch improvements. And from what I've read, the expectation is that they'll be monolithic instead of chiplets. Backporting arch improvements happens very rarely, I would expect the only major change for the 3000 series APUs to be upgrading the iGPU to Navi.
|
# ? Jun 20, 2019 21:51 |
|
Paul MaudDib posted:Knowing Intel it'll be something dopey like the i7 is 10C but has hyperthreading disabled and then they screw up the rest of the lineup to follow (8C16T would beat 10C10T so it needs to be gone there too, etc etc). The only way I see the math working is if they just straight-up make the i7 10C20T, any other configuration they're just handing AMD this generation and probably next generation too.
|
# ? Jun 20, 2019 21:58 |
|
Arzachel posted:Backporting arch improvements happens very rarely, I would expect the only major change for the 3000 series APUs to be upgrading the iGPU to Navi. Huh? They've already been announced to be 12 nm versions of the existing lineup (with Vega graphics) and are releasing on the 7th.
|
# ? Jun 20, 2019 22:11 |
|
Llamadeus posted:https://www.amd.com/en/products/apu/amd-ryzen-5-3400g Oh, then it's even more underwhelming
|
# ? Jun 20, 2019 22:12 |
|
mdxi posted:Ah, cool, thanks! Done, for the machines I have now. I'll update when I get my hands on a 3900X: https://firepear.net/grid/ryzen3900/#stockfish-chess-benchmark Thanks so much!
|
# ? Jun 20, 2019 22:31 |
|
Paul MaudDib posted:What do you mean "moving"? Zen/Zen+ were an 8-core die too... two 4-core CCXs per die. I dunno what the gently caress I was thinking, anymore. All of the hypothetical topology mockups have, predictably, gotten buried by reporting on the real things now that AMD's released their official information so I can't find the thing that made me think what I thought, but the crux of the matter is that I thought that in re-arranging the two CCXes from a side-by-side configuration to over/under, it meant that Zen 2 dies were going to be one 8-core CCX per die with a completely coherent L3 cache per-die, instead of two four-core CCXes with their separate L3s per die.
|
# ? Jun 21, 2019 06:32 |
|
Paul MaudDib posted:The absolute cheapest display adapters are still Tesla-based (GT210/GT730/etc) or Terascale-based (Radeon 5450/6450). If you want something with actual modern driver support I'd look at a RX 460, GT 1030, RX 570, or GTX 1050. If you want lots and lots of displayport outputs, probably a Quadro P400 or P600.
|
# ? Jun 21, 2019 07:44 |
Maybe they don’t think Ryzen APU’s are going to be very impressive regardless of if it’s Vega, Navi, or something actually good in there since they’re all saddled with DDR4 at the moment. We’re not going to be able to guess how much DDR5 will boost those APU’s(hopefully into decent 1080p 3D territory) at the moment, but maybe they’ve got some idea.
|
|
# ? Jun 21, 2019 10:06 |
|
If they're not busy working on 3700X with a GPU as the second chiplet, they're surely being crazy? There's got to be a reasonable market for people who want the high core count options for productivity reasons without the need for a dGPU. I know I'd love one to finally upgrade my linux box where a dGPU would just be a waste.
|
# ? Jun 21, 2019 12:15 |
|
It is really lovely that they made the embargo date the release date of the CPU. I know this is becoming more common but it still sucks.
|
# ? Jun 21, 2019 13:01 |
|
Epyc pricing leaked;
|
# ? Jun 21, 2019 13:05 |
|
|
# ? Apr 24, 2024 04:33 |
|
The Gunslinger posted:It is really lovely that they made the embargo date the release date of the CPU. I know this is becoming more common but it still sucks. I feel the same way but I'm convincing myself that they'll also be out of stock for months as a consolation.
|
# ? Jun 21, 2019 13:38 |