|
punk rebel ecks posted:I see. Is it likely the first half of 2022 or the latter half of 2022? I think that will determine whether or not I should wait. The leaked roadmap suggests that it is mid 2022, but it is probably too far out for even AMD to know for sure today. The original leak that lots of people have confirmed: https://videocardz.com/newz/amd-ryzen-2021-2022-roadmap-partially-leaks And here is a version that someone has expanded on with further leaks/confirmations (THIS IS NOT A FURTHER LEAK just a forum dude creation): The thing after Van Gough has been confirmed now to be "Dragons Crest"
|
# ? Oct 9, 2020 06:11 |
|
|
# ? May 16, 2024 18:27 |
|
apology for poor english when were you when Intel dies? i was sat at home watch Lisa in AMD Zen 3 launch and F5ing for Ampere GPU when Pjotr ring ‘Intel is kill’ ‘no’ You know we had a caesura and are entering a new era, when your gaming daddy chat group that shrugged off AMD because “Gaming with Intel is faster” and with top end Intel 9900 and 10900 CPUs instantly jumps on the 5950.
|
# ? Oct 9, 2020 06:12 |
|
Mr.PayDay posted:apology for poor english Seeing the CS:GO benchmark was the last straw for me, Intel lost the crown and it's nice to see AMD back on top because my first self-build PC was an Athlon XP back in the day.
|
# ? Oct 9, 2020 06:32 |
|
According to CNBC AMD is looking to acquire Xilinx for $30 billion. I guess they're no longer worried about their debt. Xilinx make FPGAs right? Why would AMD want to buy them?
|
# ? Oct 9, 2020 07:40 |
|
Curious to see what a 5700X looks like and costs. I'm still thinking I'll be riding my 3600X for a while, and seeing what the next AM4 spin is. With AMD starting to really trounce Intel though it does make me a bit worried about the price of CPUs jumping even further.
|
# ? Oct 9, 2020 07:53 |
|
gradenko_2000 posted:* if your motherboard supports "BIOS Flashback", then you can load the latest BIOS into a flash drive, plug the drive into the USB port designed for flashback, press a button, and the new BIOS will be flashed even if you don't have a working CPU in the socket. Thank you!
|
# ? Oct 9, 2020 08:06 |
|
hobbesmaster posted:I could absolutely be misreading the pcie bandwidth table. I think it's going to take PCIe5 and 6 before it's possible to have any serious number of NVME ports. At which point I wonder if there will be any interest in U.2 in the consumer space. It seems to me the M.2-on-the-motherboard form factor isn't particularly scalable. If mechanical drives eventually die out, do we get stuck with SATA drives for home-brew storage severs or will there be affordable U.2 solutions?
|
# ? Oct 9, 2020 08:48 |
|
I do wish some cheaper motherboards would eliminate the x1 slots altogether, consolidate those lanes and give me another x4 slot instead.
|
# ? Oct 9, 2020 09:03 |
|
Are there many boards with four x1 slots? They seem to have been more common with older X470 and B450 boards, there's only one X570 I can see. For a board with three, x2 and x1 might be more useful, or a single x2 vs 2 x1.
|
# ? Oct 9, 2020 09:57 |
|
Drakhoran posted:According to CNBC AMD is looking to acquire Xilinx for $30 billion. I guess they're no longer worried about their debt. I wonder if the Xilinx acquisition isn't related to improving AMD's SoC game. Between their APUs, semi-custom division, and console chips, there's a real need for SoC expertise at AMD, I think. I'm also wondering if this isn't also related to making better chips for the ultra-low-power segment, like that 5W Pollock chip that Microsoft was reportedly loving around with for a Surface Go 3. ...actually, now I wonder if their ongoing work with Pollock was why AMD put so much focus on improving IPC while keeping TDPs locked in place for Zen 3.
|
# ? Oct 9, 2020 10:03 |
|
I would need a B450 beta bios anyways so might as well wait for the inevitable 5700x. Maybe by then I'll be able to buy a 3070 or equivalent AMD card too!
|
# ? Oct 9, 2020 10:03 |
|
Good to see AMD buying Xilinx before Nvidia snaffle them up. There is a long history of collaboration with Xilinx, there was even a Xilinx FPGA that was socketed for the Opteron servers and could be flashed to accelerate workloads. I'm sure AMD has had them in their sights for a while and just been too broke to afford them.
|
# ? Oct 9, 2020 11:03 |
|
FuturePastNow posted:I do wish some cheaper motherboards would eliminate the x1 slots altogether, consolidate those lanes and give me another x4 slot instead. Ryzen has 24 PCIe lanes total. x16 GPU (or x8 x8), x4 nVME/M.2 SATA, x4 chipset. Absolutely everything else goes through that x4 chipset link. Sound, network, thermal monitoring, all IO except for 4 USB 3.1 lanes that are served by the ryzen itself. You don't gain anything by consolidating x1 into x4 because that's all ultra-lowspeed stuff. If you need anything other than a single GPU and nVME drive you have to to sacrifice lanes somewhere or move to threadripper/epyc/xeon.
|
# ? Oct 9, 2020 13:10 |
|
Drakhoran posted:According to CNBC AMD is looking to acquire Xilinx for $30 billion. I guess they're no longer worried about their debt. Considering how popular those classic mini consoles have been, maybe amd is looking to make an Athlon XP classic.
|
# ? Oct 9, 2020 13:56 |
|
Multiple 1x slots are also more useful in a lot of cases than an extra 4x, especially on cheaper boards. My current board was extremely expensive and I don't need extra add in cards, but for the five years before that I had three separate 1x cards for extra networking and high speed USB
|
# ? Oct 9, 2020 14:05 |
|
Current boards you get 2 m.2 slots and a spare 16x slot (may or may not have 16x lanes). If you need more than 3 NVMe drives you're in an unusual position and either need to buy a threadripper -- which can use the 16x -> 4*4x splitter cards -- or you just need to stop buying undersized nvme drives.
|
# ? Oct 9, 2020 14:19 |
|
MSI's Meg Unify has 3 M.2 slots right now at least.
|
# ? Oct 9, 2020 16:15 |
|
Drakhoran posted:According to CNBC AMD is looking to acquire Xilinx for $30 billion. I guess they're no longer worried about their debt. Intel bought Altera as a data center protection play I thought, probably the same reason for AMD
|
# ? Oct 9, 2020 16:24 |
|
So DDR5 and PCIe5 is slated 2022 only? I sure hope so, because I'm half-expecting EPYC and Threadripper to be DDR5 already (including yet another socket modification for the TR).
|
# ? Oct 9, 2020 16:24 |
|
Gwaihir posted:MSI's Meg Unify has 3 M.2 slots right now at least. The PCH only really has 4 lanes to the CPU so those cards can't both be full speed, right?
|
# ? Oct 9, 2020 16:37 |
|
Combat Pretzel posted:So DDR5 and PCIe5 is slated 2022 only? We don't know. A year or so ago it seemed confident that DDR5 platforms would be in 2021, but that was before a lot of poo poo happened.
|
# ? Oct 9, 2020 16:51 |
|
Combat Pretzel posted:So DDR5 and PCIe5 is slated 2022 only? Seems that way from the leaked AMD roadmaps, yeah. Intel doesn't seem likely to hit DDR5 before then, either. Milan is DDR4, which is slated to continue 'till at least mid 2021. Q4 2020 seems like the earliest Epyc/TR might move to a new socket and DDR5, with desktop boards trailing by another year-ish.
|
# ? Oct 9, 2020 16:54 |
|
Its obvious that there are FPGA users/experts based on posts upthread but yes the Xilinix acquisition makes a lot of sense, especially post-NVDA-Mellanox acquisition. The fact that AMD is getting design wins in CPUs for supercomputers doesn't really mean much when NVDA is supplying the accelerators and the fabric. Xilinix is also very cosy with the 5G/comms industry, for what that's worth. Bad for consumers, ultimately, imo, but AMD basically has to do this in order to try and make their own holistic compute solution to hold NVIDIA off, especially if they decide to (as rumored) make baby ARM chips that are glorified PCIe complexes to just be able to throw GPUs/floating point performance at problems, especially for AI/inference. Much the same as the ATI acquisition, this will either make or break them long-term. Let's hope its the right choice. Stock sure as poo poo didn't like it this morning though. That's a lot of debt for a company just emerging from some real dark times. Crunchy Black fucked around with this message at 16:59 on Oct 9, 2020 |
# ? Oct 9, 2020 16:56 |
|
Last I saw DDR5 DIMMs won't be generally available until Q3 2021 at the earliest anyway so you're not missing much if desktop board support doesn't show up until 1H 2022.
|
# ? Oct 9, 2020 16:57 |
|
New DDR specs are always expensive as gently caress when they come out it's probably going to take 1-2 years after that for supply to catch up and make them reasonable.
|
# ? Oct 9, 2020 17:03 |
|
The Gadfly posted:Ah I see. I haven't messed around with anything like this before, so thanks for correcting all the stuff I looked up just now now that I'm at a PC: this is what you want. If you can't live without the PCIe 4.0 capability it's going to be expensive for sure as that tech is just emerging onto the market, this is the first device with a PCIe 4.0 switch chip that I know of. One note is that there is bifurcation capability on some workstation/server-ish AM4 boards, like the Asrock Rack X570D4i, so in that situation you could use the cheap Asus Hyper card. That board has an IPMI controller so you don't need standalone graphics (but the IPMI also has very little graphics performance and won't do much beyond run a desktop). Or there are probably some ASUS workstation boards that support it too and you could run graphics on the chipset slots. Obviously doesn't do much for you if you've already bought the hardware and are trying to jam in more storage of course. Paul MaudDib fucked around with this message at 17:11 on Oct 9, 2020 |
# ? Oct 9, 2020 17:03 |
|
hobbesmaster posted:The PCH only really has 4 lanes to the CPU so those cards can't both be full speed, right? The x570 chipset can do 8GB/s so it wouldn't handle two pcie 4.0 SSDs running full speed simultaneously. There supposedly is a small latency for going through the chipset at all but from what I've heard it isn't noticeable. I was leaning towards this board and wasn't too worried as I was planning to put pcie3.0 drives in those chipset slots anyways.
|
# ? Oct 9, 2020 17:05 |
|
hobbesmaster posted:The PCH only really has 4 lanes to the CPU so those cards can't both be full speed, right? lanes are multiplexed in realtime, not at boot, so you could pull from either drive at full speed (with no restart) but not both at the same time, you'd functionally only get x2 speed from each drive if you were pulling from both drives at the same time. You are limited by the x4 link on the backhaul, you can't exceed that. that said, that really only applies to sequential transfers where the flash can produce enough bandwidth to saturate the link, random IO is limited by flash latency and the card will only be able to put out like 1 MB/s anyway, which the PCH can obviously trivially serve. In this case it could actually be faster if you have RAID1 set up so either drive can serve the requests, you would get twice the random performance. (Optane is much better but much more expensive per GB. An Optane 900P can do like 10 MB/s at 4K random QD=1.) Also, if you are just doing, let's say video editing, where the speed of one drive is sufficient and you just want more capacity, there's no problem with having two drives plugged into the PCH, you will still get functionally the same speed as a single drive (4.0x4 speed), just not the speed of two drives. Paul MaudDib fucked around with this message at 17:24 on Oct 9, 2020 |
# ? Oct 9, 2020 17:09 |
|
Yeah, and real time use case, I'm never ever trying to max out both at once. Nor would I really be able to tell if I was missing bandwidth outside of benchmarks.
|
# ? Oct 9, 2020 17:21 |
|
Makes sense especially with PCIe 3 drives, I guess I sort of implicitly thought if you had PCIE 3 kicking around it would all be PCIE 3 and thats clearly not the case.
|
# ? Oct 9, 2020 17:23 |
|
Crunchy Black posted:Bad for consumers, ultimately, imo, but AMD basically has to do this in order to try and make their own holistic compute solution to hold NVIDIA off, especially if they decide to (as rumored) make baby ARM chips that are glorified PCIe complexes to just be able to throw GPUs/floating point performance at problems, especially for AI/inference. I believe Mellanox already has something like this for NVMeOF
|
# ? Oct 9, 2020 17:28 |
|
pixaal posted:New DDR specs are always expensive as gently caress when they come out it's probably going to take 1-2 years after that for supply to catch up and make them reasonable. It's basically this that's convincing me to upgrade from my 4770K to a 5900/5950X this year. Should be plenty until 2024 when we're all dead.
|
# ? Oct 9, 2020 17:35 |
|
WhyteRyce posted:I believe Mellanox already has something like this for NVMeOF That is correct, they even announced it before the acquisition.
|
# ? Oct 9, 2020 17:37 |
|
The extra lanes presumably help when you want to do stuff like use old PCIe2 x8 device*. It's never going to saturate the 4x4 bridge but it will be hobbled if it's forced to run in a x1 slot downgraded to gen2 speeds. * Cheap second hand 10GbE solutions...
|
# ? Oct 9, 2020 17:39 |
|
Drakhoran posted:According to CNBC AMD is looking to acquire Xilinx for $30 billion. I guess they're no longer worried about their debt. A lot of reasons. Xilinx make FPGAs, NICs (with solarflare), AI accelerators and have huge market share in automotive and RF The wombo combo of fpga + server CPU is why Intel bought Altera but Intel was lead by morons and couldn't make it work AMD has good leadership but tbh the main issue is software, not hardware. They are going to finance this using stock not debt since every dumbass like me will buy AMD stock tho in my defense my current gains sit around 1000%
|
# ? Oct 9, 2020 17:50 |
|
Malcolm XML posted:They are going to finance this using stock not debt since every dumbass like me will buy AMD stock tho in my defense my current gains sit around 1000% Happy to report that my buying AMD stock has more than offset my buying Intel stock at roughly the same time.
|
# ? Oct 9, 2020 17:56 |
|
Crunchy Black posted:Its obvious that there are FPGA users/experts based on posts upthread but yes the Xilinix acquisition makes a lot of sense, especially post-NVDA-Mellanox acquisition. The fact that AMD is getting design wins in CPUs for supercomputers doesn't really mean much when NVDA is supplying the accelerators and the fabric. I think AMD is smart enough to acquire them and leave them the gently caress alone, for datacenter plays, etc, as you said. x86 is not appropriate to combine with FPGAs onto the same die / package, IMO (every Altera/Intel attempt at this was garbage IMO, from Stellarton to whatever Arrias they tried it on) and Xilinx has a great thing going with Zynq. Maybe chiplets / MCMs would change that up a bit, but US+ is a great product IMO and x86 can't play in that arena right now given the power / thermal constraints. Xilinx has no effective competition in certain segments of the FPGA market. Xilinx is one of the leaders in SerDes / transceiver IP as well, and they've got some interesting patents stacking up from the RFSoC side of things. The thing that makes me curious is when any of the big big boys will eat someone with a healthy PCIe switch portfolio. PLX got eaten up into Broadcom / Avago, Microchip has introduced some switches... I remember using PEX8500, 8600 and 8700 series PCIe switches and there was nothing better around at the time.
|
# ? Oct 9, 2020 17:57 |
|
Hrm, I still haven't opened the 3600 I bought on sale earlier this year. Glad I didn't open it, should make it easier to sell it.
|
# ? Oct 9, 2020 18:07 |
|
Malcolm XML posted:A lot of reasons. I was thinking that when AMD finally delivers the knockout punch with Zen 3, Intel's stock would tank a lot more. Now we know that Intel is a company with a lot of money, assets and resources so while it seems like doom and gloom for them now it's very much possible someone at Intel has decided to skip ahead to whatever the new die size is and if you buy intel now and quietly park your money you will get massive payouts later on when they catchup and beat AMD again. This wasn't the first time so I don't think Intel is out of the fight. Could be a great stock purchase, especially if they tank some more.
|
# ? Oct 9, 2020 18:10 |
|
|
# ? May 16, 2024 18:27 |
|
Kraftwerk posted:I was thinking that when AMD finally delivers the knockout punch with Zen 3, Intel's stock would tank a lot more. Intel's stock getting crushed in July was 100% due to their own admission that 7nm would be delayed, and has shown basically no movement due to any AMD announcements before or since then. Which...kinda makes sense. Despite trailing TSMC in terms of nodes, Intel's revenue has steadily increased over the last 4 years, and they've been selling chips almost as fast as they can produce them. AMD's been making big inroads with DIY enthusiasts, but that's actually not a particularly large market segment. From a pure business standpoint, Intel has been doing just fine. Intel still has enormous support in the datacenter due to their ecosystem, and it's still hard to find non-poo poo-tier pre-built PCs at like BestBuy or whatever, and AMD is still struggling to work their way into laptops in a meaningful way. I mean, don't get me wrong, Intel is fighting a losing battle right now, but it's a battle they're losing slowly enough that they've got plenty of opportunity to un-gently caress themselves.
|
# ? Oct 9, 2020 18:22 |