- Joined
- Nov 15, 2021
He flooded Cadence with chinks, and now their software is garbage. But at least they make a lot of money, I guess.
GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.
- Thread starter Smaug's Smokey Hole
- Start date
-
🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
- Joined
- Aug 28, 2019
Second part of the video elaborates on the two "LP" cores that may come to Zen 6 desktop CPUs and various APUs. This internal diagram shows that AMD is targeting around 1 Watt power consumption (or less?) for each LP core (it's undecided whether these would be based on Zen 5 or Zen 6 for Olympic Ridge desktop CPUs):
The cores do have SMT, so you're getting 4 threads, but probably not AVX-512. A simulation of Zen 5 LP put it at "65-75% the IPC of Zen 5 classic, and 50-60% the frequency". But that was with both on 4nm, whereas TSMC N3P is now expected for the I/O die. And it could end up being Zen 6 LP on desktop for a modest IPC gain. I think the reduced IPC is reflecting the difference in L3 cache.
I initially assumed they wouldn't even have L3 cache, like Intel's LP E-cores. But now that the LP cores may have been identified on this Medusa Point render, it looks like the two cores are sharing some cache between them. It's around 3-4 MiB by my guess, compared to 48 MiB for the 12-core chiplets. The cores look like less than half the size of the standard Zen 6 cores. Pixel counting the image may be worthless if dimensions have been altered to protect the source(s).
Hopefully, scheduling issues are minimal if these are simply treated as the slowest cores among the up to 26 cores available. I think the real scheduling issue for AMD has been their mixing of 3D V-Cache on one chiplet and a higher clocking chiplet with no 3D V-Cache.
- Joined
- Jul 18, 2019
Not to go too off topic but one of my biggest takeaways of working with non-American/Euro developers at [day job] is how much they care about v e l o c i t y over delivering actually good software. I've had to ship things with terrible fail states baked into the architecture simply because the boss nigger said we need it out for Q2. And all the foreigners on the project get big sloppies from management because "look at how fast they delivered X feature!"
One of AMD's biggest shots at Intel was over 'economy cores' and them not supporting the full instruction set. Foregoing AVX-512 on their LP cores will be such a bad look imo.
- Joined
- Aug 28, 2019
G.Skill Rolls Out 128 GB Capacity DDR5 Kit, Trident Z5 Royal With 64 GB Per DIMM & 8000 MT/s Speeds
The first 64 GB DDR5 UDIMMs are finally appearing.
The first 64 GB DDR5 UDIMMs are finally appearing.
We'll see. I'm not convinced the purpose of LP cores is to boost multi-threading, although I guess two of what's described could by 4-10% (24/10 cores). I think they're intended to cut power consumption for PCs that are idling 90% of the time, watching videos, etc. 2-core Zen 5 LP might be faster than 4-core N100.
- Joined
- Dec 19, 2022
And Intel griped over how AMD’s chiplets were just crudely glueing dies together. It’s just marketing, both sides immediately set about studying their rival’s innovations.
Unfortunately it’s a CUDIMM, meaning zen 4 won’t boot with it at all and zen 5 can only use it with the clock turned off, ie at low speeds.
- Joined
- Aug 28, 2019
I don't see any mention of it being CUDIMM. DDR5 memory was able to hit that kind of speed in tests long before we heard about CUDIMM. And it was demonstrated with a 9950X:
- Joined
- Dec 19, 2022
At 4000MT/s using a zen 5, which are able to boot with CUDIMMs (at low speeds, which 4000MT/s or 2000MHz certainly is).
- Joined
- Aug 28, 2019
That's clearly the clock rate before being multiplied by 2, aka the secret behind the famed "Double Data Rate" memory. You can confirm this by running CPU-Z yourself.
- Joined
- Dec 19, 2022
My bad, I was looking at the uncore frequency as actual frequency and saw the 4000 as MT/s.
Never used CPU-Z, I’m on Linux.
- Joined
- Nov 15, 2021
You can pretty easily run into signal integrity issues with memory running over 3 GHz. Don't buy this 4 GHz memory if you're putting it on a cheap board.
Yeah, and they think QA is a waste of time because it slows down feature delivery. AFAIK, Cadence has no formal QA groups.
AVX-512 is such a marginal feature with so few real-world use cases outside HPC, where GPUs are taking over anyway. Linus T's rant about it was partially right. SIMD in general has been a mixed bag.
Yeah, and they think QA is a waste of time because it slows down feature delivery. AFAIK, Cadence has no formal QA groups.
AVX-512 is such a marginal feature with so few real-world use cases outside HPC, where GPUs are taking over anyway. Linus T's rant about it was partially right. SIMD in general has been a mixed bag.
- Joined
- Mar 30, 2023
Is a 13" Lenovo Yoga with a Snapdragon 850 cpu a good laptop if I only want to install Debian/Ubuntu (Linux Mint doesn't do arm) and use it occasionally to browse the internet or watch shows without ha ing to worry about battery life?
- Joined
- Jul 18, 2019
Snapdragon 850 is quite old at this point and not particularly power efficient. Honestly if you want an ARM laptop for fucking around on, see if you can find a used M1 Macbook Air.
- Joined
- Sep 5, 2019
Why the fuck would I care about tariffs when I can't even get an MSRP 9070? Seriously, where the fuck is the refresh?
- Joined
- Aug 28, 2019
With tariffs, the cards will probably be above MSRP even if a plentiful supply arrives on your shores. Instead of high demand/low supply driving up the price, it would be the tariff. If tariffs are not actually affecting the components, middlemen might charge more anyway to take advantage of the panic. But the 9070 cards look out of stock or above MSRP from shady sellers all over the place, so you'll get to experience both at the same time.
AMD accumulated 9070 series cards at retailers/warehouses for almost 2 months instead of launching them in January, and claimed that first week sales were "10x more than previous generations". So any resupply is going to feel slow, way less than the initial rush of units sold.
If your GPU breaks, I recommend touching grass, or playing games that run on integrated graphics.
China’s General Purpose Media Interface “GPMI” set to deliver up to 192Gbps of bandwidth and 480W power through a single USB cable
Last edited:
- Joined
- Sep 10, 2020
My brother and I installed my 5950x cpu and the asus 9070XT gpu, and while I'm noticing more frames, the cabling on the GPU prevents me putting the glass back onto the case and the noise is pretty noticeable, tho I use headphones so once I'm gaming that isn't such a factor. However, the cpu seems to want to hang around 90C when I'm playing something, and sometimes when I'm not running games it'll slide back and forth from 40C to 84C. I've ordered a new cooler just in case.
- Joined
- Dec 17, 2019
NVIDIA PhysX and Flow Made Fully Open-Source
Nvidia's Github announcementJust fix it yourself faggots
- Joined
- Apr 11, 2023
- Joined
- Aug 28, 2019
What's the 16-core CPU for? Maybe a 5700X3D would've been more your speed...
(early Brave result for "5700X3D") https://old.reddit.com/r/buildapc/comments/1j3e6wg/5700x3d_9070xt_smart_upgrade_or_waste_of_money/
- Joined
- Dec 12, 2022
More cores are better lol. I can say that with certainty with my 12900k. Infinite chrome (well Brave) tabs can be open lmao (assuming you weren't stingy on RAM). That and it keeps pace with the newer generations.
- Joined
- Sep 10, 2020
It was on sale for more than half off and seemed to have a great performance increase to the 5600x.
Interesting link, ty