Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
AMD Ryzen R9 3950X Review: 16 Cores, 32 Threads of Power - MMORPG.com
System
Member UncommonPosts: 12,599
AMD Ryzen R9 3950X Review: 16 Cores, 32 Threads of Power - MMORPG.com
It’s been a long wait, but today is an exciting day for those of us here at MMORPG.com. Today, we get to bring you our review of the AMD Ryzen R9 3950X. Featuring an incredible 16-cores and 32-threads of processing power with a boost speed all the way up to 4.7GHz, this CPU has “prosumer” written all over it. Coming to market at a comparatively cheap $749, is this the right CPU for you? We have the answer and this is our review.
Comments
https://wccftech.com/amd-zen-4-5-nm-launching-2021/ AMD is already shipping 7nm parts that were designed last year. If this rumor is true, then AMD is already working on 5nm and next year we will see a dramatic increase in speed and decrease in power consumption.
------------
2024: 47 years on the Net.
There are so many things wrong with that article that the headline really should be "WCCFtech runs clickbait articles without caring whether they're true."
Still running 1080p here on 27" and my 970 still feels op af.
So you don't think AMD is working on 5nm Ryzen parts?
------------
2024: 47 years on the Net.
For starters, there's really no way that 5 nm would already have better yields than 7 nm in any meaningful sense. It's certainly plausible that testing is going well. Unless they're comparing 5 nm testing results to analogous testing results on 7 nm from two years ago, it's hard to imagine how such a comparison could even be meaningful. Yields on a tiny test die as a percentage of chips that worked but not necessarily at a good frequency or voltage is not at all similar to the percentage of much larger dies that meet your binning requirements in a commercial product. It sounds like someone is trying to compare yields on TSMC's ~18 mm^2 test part to something like AMD's 7 nm chips that are an order of magnitude larger and have to meet particular frequency and voltage requirements.
AMD has announced that Zen 3 parts will be on "7+ nm", or TSMC's 7 nm EUV process node. AMD's CEO has announced that we'll see Zen 3 this year. This hasn't officially been announced, but considering that they're still in the process of launching 7 nm parts, a launch on 7+ surely isn't imminent. I'd expect to see Zen 3 late this year, though closer to the middle of the year is also plausible. What's not plausible is that that will happen and then be followed by a Zen 4 launch on 5 nm in early 2021. Late 2021, sure, but the article explicitly says, "AMD's 5nm products will be landing in early 2021". I could believe a tape-out at that time, but not a launch.
The article says that the reason Nvidia hasn't moved to 7 nm yet is because of yield issues. That's very unlikely. Process nodes are chosen years in advance, long before you understand how good the yields are. Even the awful yields of Fermi didn't prevent Nvidia from launching products--and constantly telling anyone who would listen how awesome Fermi would be while it was delayed. What's far more likely is that 7 nm wasn't going to be ready by the time they wanted to launch Turing, and they didn't want to go through all of the expense to move Turing to 7 nm rather than skipping it and moving on to 7+ so that they could get GPUs out on 7+ sooner. It's not like Nvidia is in a terrible position with their GPUs today and desperately needs to catch up.
The article also pretty severely misinterprets TSMC's comments about clock speeds. For example, check here:
https://images.anandtech.com/doci/15219/wikichip_tsmc_logic_node_q2_2019.png
TSMC's official approximation of how 5 nm compares to 7 nm is that you'll be able to get either 15% more performance in the same power envelope or get the same performance as before while using 30% less power. Those numbers are geared toward situations where you're tightly constrained on power, and with the assumption that the same performance as before at 30% less power will be an interesting option for a lot of customers. That really isn't a good description of desktop CPUs that clock in the 4-5 GHz range.
How high a process node can reasonably clock varies tremendously with the design, and it's not at all the case that you could just move an arbitrary design built for 7 nm to 5 nm and get a 15% clock speed, which is what WCCFtech claims. That's why AMD's CPUs on 7 nm can clock above 4 GHz, and their GPUs generally don't reach 2 GHz apart from extreme overclocking. Or look at TSMC's guidance on the previous big jump: 7 nm would offer 40% more performance than 16 nm at the same power. For 4 W cell phone chips, that's a decent enough approximation. But we didn't see top end CPU clock speeds jump by 40% in the last two years, and it's not just because AMD used GlobalFoundries rather than TSMC for their Zen 1 and Zen 2 CPUs.