https://www.tomshardware.com/news/amd-ryzen-3000-series-matisse-specs,38310.htmlhttps://www.tomshardware.com/news/amd-third-gen-threadripper-roadmap,39254.htmlhttps://www.tomshardware.com/reviews/cpu-buying-guide,5643.htmlhttps://www.tomshardware.com/news/dell-amd-epyc-rome-server,39213.htmlSomething's up with AMD's latest just released roadmaps --- changes are showing up that are possibly being driven by TSMC's recent 6nm announcements and/or some new hard yield data on AMD 7nm chiplet production that are showing up some new post production trends to AMD management.
AMD could also be reacting to a general market shift that may actually be going on --- simply put AMD doesn't really need to prime the pump any further, but instead AMD may need to implement what they are already committed to do, then get a firm grip on the saddle horn (holding on real tight) and then simply ride the tiger as it explodes underneath them ......
Lisa Su and her opening remarks on the 27th of this month will tell us which functional pathway AMD is gonna be taking, which could be in a range from "sandbagging again" to "slowly implementing" to "blowing our minds with unexpected rapid progress" all over again.
Tom's Hardware states:Threadripper, Ryzen, and Epyc Rome all use the same 7nm compute chiplets, meaning the use of these chiplets will compete with each other for the best wafers and for the best sorted chiplets, of which there likely aren't that many totally 100% flawless all top of the line wafers jest slam full of first class chiplets -- simply because the 7nm node is still quite new. Testing and sorting of the chiplets are all part of this equation as AMD needs to correctly state mhz and boost speeds on each type of chiplets that goes into their various chipsets.
The newer the node, the more defects, meaning more low-quality or defective chiplets. Rome will, of course, get the best of the bunch, while Threadripper will need to get the chiplets that clock high at decent voltages, leaving Ryzen with the worst chiplets. However, as Threadripper is a niche product and chiplet supply could be tight right at first so there might not be any room left for Threadripper to snag some of the best chips that would make more money if AMD sold them as server CPUs.
Motherboards are also a complicating factor. TR4, the Threadripper socket, hasn't seen an update since 2017 and still uses the X399 chipset. Granted, X399 boards can be quite good, but AMD will likely want to update it for Threadripper 3000. While we’ve seen leaks and news about 500-series boards for Ryzen, there hasn't been any news of a new chipset for Threadripper, making it unlikely the new Threadripper chips are on the near horizon. We may just need to wait for the next half-year horizon to roll around for Threadripper 3000.OK, TSMC has just now stated that 6nm lithography is now out there now saying that all 7nm designs from all of their customers can now be run on 6nm lithography at no real added cost and that all the development tools are exact the same as 7nm+ and 7nm++ and 7nm+++ so no design work is wasted in the 6nm conversion.
So, to TSMC 6nm is going to be the long lasting, large, persistent lithography node, not 7nm+++ --- so you can see that TSMC really wants to collect and accumulate all of their various 7nm customer's past ordered stuff and to run them in the future under the 6nm banner for ease of scheduling. This will happen soon enough anyway due to pricing, but not overnight.
This is all being put out there now, brand new, like a month before the 2019 Computex show's go date and all the big AMD announcements that are already planned to be made by Lisa Su at Computex.
It is obvious that the TSMC 6nm change over has to be addressed at the Computex roll out by Lisa Su on the 27th of this month in some fashion as it affects all the previously laid out plans and road maps.However, getting better isn't a bad thing. It just requires Lisa Su to tee it up correctly and to describe it in a way that does not confuse the computer press and the AMD fanboy customers.
Tom's Hardware is also stressing an interesting point about Intel and their "always claimed higher clock speeds". Intel only makes their exaggerated speed claims off of the best ONE (1) of their 4 or 6 or 8 cores and then by their own published sorting standard allows the rest of the compute cores to fall within a broader (lower) stated range of performance. It has always been that way, this trick has never been a secret with Intel (they just never talked much about it).
AMD however makes their performance claims more accurately and AMD covers the entire processor and that carefully includes
all the enclosed compute chiplets that make up the AMD processors.
Both Intel and AMD have communications and power bridges in their products (the AMD 14nm center core is a good example of this) --- communications and power bridges that intentionally run at larger lithography as they are pushing the CPU power requirements down these same traces at the same time as they handle inner chip communications between the center communications core and the compute chiplets/cores.
So, saying you are rating the compute cores/chiplets is somewhat more accurate, but the entire processor also includes AI blocks and Graphics blocks and some other items that are not strictly part of the compute rating per se. Saying you are judging the entire CPU is also a more accurate way to say things. But, note please, Intel doesn't do this at all ....... they just pay attention to the very fastest single core that they find inside a given processor.
Given the additional 15% compute speed and 20% energy performance boost the 6nm change over may eventually bring to the AMD party, Intel is likely going to start to choke a bit over their cute little "only 1 core is the fast one" marketing tricks. Tom's Hardware is now carefully keeping independent real world performance / price ranking data on all Intel and AMD processors performance-wise using a range of standard benchmark tests on a sample set of processors to rank them correctly against each other.
We now note that Tom's is already reality ranking some Ryzen classes of much cheaper AMD processors as being functionally faster & better than some of Intel's older big expensive energy sucking Core i7 kahunas. Tom's is also keeping track of the selling prices and tracking the various resulting "value per dollar" effects.
Quite frankly, Intel sucks at "value per dollar" lately and that situation is getting worse, not better ....
https://www.tomshardware.com/reviews/cpu-buying-guide,5643.htmlToss into the mix the fact that ALL the AMD cores just naturally perform that fast, while the Intel processors have to run on the very most modern style of Intel motherboards in systems that are LOADED UP with several different layers of expensive Intel Optane memory types WITH ALL OF IT BEING REQUIRED TO WORK TOGETHER to get to that good advertised Intel "processor" performance level ..... and that means you the consumer will have to pay all those hidden costs when you go get "Intel Inside" on your new machine purchase.
..... and you always need to remember that your Intel processor is rated just off that ONE (1) very best core that it contains ..... and the entire shebang never actually runs as fast as that ONE (1) very best core as
all the cores count towards processor performance numbers.
..... it ain't the whole Intel processor, boys and girls ..... it never has been
P.S. please don't forget to calculate in your Intel mitigations for the Spectre, Meltdown and Spoiler vulnerabilities (20% off the top, minimum) .....Now, as far as what is standard processor clock speed, what is "overclocked" and how much higher is it actually ..... these get off into marketing BS with Intel so deeply there is really no way to explain or to compare them.
This is why Tom's set up their own reporting and ranking system, one that is impartial and structured to show the comparative merits of a computer CPU system with all its multiple multiple cores that can actually run at different speeds under different thermal loads.
....... and yes, a better CPU cooler and better memory does mean better and improved performance, generally speaking ....... as does running (or not running) your Meltdown/Spectre/Spoiler mitigations ...... or even which OS systems you run the test under ......===================================================
https://www.tomshardware.com/news/dell-amd-epyc-rome-server,39213.htmlSignificant new SKU announcements from both of the Hewlett Packard and the Dell Server and Consumer sides ---- AMD will immediately take over 30% of HP's server output due to
Intel failing to meet existing CPU Processor orders.
HP and Dell Consumer (desktop, laptop and Chromebook) and the HP and Dell Server Areas are all showing multiple new AMD based units popping up within the last 2 weeks.
On the same note, Dell will immediately TRIPLE the number of Server SKUs with EPYC Server processors inside them instead of having "Intel Xenon Inside". This Dell change goes well beyond the 30% mentioned earlier.
This is a set of simple factual announcements coming out from HP and Dell less than a month ahead of the AMD Computex dog and pony show announcements from AMD. HP is the world's largest supplier of PC products, and coupled with #3 Dell's new AMD product announcements this may signify a large real shift in computing away from Intel and towards AMD.
Lenovo at #2 has not been heard from yet, but Lenovo will be announcing something fairly soon it is believed.
Dell and HP are the most SIGNIFICANT USA finished box producers of consumer and server products ...... and for both of them to be showing such large pro-AMD movements on their own, independently, means that a large pro-AMD market shift really is likely taking place in the US market as we speak.