News Intel Offers $2 Billion for RISC-V Startup SiFive: Bloomberg

waltc3

Reputable
Aug 4, 2019
420
223
5,060
Makes sense as Intel doesn't seem to know where to turn for help in designing new, ground-up CPU architectures. My thought would be that Intel wants to enhance its controller business as Intel is a chip company, not just a CPU company--putting all that money into FABs means you have to keep them running making something, even if it isn't CPUs.
 

InvalidError

Titan
Moderator
Makes sense as Intel doesn't seem to know where to turn for help in designing new, ground-up CPU architectures.
There isn't any such thing as a fundamentally new CPU architecture anymore: compare AMD, ARM, Intel, IBM, etc. designs from the past 10+ years and you find many more similarities than difference regardless of ISAs. The only major difference between architectures is how much silicon and power gets allocated to each part of the pipeline and even that tends to be similar across the board since all instruction sets are affected similarly by cache hit rate, latency, branch prediction accuracy, re-order queue depth, etc. which are themselves dictated by typical software design.

Software design is the biggest bottleneck.
 

VforV

Respectable
BANNED
Oct 9, 2019
578
287
2,270
I would not stress about Intel's future... they would need to lose for years and years to actually have issues and worry about being done.

As an AMD fan, I'm more worried about this leak, if it turns true and manages to exceed expectations:
https://videocardz.com/newz/intel-a...lake-hx-series-for-enthusiast-notebook-series
View: https://www.youtube.com/watch?v=YCRoo0tqJL4


If MLiD is right as he's been many times in the past, looks like the competition will get fierce and although I want intel competitive, I prefer AMD winning, even by a bit - they still are the little dog in this fight (and vs nvidia too).
 

zodiacfml

Distinguished
Oct 2, 2008
1,228
26
19,310
Intel needs fabs not fabless.😂 Seriously, this is a purchase way too early but could prove beneficial in the future after Intel gets its new fabs
 

watzupken

Reputable
Mar 16, 2020
1,007
507
6,070
I guess this move shows that Intel is really concerned about ARM and conceding that their products are no longer that sought after as they used to. While AMD is also chipping away at Intel's market share across all sectors, but ARM is chipping chunks away, in particular at the enterprise/ data center segment which hurts them the most. With Apple successfully introducing a worthy ARM chip to the consumer market, it is no wonder why Intel is so aggressive putting up marketing materials to downplay the threat from a certain lifestyle company.
 

watzupken

Reputable
Mar 16, 2020
1,007
507
6,070
I would not stress about Intel's future... they would need to lose for years and years to actually have issues and worry about being done.

As an AMD fan, I'm more worried about this leak, if it turns true and manages to exceed expectations:
https://videocardz.com/newz/intel-a...lake-hx-series-for-enthusiast-notebook-series
View: https://www.youtube.com/watch?v=YCRoo0tqJL4


If MLiD is right as he's been many times in the past, looks like the competition will get fierce and although I want intel competitive, I prefer AMD winning, even by a bit - they still are the little dog in this fight (and vs nvidia too).
I actually won't be too concerned. Not until I see the actual performance. Having Gracemont as powerful as a Skylake processor sounds fairly formidable. But that being a efficient core means that it probably can't run at very high clock speed. If you are looking at games now not being able to fully utilise more than 8 cores, then the 8 or 16 efficient cores will only be stowed away under gaming load. I also suspect an additional software layer to help the application know which cores to tap on, may result in some performance penalty.
 
  • Like
Reactions: VforV

TheHughMan

Prominent
May 15, 2020
18
1
515
Open source or not, Intel is on a mission to turn RISC-V into RISC-0, as in Zero Risk & Threat to Intel like all their merg, I mean acquisitions.
 
Open source or not, Intel is on a mission to turn RISC-V into RISC-0, as in Zero Risk & Threat to Intel like all their merg, I mean acquisitions.
In the last 5 years intel bought habana, nervana (AI) altera, omnitek (fpga) and we have seen products based on all of them already, all the iGPUs now have AI.
Even if risc-v wasn't open source and thus not part of this deal, intel wouldn't disappear it.
It's probably cheaper and definitely faster to just buy a completed product and integrate it then it is to R&D it yourself.
A basic risc-v controller would make sense for their new GPUs to add more features or even make them boot without any external CPU, it could also add more power to their fpga boards, heck the smaller cores in future alder lake CPUs could be risc-v to improve efficiency at idle/low loads.
 
  • Like
Reactions: TheHughMan
Jun 14, 2021
3
0
10
There isn't any such thing as a fundamentally new CPU architecture anymore: compare AMD, ARM, Intel, IBM, etc. designs from the past 10+ years and you find many more similarities than difference regardless of ISAs. The only major difference between architectures is how much silicon and power gets allocated to each part of the pipeline and even that tends to be similar across the board since all instruction sets are affected similarly by cache hit rate, latency, branch prediction accuracy, re-order queue depth, etc. which are themselves dictated by typical software design.

Software design is the biggest bottleneck.

Depends what software I guess? No wait, if it were just software, why did Itanium fail so horrendously? It could run Linux, Windows, OpenVMS, what more could anyone possibly need? </sarcasm>

At least RISC-V seems off to a better start with Linux, FreeBSD, and most recently OpenBSD announcing support for it. LLVM, golang, gcc and other very important toolchains are also already supporting RISC-V.

To think that in 2021, we can still buy new MIPS64 based systems which have a wide range of operating system support, even from volunteer run projects such as OpenBSD, yet we can't buy new Itanium systems (no, old stock on eBay for a couple hundred bucks doesn't count), and HP will officially discontinue their Enterprise support for Itanium in July of this year kind of makes me wonder, is software design the biggest bottleneck?

Heck, this site is still tom's HARDWARE, not tom's SOFTWARE, right?

I think hardware plays a significant part, software doesn't run on thin air, no matter who might be investing $ to support a hardware platform. If a hardware platform doesn't have customers, that to me is a bigger red flag than whether it has some limited amount of software for it. Commodore Amiga enthusiasts, still have new hardware being produced for their niche market, decades after Commodore declared bankruptcy. That certainly helps individuals who still develop software for that platform. However, even those people seem to be gushing over the Vampire FPGA and ARM and PowerPC accelerator upgrades that have been available for their systems over the ensuing decades so that they can run their demos and whatnot faster than stock hardware from 1994. Who knows, in another decade or two, maybe there will be RISC-V based Commodore Amiga accelerator cards for the likes of them. Regardless, they still want and clearly prefer hardware to emulators for most appreciable applications, warez kiddies and PiMIGA projects notwithstanding.
 
Last edited:
Jun 14, 2021
3
0
10
In the last 5 years intel bought habana, nervana (AI) altera, omnitek (fpga) and we have seen products based on all of them already, all the iGPUs now have AI.
Even if risc-v wasn't open source and thus not part of this deal, intel wouldn't disappear it.
It's probably cheaper and definitely faster to just buy a completed product and integrate it then it is to R&D it yourself.
A basic risc-v controller would make sense for their new GPUs to add more features or even make them boot without any external CPU, it could also add more power to their fpga boards, heck the smaller cores in future alder lake CPUs could be risc-v to improve efficiency at idle/low loads.

Intel can't "disappear" RISC-V anymore than Apple or Microsoft could "disappear" FreeBSD or OpenBSD. The licensing of RISC-V permits and encourages reuse freely. There are already other established vendors in the market with RISC-V based designs aside from SiFive. That's entirely the point of having an open and freely reusable CPU ISA. Whether Western Digital's SweRV, or Olof Kindgren's SERV, there is no singular RISC-V hegemony, there is apparent strength through diversity with multiple implementations already demonstrated in the field and many more prototypes and projects announced than I can easily summarize here without this post becoming rather long.

SiFive has, near as I can discern, served as something closer to a "reference implementation" of RISC-V given that it was founded by Krste Asanović, Yunsup Lee, and Andrew Waterman, three researchers from the University of California Berkeley where RISC-V has its roots. Presumably, much like the BSD derived TCP/IP stack becoming the reference implementation from which many other implementations were derived, so too will SiFive's part play a similar role to others who may choose to make their own RISC-V derived implementations.

Per your GPU prognostications, NVidia has been rather open that their next generation GPU designs will be 64bit RISC-V based (their existing GPU designs have had 32bit ARM cores, presumably why Nvidia has been eager to attempt to acquire ARM from SoftBank since unlike Apple, Nvidia does not have a perpetual ARM license).

Also see: "Nvidia RISC-V Evaluation Story" presented by Joe Xie in 2016 and viewable on youtube. Additional research has been presented by Joe Xie and other Nvidia researchers on their iterations of RISC-V research.

Of course, NVidia's publicly expressed and vetted interests in RISC-V aren't constrained to just their next generation GPU designs.

Also see: "Case study on RISC-V Verification with Nvidia Networking" (aka what was previously known as Mellanox Technologies/Networking) as presented by Larry Lepitas of Imperas at DVCon 2021, also viewable on youtube.