Chips and Cheese
Intel Finally Solving the Right Problem
#ChipAndCheese
Telegraph | source
(author: brians256)
Intel Finally Solving the Right Problem
#ChipAndCheese
Telegraph | source
(author: brians256)
Chips and Cheese
Intel’s HEDT Roadmap
#ChipAndCheese
Long the king of High-End Desktop (HEDT) computing, Intel’s once undisputed position has become a much more awkward one since the arrival of a resurgent AMD and its Threadripper CPUs.
X299 is Dead!
The header says it all: X299 is dead, at least in the original form that Intel launched it. Available for pre-order on the 26th of June, 2017 with the aim of being the fastest consumer/prosumer CPU platform on the planet, with the most PCIe lanes, Intel did accomplish that goal… For a short while.
The crown would not last long, as Intel would lose the title of “most PCIe lanes” when AMD released X399 on the 10th of August of 2017 (a mere 6 weeks later) but, the platform still held the world’s fastest HEDT CPU until AMD’s 2990WX was launched nearly a year later, in August of 2018. Unfortunately for AMD, the 2990WX+X399 combo wouldn’t quite take the crown away from Intel as Windows’ scheduler had issues with the AMD platform, keeping X299 on top for the fastest Windows platform available.
Intel’s loss of the HEDT war wouldn’t be fully realized until AMD released the Threadripper 3000 series, and finally, in one product release, X299 was obsoleted almost over night. Now with the launch of Zen 3, even the plebian consumer AM4 platform’s 16-core 5950X makes the newest 18-core CPU on the X299 platform, the 10980XE, pointless for all but 3 groups of people: overclockers that need more than 20 threads for benchmarks that can take advantage of them; people that need lots of PCIe lanes for fairly cheap; the people that need a lot of memory bandwidth.
But that is enough of the past, let’s look forward and see: What does Intel have coming out for HEDT?
Long Live X299!
In a word, nothing. Nada. Zip. Zilch. Sorry for the letdown, but the sad truth is that Intel has nothing on their HEDT roadmap until Sapphire Rapids which is not releasing to the general public until 20231 at the earliest.
Now, I hear you over there screaming “But Cheese, what about Ice Lake! That is coming to HEDT right?” Unfortunately, none of the roadmaps I have seen have shown any sign of an Ice Lake HEDT product coming to market. The most likely reason for this are poor yields on the large Ice Lake dies.
Intel and HEDT: An on-again, off-again Relationship
So, it appears AMD has won this round. Intel has completely ceded the HEDT market to AMD until at least 2023 which begs two questions:
1. What will AMD do with HEDT and,
2. Can Intel recover in the HEDT market when Sapphire Rapids launches?
To answer the first question, Threadripper is not going anywhere soon, however, do not be surprised by future price increases. And to answer the second question, I do not believe that Intel can recover to the same near absolute stranglehold they had on the HEDT market, especially with Zen 4 coming out before Sapphire Rapids.
So… X299 is dead! Long live X299!
1New changes in Intel’s internal timelines made available to Chips and Cheese suggests that HEDT SPR may be available earlier than originally thought, in Q4 2022.
source
Intel’s HEDT Roadmap
#ChipAndCheese
Long the king of High-End Desktop (HEDT) computing, Intel’s once undisputed position has become a much more awkward one since the arrival of a resurgent AMD and its Threadripper CPUs.
X299 is Dead!
The header says it all: X299 is dead, at least in the original form that Intel launched it. Available for pre-order on the 26th of June, 2017 with the aim of being the fastest consumer/prosumer CPU platform on the planet, with the most PCIe lanes, Intel did accomplish that goal… For a short while.
The crown would not last long, as Intel would lose the title of “most PCIe lanes” when AMD released X399 on the 10th of August of 2017 (a mere 6 weeks later) but, the platform still held the world’s fastest HEDT CPU until AMD’s 2990WX was launched nearly a year later, in August of 2018. Unfortunately for AMD, the 2990WX+X399 combo wouldn’t quite take the crown away from Intel as Windows’ scheduler had issues with the AMD platform, keeping X299 on top for the fastest Windows platform available.
Intel’s loss of the HEDT war wouldn’t be fully realized until AMD released the Threadripper 3000 series, and finally, in one product release, X299 was obsoleted almost over night. Now with the launch of Zen 3, even the plebian consumer AM4 platform’s 16-core 5950X makes the newest 18-core CPU on the X299 platform, the 10980XE, pointless for all but 3 groups of people: overclockers that need more than 20 threads for benchmarks that can take advantage of them; people that need lots of PCIe lanes for fairly cheap; the people that need a lot of memory bandwidth.
But that is enough of the past, let’s look forward and see: What does Intel have coming out for HEDT?
Long Live X299!
In a word, nothing. Nada. Zip. Zilch. Sorry for the letdown, but the sad truth is that Intel has nothing on their HEDT roadmap until Sapphire Rapids which is not releasing to the general public until 20231 at the earliest.
Now, I hear you over there screaming “But Cheese, what about Ice Lake! That is coming to HEDT right?” Unfortunately, none of the roadmaps I have seen have shown any sign of an Ice Lake HEDT product coming to market. The most likely reason for this are poor yields on the large Ice Lake dies.
Intel and HEDT: An on-again, off-again Relationship
So, it appears AMD has won this round. Intel has completely ceded the HEDT market to AMD until at least 2023 which begs two questions:
1. What will AMD do with HEDT and,
2. Can Intel recover in the HEDT market when Sapphire Rapids launches?
To answer the first question, Threadripper is not going anywhere soon, however, do not be surprised by future price increases. And to answer the second question, I do not believe that Intel can recover to the same near absolute stranglehold they had on the HEDT market, especially with Zen 4 coming out before Sapphire Rapids.
So… X299 is dead! Long live X299!
1New changes in Intel’s internal timelines made available to Chips and Cheese suggests that HEDT SPR may be available earlier than originally thought, in Q4 2022.
source
Chips and Cheese
NVIDIA’s Enterprise
#ChipAndCheese
Well, I guess that this will be the first non-welcome post on this website and the topic of this piece is NVIDIA Corporation, specifically the next-gen server GPU and more broadly the NVIDIA Enterprise division.
Now please remember that what is written below are leaks, rumours, and speculation, so take this with an appropriate amount of caution however what is published in this post is believed to be truthful and accurate.
Next-Generation Server GPU
As kopite7kimi has already posted on Twitter on in this tweet (https://twitter.com/kopite7kimi/status/1336900538185572357), the next-gen server GPU is an MCM design but has been delayed. Now, kopite7kimi did not go into detail as to why the next-gen server GPU has been delayed, however, which is the topic of this article.
From the information we have gathered from sources, it appears that NVIDIA is experiencing issues with making the next-gen server GPU appear as a single unified processor, instead of multiple GPUs, to the operating system because of firmware and driver issues.
NVIDIA Enterprise
We have also received reports that the Enterprise Driver team at NVIDIA has told at least one company “not to expect a response on any issues” from them; this is apparently due to people being moved to a different team at NVIDIA.
It is possible that these two events are linked, and that people were moved to work on the next-gen server GPU’s drivers, however, this is speculation by us here at Chips and Cheese.
The Future of NVIDIA in the Enterprise Space
With the rumored issues that NVIDIA is having in the Enterprise Space and with the now resurgent AMD along with Intel trying to break into the Enterprise GPU as well, NVIDIA looks like it may hit a rough spot in the near future.
source
NVIDIA’s Enterprise
#ChipAndCheese
Well, I guess that this will be the first non-welcome post on this website and the topic of this piece is NVIDIA Corporation, specifically the next-gen server GPU and more broadly the NVIDIA Enterprise division.
Now please remember that what is written below are leaks, rumours, and speculation, so take this with an appropriate amount of caution however what is published in this post is believed to be truthful and accurate.
Next-Generation Server GPU
As kopite7kimi has already posted on Twitter on in this tweet (https://twitter.com/kopite7kimi/status/1336900538185572357), the next-gen server GPU is an MCM design but has been delayed. Now, kopite7kimi did not go into detail as to why the next-gen server GPU has been delayed, however, which is the topic of this article.
From the information we have gathered from sources, it appears that NVIDIA is experiencing issues with making the next-gen server GPU appear as a single unified processor, instead of multiple GPUs, to the operating system because of firmware and driver issues.
NVIDIA Enterprise
We have also received reports that the Enterprise Driver team at NVIDIA has told at least one company “not to expect a response on any issues” from them; this is apparently due to people being moved to a different team at NVIDIA.
It is possible that these two events are linked, and that people were moved to work on the next-gen server GPU’s drivers, however, this is speculation by us here at Chips and Cheese.
The Future of NVIDIA in the Enterprise Space
With the rumored issues that NVIDIA is having in the Enterprise Space and with the now resurgent AMD along with Intel trying to break into the Enterprise GPU as well, NVIDIA looks like it may hit a rough spot in the near future.
source
开题答辩，答辩组长做NLP的，提问：分支是什么？
🤯
🤯
非常遗憾RVI最终以在RVA profile中保留C扩展的决定结束此次高通proposal的讨论。
我认为此次决定才是真正会给fragmentation埋下隐患的。
我认为此次决定才是真正会给fragmentation埋下隐患的。
#今日看了什么
https://lists.riscv.org/g/tech-profiles/topic/rvi_bod_decision_regarding/102522954
In particular, RVI made GC the defacto standard back in 2016 and reaffirmed C as mandatory in RVA profiles up through this past year. The BoD discussion was led by Dave Patterson. They felt it was important to resolve this now and not let the discussion continue. The BoD felt that not only was this the right decision but that other parties were already using this discussion to proclaim RISC-V fragmentation and lack of dependability.
The BoD has sent a strong message that unless there is something catastrophic (e.g. a security issue -- which we have none), that they want RVI members to honor the bond of RVI's word for items like this that are published as mandatory in profiles.
https://lists.riscv.org/g/tech-profiles/topic/rvi_bod_decision_regarding/102522954
In particular, RVI made GC the defacto standard back in 2016 and reaffirmed C as mandatory in RVA profiles up through this past year. The BoD discussion was led by Dave Patterson. They felt it was important to resolve this now and not let the discussion continue. The BoD felt that not only was this the right decision but that other parties were already using this discussion to proclaim RISC-V fragmentation and lack of dependability.
The BoD has sent a strong message that unless there is something catastrophic (e.g. a security issue -- which we have none), that they want RVI members to honor the bond of RVI's word for items like this that are published as mandatory in profiles.
Daniel Lemire's blog
Science and Technology links (November 12 2023)
1.
1. Vitamin K2 supplements might reduce the risk of myocardial infarction (heart attacks) and of all-cause death (Hasific et al. 2022). You find vitamin K2 in some Gouda cheeses and in eggs.
2. Most of the water on Earth is salinated (in the oceans) and cannot be consumed. Fresh water is often scarce. Yet Israel is desalinating water for less than a dollar per cubic meter.
3. People living in South America engaged in warfare for 10,000 years before the arrival of the Europeans (Standen et al. 2023).
4. The last glacial period ended about 12,000 years ago and last about 100,000 years. About 26,000 years ago, all of Canada was covered by a permanent ice sheet. Thus many of us were taught in school that human beings first colonized America about 12,000 years ago by the Bering land bridge, that existed back then between modern-day Russia and modern-day Alaska. The evidence accumulates that there were human beings in America much earlier than initially thougth. They would have been present 21,000 to 23,000 years ago in New Mexico. We even have their footprints.
5. As recently as 20,000 years ago—not long in geological terms—Britain was not, in fact, an island. Instead, the terrain that became the British Isles was linked to mainland Europe by Doggerland, a tract of now-submerged territory where early Mesolithic hunter-gatherers lived, settled and traveled (McGreevy, 2020). Correspondly, there were human beings in Ireland 31,000 years ago.
6. Gray et al. (2023) argue that the limited freedom that children enjoy in our modern societies is leading to a rise in mental disorders.
7. Most people cannot understand the bat and ball problem, even after the solution is given. The problem can be stated as follows: “A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost?”
8. When hiring, we find a slight bias in favour of females in male-dominated fields, and a strong bias in favour of females in female-dominated fields (Schaerer et al., 2023). Overall, people greatly overestimate gender biases in hiring.
9. Retinol, a common cosmetic product, keeps one’s skin younger.
10. Unwarranted financial optimism might be the result of low cognitive abilities.
source
Science and Technology links (November 12 2023)
1.
1. Vitamin K2 supplements might reduce the risk of myocardial infarction (heart attacks) and of all-cause death (Hasific et al. 2022). You find vitamin K2 in some Gouda cheeses and in eggs.
2. Most of the water on Earth is salinated (in the oceans) and cannot be consumed. Fresh water is often scarce. Yet Israel is desalinating water for less than a dollar per cubic meter.
3. People living in South America engaged in warfare for 10,000 years before the arrival of the Europeans (Standen et al. 2023).
4. The last glacial period ended about 12,000 years ago and last about 100,000 years. About 26,000 years ago, all of Canada was covered by a permanent ice sheet. Thus many of us were taught in school that human beings first colonized America about 12,000 years ago by the Bering land bridge, that existed back then between modern-day Russia and modern-day Alaska. The evidence accumulates that there were human beings in America much earlier than initially thougth. They would have been present 21,000 to 23,000 years ago in New Mexico. We even have their footprints.
5. As recently as 20,000 years ago—not long in geological terms—Britain was not, in fact, an island. Instead, the terrain that became the British Isles was linked to mainland Europe by Doggerland, a tract of now-submerged territory where early Mesolithic hunter-gatherers lived, settled and traveled (McGreevy, 2020). Correspondly, there were human beings in Ireland 31,000 years ago.
6. Gray et al. (2023) argue that the limited freedom that children enjoy in our modern societies is leading to a rise in mental disorders.
7. Most people cannot understand the bat and ball problem, even after the solution is given. The problem can be stated as follows: “A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost?”
8. When hiring, we find a slight bias in favour of females in male-dominated fields, and a strong bias in favour of females in female-dominated fields (Schaerer et al., 2023). Overall, people greatly overestimate gender biases in hiring.
9. Retinol, a common cosmetic product, keeps one’s skin younger.
10. Unwarranted financial optimism might be the result of low cognitive abilities.
source
Chips and Cheese
ARM’s Cortex A72: aarch64 for the Masses
#ChipAndCheese
Telegraph | source
(author: clamchowder)
ARM’s Cortex A72: aarch64 for the Masses
#ChipAndCheese
Telegraph | source
(author: clamchowder)
想想你日常生活中用的是狗屁论文，还是能跑的代码
BIRD Linux LLVM GCC Rocket真的能跑
ASPLOS ISCA MICRO 废纸一张
这不比什么狗屁论文好多了
内容：Soha为第一作者、我为第二作者向BIRD（BIRD Internet Routing Daemon）提交的关于OSPF路由协议的patch在昨天被社区接收了。这是我们团队首次在路由协议软件中做出贡献。此前，我们团队曾多次向Linux内核、LLVM编译器、GCC编译器、gem5模拟器、Rocket处理器核等社区提交过多份patch并获得接收。
Chips and Cheese
Core to Core Latency Data on Large Systems
#ChipAndCheese
Telegraph | source
(author: clamchowder)
Core to Core Latency Data on Large Systems
#ChipAndCheese
Telegraph | source
(author: clamchowder)
Daniel Lemire's blog
Generating arrays at compile-time in C++ with lambdas
Suppose that you want to check whether a character in C++ belongs to a fixed set, such as ‘\0’, ‘\x09’, ‘\x0a’,’\x0d’, ‘ ‘, ‘#’, ‘/’, ‘:’, ‘<‘, ‘>’, ‘?’, ‘@’, ‘[‘, ‘\\’, ‘]’, ‘^’, ‘|’. A simple way is to generate a 256-byte array of Boolean values and lookup the value. This approach is sometimes called memoization (and not memorization!!!). You might do it as follows:
It is reasonably efficient in practice. Some people might object to how the table is generated. Can you have the C++ compiler generate the array at compile-time from a function?
Using C++17, you might do it with an std::array as follows:
These two approaches should be equivalent in practice.
source
Generating arrays at compile-time in C++ with lambdas
Suppose that you want to check whether a character in C++ belongs to a fixed set, such as ‘\0’, ‘\x09’, ‘\x0a’,’\x0d’, ‘ ‘, ‘#’, ‘/’, ‘:’, ‘<‘, ‘>’, ‘?’, ‘@’, ‘[‘, ‘\\’, ‘]’, ‘^’, ‘|’. A simple way is to generate a 256-byte array of Boolean values and lookup the value. This approach is sometimes called memoization (and not memorization!!!). You might do it as follows:
constexpr static bool is_forbidden_host_code_point_table[] = {
1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
bool is_forbidden_host_code_point(char c) {
return is_forbidden_host_code_point_table[uint8_t(c)];
}
It is reasonably efficient in practice. Some people might object to how the table is generated. Can you have the C++ compiler generate the array at compile-time from a function?
Using C++17, you might do it with an std::array as follows:
constexpr static std::array<uint8_t, 256> is_forbidden_array = []() {
std::array<uint8_t, 256> result{};
for (uint8_t c : {'\0', '\x09', '\x0a','\x0d', ' ', '#', '/', ':',
'<', '>', '?', '@', '[', '\\', ']', '^', '|'}) {
result[c] = true;
}
return result;
}();
bool is_forbidden_host_code_point_array(char c) {
return is_forbidden_array[uint8_t(c)];
}
These two approaches should be equivalent in practice.
source
