# What's wrong with AMD?

**URL:** <https://boards.straightdope.com/t/whats-wrong-with-amd/761200>\
**Category:** In My Humble Opinion\
**Created:** [July 29, 2016, 4:32pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200 "2016-07-29T16:32:27Z")\
**Posts on this page:** 8\
**Page:** 2

<div class="post-metadata">

**Author:** ![ReticulatingSplines](https://avatars.discourse-cdn.com/v4/letter/r/e274bd/32.png) [@ReticulatingSplines](https://boards.straightdope.com/u/ReticulatingSplines)\
**Post date:** [August 2, 2016, 5:21pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/21 "2016-08-02T17:21:08Z")

</div>

> [@Mangetout](#):
>
> The benchmarks on the FX-8350 (which I just searched today after reading your post) [surprised me](https://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8350+Eight-Core) - right up there with the big guys.

That’s the multithread score though. Look at how abysmally low the single thread score is.

---

<div class="post-metadata">

**Author:** ![ReticulatingSplines](https://avatars.discourse-cdn.com/v4/letter/r/e274bd/32.png) [@ReticulatingSplines](https://boards.straightdope.com/u/ReticulatingSplines)\
**Post date:** [August 2, 2016, 5:41pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/22 "2016-08-02T17:41:38Z")

</div>

> [@MichaelEmouse](#):
>
> Any ballpark figures on Nvidia’s R&D budget?

348.25M per quarter

> [@](#):
>
> What made it better?

It wasn’t so much the Athlon being spectacular as it was the Pentium 4 being a huge dud. At a given clock speed the Athlon chips handily outperformed their Intel counterparts, and the P4 chips required a new kind of RAM. Hopefully, someone with a better technical background than me can tell you why, from a hardware perspective, the Athlon performed better.

---

<div class="post-metadata">

**Author:** ![MichaelEmouse](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/michaelemouse/32/101_2.png) [@MichaelEmouse](https://boards.straightdope.com/u/MichaelEmouse)\
**Post date:** [August 7, 2016, 10:48pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/23 "2016-08-07T22:48:52Z")

</div>

> [@ReticulatingSplines](#):
>
> That’s the multithread score though. Look at how abysmally low the single thread score is.

Are there applications where it would make sense to use a CPU with multiple cores and threads rather than a GPGPU?

> [@ReticulatingSplines](#):
>
> 348.25M per quarter

That’s surprisingly low.

Unless you’re making a “3 fiddy” joke.

> [@ReticulatingSplines](#):
>
> It wasn’t so much the Athlon being spectacular as it was the Pentium 4 being a huge dud. At a given clock speed the Athlon chips handily outperformed their Intel counterparts, and the P4 chips required a new kind of RAM. Hopefully, someone with a better technical background than me can tell you why, from a hardware perspective, the Athlon performed better.

Can anyone explain (to non-CS people) what made the Pentium 4 a dud?

> [@ftg](#):
>
> Also, the PC cpu market isn’t doing too great. Even Intel knows that they aren’t going to be making tons of dough here going forward. The ARM chipmakers are making truckloads of CPUs. AMD isn’t competing in this market and Intel has blown its chances to be a major player. _Both_ companies have a big problem.

What else can Intel do?

Could the ARM chipmakers enter markets like gaming GPGPUs? Could computing as a whole switch to the ARM instruction set?

---

<div class="post-metadata">

**Author:** ![Dr.Strangelove](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/dr.strangelove/32/6613_2.png) [@Dr.Strangelove](https://boards.straightdope.com/u/Dr.Strangelove)\
**Post date:** [August 7, 2016, 10:58pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/24 "2016-08-07T22:58:38Z")

</div>

> [@MichaelEmouse](#):
>
> Can anyone explain (to non-CS people) what made the Pentium 4 a dud?

The architecture had low IPC (instructions per clock), but they expected to make up for it with very high clock rates. At one point they were talking about 10 GHz chips. Obviously, this never happened–they never got much above 4 GHz. Silicon just doesn’t scale very well to these high clocks. Better off sacrificing a few hundred MHz and going for high IPC.

The current lineup of chips (Core series) has its architectural heritage from a line that was intended for low-power mobile chips (out of their Israeli office). The P4 line was essentially abandoned. As I’ve said earlier, power efficiency is really where its at, so it’s no coincidence that their current line is based on one where power efficiency was a goal from the very start.

---

<div class="post-metadata">

**Author:** ![MichaelEmouse](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/michaelemouse/32/101_2.png) [@MichaelEmouse](https://boards.straightdope.com/u/MichaelEmouse)\
**Post date:** [August 8, 2016, 10:31am UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/25 "2016-08-08T10:31:56Z")

</div>

> [@Dr.Strangelove](#):
>
> The architecture had low IPC (instructions per clock), but they expected to make up for it with very high clock rates. At one point they were talking about 10 GHz chips. Obviously, this never happened–they never got much above 4 GHz. Silicon just doesn’t scale very well to these high clocks. Better off sacrificing a few hundred MHz and going for high IPC.

I understand that a complete answer to the following may require a lot more background knowledge than I have:

Earlier, you talked about how the GTX 1060 does less work per clock but can be clocked 40%+ higher than the RX 480. Here, you say that more IPC may be preferable. I don’t see a contradiction but it does make me curious about the principles and trade-offs involved. What governs whether there should be more instructions or more clocks? Can you give an example?

Roughly speaking do power & heat increase linearly with instructions but quadratically with clocks?

Also, am I mistaken in thinking that coming up with middleware and specialized hardware combinations like the different GameWorks suites makes a lot of sense? Any major downsides? If it does make sense, any guesses as to why it took until 2016 for AMD to start offering it?

---

<div class="post-metadata">

**Author:** ![jz78817](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/jz78817/32/1028_2.png) [@jz78817](https://boards.straightdope.com/u/jz78817)\
**Post date:** [August 8, 2016, 2:35pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/26 "2016-08-08T14:35:14Z")

</div>

> [@MichaelEmouse](#):
>
> I understand that a complete answer to the following may require a lot more background knowledge than I have:
> 
> Earlier, you talked about how the GTX 1060 does less work per clock but can be clocked 40%+ higher than the RX 480. Here, you say that more IPC may be preferable. I don’t see a contradiction but it does make me curious about the principles and trade-offs involved. What governs whether there should be more instructions or more clocks? Can you give an example?
> 
> Roughly speaking do power & heat increase linearly with instructions but quadratically with clocks?  
> Also, am I mistaken in thinking that coming up with middleware and specialized hardware combinations like the different GameWorks suites makes a lot of sense? Any major downsides? If it does make sense, any guesses as to why it took until 2016 for AMD to start offering it?

GPUs are a different animal, in that the job they’re designed for is “embarrassingly parallel.” GPUs basically carry out the same instructions/calculations over and over, so you can get more performance by either increasing the clock _or_ throwing more execution units in it. More units means the chip can have more pixels in the pipeline.

---

<div class="post-metadata">

**Author:** ![control-z](https://avatars.discourse-cdn.com/v4/letter/c/eada6e/32.png) [@control-z](https://boards.straightdope.com/u/control-z)\
**Post date:** [August 8, 2016, 3:53pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/27 "2016-08-08T15:53:31Z")

</div>

> [@MichaelEmouse](#):
>
> Are there applications where it would make sense to use a CPU with multiple cores and threads rather than a GPGPU?

For a gamer ideally you want both. Good games support multiple cores, but not all do it or do it well.

---

<div class="post-metadata">

**Author:** ![sweat209](https://avatars.discourse-cdn.com/v4/letter/s/9fc348/32.png) [@sweat209](https://boards.straightdope.com/u/sweat209)\
**Post date:** [August 8, 2016, 9:29pm UTC](https://boards.straightdope.com/t/whats-wrong-with-amd/761200/28 "2016-08-08T21:29:25Z")

</div>

AMD market is different than Intel market.

AMD more for gamers and people building their own computer.

Well Intel more for businesses and servers.

Likewise a Intel Xeon E5-2697 12 core would not perform as well as gaming computer.

A a Intel Xeon E5-2697 12 core would be better for businesses, servers, CAD or video editing.

I don’t see market changing any time soon.

[Previous page](https://boards.straightdope.com/t/whats-wrong-with-amd/761200.md?page=1)
