How bad was AMD Bulldozer and its variants

[wildhorse2k]

Yeah it beats it by a little and 1800X is clocked 400-500Mhz higher than the 6900K at stock clock for non turbo and turbo. How would it fare a the same clock speed.

But yes if you do not overclock and intend to leave it at stock speed, the Ryzen 1800X is a much better buy especially for the price than the 6900K if gaming is not your primary usage.

I am no Intel fanboy, just stating the facts. I wish AMD could beat Intel or be neck and neck with them like back in the AMD Athlon 64 X2 days were AMD spanked Intel or in the P4 Northwood and Athlon XP Barton and Athlon 64 days where they were neck and neck.

Yeah AMD fans fail to realize Ryzen has performance of Haswell. That's quite bad. It's also quite a buggy CPU. FM3 bug, VME bug, high speed RAM problems, cold boot DRAM voltage bug, FMA4 is implemented but CPU claims it isn't supported. Instructions work, but sometimes give "wrong" results. AMD is giving it very cheaply in form of Ryzen 1700 which can be OCed so people are happy even though its buggy.

Trolling is not allowed.
Markfw
Anandtech Moderator

 

[Hi-Fi Man]

Yeah AMD fans fail to realize Ryzen has performance of Haswell. That's quite bad. It's also quite a buggy CPU. FM3 bug, VME bug, high speed RAM problems, cold boot DRAM voltage bug, FMA4 is implemented but CPU claims it isn't supported. Instructions work, but sometimes give "wrong" results. AMD is giving it very cheaply in form of Ryzen 1700 which can be OCed so people are happy even though its buggy.

It's interesting that it's almost like K10 vs Core 2 like someone else mentioned. Core 2 clocked much higher and had higher IPC in most scenarios but AMD countered with three sometimes four cores for the price of Intel's two.

 

[scannall]

Yeah AMD fans fail to realize Ryzen has performance of Haswell. That's quite bad. It's also quite a buggy CPU. FM3 bug, VME bug, high speed RAM problems, cold boot DRAM voltage bug, FMA4 is implemented but CPU claims it isn't supported. Instructions work, but sometimes give "wrong" results. AMD is giving it very cheaply in form of Ryzen 1700 which can be OCed so people are happy even though its buggy.

I'm guessing you don't remember the X58 platform when it came out.....

 

[DrMrLordX]

Yeah AMD fans fail to realize Ryzen has performance of Haswell.

I like how every thread becomes a Ryzen thread now. Why would that be? Oh, and funny joke.

That's quite bad.

. . . whaaaaat? Even if Ryzen had Haswell performance - which it doesn't, check Cinebench or whatever - calling that "quite bad" is pretty ignorant, especially when you consider where AMD's product stack was as recently as this past February.

It's also quite a buggy CPU. FM3 bug, VME bug, high speed RAM problems, cold boot DRAM voltage bug, FMA4 is implemented but CPU claims it isn't supported.

Wow, these chips are completely unusable! It's not like any of that can be patched out with microcode or . . . huh. Woops, guess you can!

And funny joke on the DRAM cold boot issue as well! After all, it's not board-specific or anything. Oh wait . . .

Instructions work, but sometimes give "wrong" results.

Gee good thing nobody actually uses xOP/FMA4 for anything. I think I ran one xOP app the entire two years I owned my Steamroller. Thanks y-cruncher! For supporting everything!

AMD is giving it very cheaply in form of Ryzen 1700 which can be OCed so people are happy even though its buggy.

No, they're happy 'cuz it's awesome.

 

[itsmydamnation]

Yeah AMD fans fail to realize Ryzen has performance of Haswell. That's quite bad.

sigh i guess someone has to continue the intel rear guard action, 3ghz and sandybridge IPC with HT at prescott levels i hear. According to someone who actually did a good job at benchmarking 103% perf per clock of haswell with skylake 7% down the road. BD to PD did that, when is intels next actual uarch update, 2087?

absolutely terrible performance....

It's also quite a buggy CPU. FM3 bug,

You mean running a fake workload( bet you haven't even look at the code) that will never happen in the real world, i guess skylake is just as terrible with FMA.

VME bug

yeah its a pain but we can see how big an actual deal it is by how long it took to figure it out, the ESXi pink screen problem is far more annoying then a 16bit real time mode bug that can easily be worked around with a hypervisor flag.

high speed RAM problems,

Not really, just lack of configurations options mean you hit walls quicker, 1.0.0.6 looks like it opens up all the options you would expect to see.

cold boot DRAM voltage bug,

That seems to be MB vendors as there are plently of boards without those problems.

FMA4 is implemented but CPU claims it isn't supported. Instructions work, but sometimes give "wrong" results.

Now your really scraping the bottom of the barrel, really, are you actually serious?
3Dnow, XOT, FMA4 are all gone because there is on point in having them its just extra validation work. It would also take work to remove FMA4 from decoders and FP scheduler

AMD is giving it very cheaply in form of Ryzen 1700 which can be OCed so people are happy even though its buggy.

None of the bugs you have listed really matter, its funny actually , before Zen released i got lambasted by intel fans for calling DDR4 3200 CL16 middle of the road memory and now that its apparent that getting 3400,3600 was a challenge and need bclock oc to hit it (1.0.0.6 looks the goods in this regard), we hear Zen has memory speed problems.

 

[wildhorse2k]

sigh i guess someone has to continue the intel rear guard action, 3ghz and sandybridge IPC with HT at prescott levels i hear. According to someone who actually did a good job at benchmarking 103% perf per clock of haswell with skylake 7% down the road. BD to PD did that, when is intels next actual uarch update, 2087?

absolutely terrible performance....

Speculations. If performance benchmarks reveal Skylake-X to be a flop, I won't buy it. It's plain simple.

You mean running a fake workload( bet you haven't even look at the code) that will never happen in the real world, i guess skylake is just as terrible with FMA.

That's your claim. In reality we know it can sometimes happen, probability is very low but I as a consumer want a stable CPU. Your argument about fake load doesn't load as we don't know what other loads could cause it. We have just an example one one load. I don't want random crashes.

yeah its a pain but we can see how big an actual deal it is by how long it took to figure it out, the ESXi pink screen problem is far more annoying then a 16bit real time mode bug that can easily be worked around with a hypervisor flag.

Many little bugs => poorly tested product => more hidden bugs. The result is what really matters like ESXi doesn't work because of additional hidden bugs we don't know about.

Not really, just lack of configurations options mean you hit walls quicker, 1.0.0.6 looks like it opens up all the options you would expect to see.

AGESA 1.0.0.6 is out at least on Asus C6H and people are still having memory problems. It got a little better but still requires tinkering in BIOS to get your RAM to work. Sometimes you just can't. Changing voltages to get rid of memory hole problem is not normal, as memory instability could come back at different temperature or with time as things get more worn out. As of now it is for people who like to tinker with the system and not who expect it to work out of box.

That seems to be MB vendors as there are plently of boards without those problems.

Multiple MBs have this problem, Asus C6H, ASRock Taichi. From what I have read it's not just one.

Now your really scraping the bottom of the barrel, really, are you actually serious?
3Dnow, XOT, FMA4 are all gone because there is on point in having them its just extra validation work. It would also take work to remove FMA4 from decoders and FP scheduler

It serves to demonstrate the quality of product AMD ships. They couldn't fix it in time so just disable it in CPUID. The same may happen to VME bug if they can't fix it in microcode.

Ryzen 1800X runs on the brink of stability at 4Ghz, with XFR it can reach more that could potentially be unstable in time. It is not normal to ship a CPU at frequencies that close to maximum potential (4.1Ghz).

 

[whm1974]

@wildhorse2k: I've been keeping track of AMD's Ryzen CPU. Offhand from all the people that have been buying them, I haven't yet heard of anyone returning any CPUs at all. Yes Ryzen has its problems at launch - but AMD has fixing those problems by microcode and working with motherboards manufacturers on BIOS updates.

From reading the users reports and reviews, AMD is now back on the map for high performance CPUs.

 

[DrMrLordX]

Multiple MBs have this problem, Asus C6H, ASRock Taichi. From what I have read it's not just one.

Taichi? BS, mine runs DDR4-3200 14-14-14-32 @ 1.4v vDIMM and has since I got it. Cold boot bug my buttocks. All I had to do was get b-die and all was well.

It is not normal to ship a CPU at frequencies that close to maximum potential

AMD has been doing it since the days of the 1.4 GHz Tbird. Intel did it with the 4790k.

 

[wildhorse2k]

@wildhorse2k: I've been keeping track of AMD's Ryzen CPU. Offhand from all the people that have been buying them, I haven't yet heard of anyone returning any CPUs at all.

There are shops which do not sell returned CPUs as new. It's not happening on masse but some people do return them. And you can see on such shops returned CPUs being sold as separate items. Many people don't even know about the issues I mentioned or accept them as the CPU was cheap.

All I had to do was get b-die and all was well.

It is not normal to require from customer certain brand of chips to work at advertized frequency.

AMD has been doing it since the days of the 1.4 GHz Tbird. Intel did it with the 4790k.

That doesn't make it normal. I want my CPU to remain functional for a long time at advertized frequency, not require lowering frequency after warranty expires. With 1800X I really don't know how long they will last.

 

[DrMrLordX]

It is not normal to require from customer certain brand of chips to work at advertized frequency.

Keep your normal. I'm voting for awesome.

That doesn't make it normal.

So let me get this straight, Intel's flagship Devil's Canyon CPU ships at a clockspeed within 200 MHz of its freq cap on air and . . . yeah whatever.

I want my CPU to remain functional for a long time at advertized frequency, not require lowering frequency after warranty expires. With 1800X I really don't know how long they will last.

So how many of them 4790ks have required downlocking to run at "advertized frequency"?

 

[wildhorse2k]

So how many of them 4790ks have required downlocking to run at "advertized frequency"?

My personal experience is that if CPU/GPU runs very close to limit, stable frequency gradually decreases. Therefore I have good reason not to trust Ryzen 1800X as there is extremely low reserve with XFR on at 4.1Ghz. This reserve tends to be lower with AMD than with Intel.

 

[scannall]

My personal experience is that if CPU/GPU runs very close to limit, stable frequency gradually decreases. Therefore I have good reason not to trust Ryzen 1800X as there is extremely low reserve with XFR on at 4.1Ghz. This reserve tends to be lower with AMD than with Intel.

So, the 7700K thermal problems also bother you?

 

[AtenRa]

They were horrible, some workloads were slower than Phenom 2, and power/temperature are trought the roof compared to Sandy/Ivy Bridge.

In the recent years FX8350 has managed to catch up with the 2500K in some heavy MT workloads, still the FX8350 launched after Ivy.

You really dont know what you are talking about, even FX8150 was faster in MT loads than 2500K. FX8350 was even faster than 2600K in MT loads.


FX8150 vs 2500K
http://www.anandtech.com/bench/Product/434?vs=288

FX8350 vs 2500K
http://www.anandtech.com/bench/product/697?vs=288

 

[wildhorse2k]

So, the 7700K thermal problems also bother you?

No as I intend to use custom water cooling. In comparison the Threadripper 16C is speculated to have turbo clock 3.9Ghz. This looks safer than what's going on with 1800X. I prefer more cores as then TDP will be the limit and CPU will not be running anywhere near its maximum stable frequency as it will be unreachable due to thermal problems. As long as there is sufficient temperature reserve, I consider it safer than borderline overclocking with few core CPUs which results also in high temperatures due to high leakage currents.

 

[scannall]

No as I intend to use custom water cooling. In comparison the Threadripper 16C is speculated to have turbo clock 3.9Ghz. This looks safer than what's going on with 1800X. I prefer more cores as then TDP will be the limit and CPU will not be running anywhere near its maximum stable frequency as it will be unreachable due to thermal problems. As long as there is sufficient temperature reserve, I consider it safer than borderline overclocking with few core CPUs which results also in high temperatures due to high leakage currents.

You shouldn't have to run water to not throttle at stock speeds. And yes, quite a few people have 7700K's that throttle at stock speeds. Seems clocked pretty close to it's limit...

 

[majord]

My personal experience is that if CPU/GPU runs very close to limit, stable frequency gradually decreases. Therefore I have good reason not to trust Ryzen 1800X as there is extremely low reserve with XFR on at 4.1Ghz. This reserve tends to be lower with AMD than with Intel.

This is managed by tighter voltage regulation, onboard LDO regulators and other technologies like AVFS - which lower the required headroom, compared to traditional schemes, which have to account for larger variations in supplied (vs requested) core voltage, and part-specific variations. I suggest you read up on it, you'll see your concern is unfounded.

That said The 4.1Ghz XFR is never guaranteed, that's why it's classed as XFR. the max precision boost freq is 4.0Ghz.

 

[wildhorse2k]

You shouldn't have to run water to not throttle at stock speeds. And yes, quite a few people have 7700K's that throttle at stock speeds. Seems clocked pretty close to it's limit...

Desktop Core i7 7700K also runs in Clevo laptops (for example Obsidian from Portugal) without throttling so it's more about lot of people buying it without proper cooler or without solving air circulation in their case.

This is managed by tighter voltage regulation, onboard LDO regulators and other technologies like AVFS - which lower the required headroom, compared to traditional schemes, which have to account for larger variations in supplied (vs requested) core voltage, and part-specific variations. I suggest you read up on it, you'll see your concern is unfounded.

That said The 4.1Ghz XFR is never guaranteed, that's why it's classed as XFR. the max precision boost freq is 4.0Ghz.

I have doubts how various sensors in the chip can guarantee overall chip stability at max XFR frequencies in the long term. Most people who OC may never see issues as XFR is disabled. I trust in old school sufficient frequency headroom.

 

[Topweasel]

Desktop Core i7 7700K also runs in Clevo laptops (for example Obsidian from Portugal) without throttling so it's more about lot of people buying it without proper cooler or without solving air circulation in their case.



I have doubts how various sensors in the chip can guarantee overall chip stability at max XFR frequencies in the long term. Most people who OC may never see issues as XFR is disabled. I trust in old school sufficient frequency headroom.

It's for single thread high performance applications. Besides it not really contributing to the heat load since all of the cores have to be clocked at like 1500mhz. It's more of an extra look at my clock speed number the CPU will almost never clock that high. It's so hard to get the CPU to clock to that number in use.

But that's not the problem here. AMD has a much lower temp limit and the clock speed limitations are a process/arch limitation and not a temp limitation like the 7700. We know how high the voltage can can go before instability sets. A stock clocked 1800x doesn't get anywhere close. I don't know why you would think that the 4.1ghz xfr you'll have any long term affect on the longevity of the CPU. Specially how you would feel that you knew more than AMDs engineers. Let's not forget it's not AMD but Intel that actually shipped a chip clocked so high it acted like an overclocked chip and had to be recalled. Even worse case scenarios like the 9k series FXs with their 220w TPD weren't an issue.

 

[DrMrLordX]

My personal experience is that if CPU/GPU runs very close to limit, stable frequency gradually decreases.

And what personal experience would that be? My 1.4 GHz Tbird ran "close to the limit" for about 5 years with a poorly-installed heatsink without skipping a beat. It would have gone further had the guy I donated it to given it a better PSU (not sure if he replaced the cheap-arsed one that went out on him, lost track of it).

 

[waltchan]

One of my Phenom II X2 570 Callisto can run up to 4.2GHz max on stock voltage, and 4.4GHz max with one core disabled. Single-thread performance higher than FX-4100 too.

 

[AtenRa]

One of my Phenom II X2 570 Callisto can run up to 4.2GHz max on stock voltage, and 4.4GHz max with one core disabled. Single-thread performance higher than FX-4300 too.

You know you can also OC the FX 4300 to 4,8-5.0GHz single core.
What is also wrong is comparing the dual module Bulldozer to quad core Star. Its like comparing the SandyBridge Core i3 to Core 2 Quad. Just because both have 4x threads doest make them the same.

 

[scannall]

You know you can also OC the FX 4300 to 4,8-5.0GHz single core.
What is also wrong is comparing the dual module Bulldozer to quad core Star. Its like comparing the SandyBridge Core i3 to Core 2 Quad. Just because both have 4x threads doest make them the same.

Pretty much this. I take all posted benchmarks with a very large grain of salt. We'll get real numbers soon enough.

 

[waltchan]

You know you can also OC the FX 4300 to 4,8-5.0GHz single core.
What is also wrong is comparing the dual module Bulldozer to quad core Star. Its like comparing the SandyBridge Core i3 to Core 2 Quad. Just because both have 4x threads doest make them the same.

Typo-error. I mean to write FX-4100. Never own a FX-4300. FX-4100 can only go 4.4GHz max on stock voltage with Gigabyte 78LMT-USB3 board.

 

[krumme]

You know you can also OC the FX 4300 to 4,8-5.0GHz single core.
What is also wrong is comparing the dual module Bulldozer to quad core Star. Its like comparing the SandyBridge Core i3 to Core 2 Quad. Just because both have 4x threads doest make them the same.

Stars cores were small. The comparison is relevant based on cost.
A 8150 was 315mm2 on 32nm.
Thuban x6 was 346mm2 on 45nm.
A zen core is 193mm2 btw.
Bd was a epic failure. And mostly because of sad efficiency and its huge size and not because of its lackluster performance from witch it was famous. It was worse than its reputation.

 

[amd6502]

You know you can also OC the FX 4300 to 4,8-5.0GHz single core.
What is also wrong is comparing the dual module Bulldozer to quad core Star. Its like comparing the SandyBridge Core i3 to Core 2 Quad. Just because both have 4x threads doest make them the same.

Typo-error. I mean to write FX-4100. Never own a FX-4300. FX-4100 can only go 4.4GHz max on stock voltage with Gigabyte 78LMT-USB3 board.

Good point, compare OC'd to OC'd at similar wattage.

K10 was definitely better than BD1, but with PD (what I consider the "real" dozer, rather than prototype dozer BD1) it's harder to tell (my guess is PD is slightly better).

I do think it's fair to compare 2 dozer modules to four K10's, because the areas and number of transistors are similar (K10 does not have an unfair advantage there).

I think to go from 6 wide integer core to 4 wide was too drastic a change; dozers definitely seem to be short an ALU or two. Excavator may have reached the limits of what can be done with a 4 wide core (maybe if they allowed more ALU operations on the AGU's it could be pushed a little higher?).