- Joined
- Aug 20, 2006
- Messages
- 13,000
Overclocker Roman Hartung (der8auer) has delidded a Threadripper CPU, and photos of it may be found here. I am reading that two of the four dies are disabled.
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
Waiting for Intel do drop a response before enabling the other two dies and renaming it "Butt-Hurter"?
It seems odd to populate all four dies and then disable two. It would have to be the same two disabled all the time which is not going to allow selling an Epyc with failed cores as a consumer part. If it is the case, I would have rather AMD pulled an Intel and used a smaller cheaper socket for the HEDT parts.
Its possible they have some clever trick to make it work for any of the dies. Are we dead sure that they aren't 4x 4c enabled dies with 1 memory channel enabled each?
I always found that weird, but maybe they just dont have a contract or enough money to do such contract with DF.It seems odd to populate all four dies and then disable two. It would have to be the same two disabled all the time which is not going to allow selling an Epyc with failed cores as a consumer part. If it is the case, I would have rather AMD pulled an Intel and used a smaller cheaper socket for the HEDT parts.
Looks like this was an Engineering Sample per the other thread.
Retail might not have four dies
Pencil mod that!
That makes no sense at all when looking at the diagram of epyc where EACH CPUdie has 64lanes. Not the least of which at this moment the delidding was likely an engineering model.if the 1700-1800x has 24 PCI-E lanes, then to get 64, you would need 3 dual CCX at least? Stick a 4th one in and use only silicon that would have been for a 1400 with one failed memory channel and BOOM, you have a Thread Ripper. I'm thinking AMD is not getting the such high quality yields that their fanboys are raving about and this is a pretty smart-ass way to get top dollar for an otherwise crappy silicon. Anyways the proof in in the pudding. If we can measure PCI-E latencey, I'm guess it is going vary since some lanes will have to traverse the Infinity fabric.
I'm getting all moist thinking about it.
Hahaha. you beat me to it. it'd be funny if all core clusters could be activated by tracing with a pencil. XD
That makes no sense at all when looking at the diagram of epyc where EACH CPUdie has 64lanes. Not the least of which at this moment the delidding was likely an engineering model.
Hahaha. you beat me to it. it'd be funny if all core clusters could be activated by tracing with a pencil. XD
you should read more then before making statements as if you do in fact know. I don't have the discussions saved but take this one step further... how many lanes does EPYC have with 4 dies? 128... So how does your asinine theory hold up here? It doesn't. And again he is likely delidding an ES TR so doubtful TR has 4 dies with 2 disabled.Which diagram good sir?
Everything I have seen neither confirms nor negates anything I've postulated. Nothing I have read suggest each die has 64 lanes either.
The zen core actually has 32 pcie lanes, but not all are enabled. Some good info here https://forums.anandtech.com/threads/zeppelin-ryzen-i-o-capabilities.2511227/if the 1700-1800x has 24 PCI-E lanes, then to get 64, you would need 3 dual CCX at least? Stick a 4th one in and use only silicon that would have been for a 1400 with one failed memory channel and BOOM, you have a Thread Ripper. I'm thinking AMD is not getting the such high quality yields that their fanboys are raving about and this is a pretty smart-ass way to get top dollar for an otherwise crappy silicon. Anyways the proof in in the pudding. If we can measure PCI-E latencey, I'm guess it is going vary since some lanes will have to traverse the Infinity fabric.
you should read more then before making statements as if you do in fact know. I don't have the discussions saved but take this one step further... how many lanes does EPYC have with 4 dies? 128... So how does your asinine theory hold up here? It doesn't. And again he is likely delidding an ES TR so doubtful TR has 4 dies with 2 disabled.
The zen core actually has 32 pcie lanes, but not all are enabled. Some good info here https://forums.anandtech.com/threads/zeppelin-ryzen-i-o-capabilities.2511227/
I don't have every piece of data saved nor do I wish to hunt for the info. And as you finally see above you were in fact WRONG. You made an assumption and poised it as fact. I told you, you were wrong and lo and behold you were and what I stated was in fact correct.I simply asked for that diagram you referred to which you failed at. The single die or 2 CCX (R1200-1800X) are advertised as having 24 PCIE. You have provided no proof that each EPYC die has 64 pcie lanes per CCX pair to quantify any argument. The fact that the entire package has 128 can only imply each CCX pair has at least 32.
You can continue to argue with yourself if you want but thanks for nothing in the way of evidence or even a good theory but I enjoyed a good laugh.
The fact that lanes can be disabled is all i was basing my guess with.
if the 1700-1800x has 24 PCI-E lanes, then to get 64, you would need 3 dual CCX at least? Stick a 4th one in and use only silicon that would have been for a 1400 with one failed memory channel and BOOM, you have a Thread Ripper. I'm thinking AMD is not getting the such high quality yields that their fanboys are raving about and this is a pretty smart-ass way to get top dollar for an otherwise crappy silicon. Anyways the proof in in the pudding. If we can measure PCI-E latencey, I'm guess it is going vary since some lanes will have to traverse the Infinity fabric
That makes no sense at all when looking at the diagram of epyc where EACH CPUdie has 64lanes. Not the least of which at this moment the delidding was likely an engineering model.
Because intel has always being dicks.So are retail cpu's using solder? If so why is intel being so cheap using paste.
So if a Threadripper is a gimped Epyc, Does that mean it may be possible to use an Epyc on a Threadrippper MB?
Im betting that will be the case. No loss for AMD, plus im guessing you won't get the same warranty as in server parts.So if a Threadripper is a gimped Epyc, Does that mean it may be possible to use an Epyc on a Threadrippper MB?
really? He told you there are 32 per 4core not 8core. therefore 64 per 8core = CPUdie and there are 2 of them in TR @128 lanes with 64 reserved for communication between each.ROFL. You better tell it to your 64 PCIE lanes per die.
Makes me want to fire up the ol' Urban Terror Q3 mod.
I believe the technical term is "SPLOOSH".
You joke.... but we all want this to be true! I have fond memories of my old T-Bird, a magnifying glass, and a mechanical pencil. Makes me want to fire up the ol' Urban Terror Q3 mod.
Hey, that was a man's CPU. HEHE.Ah the T-Bird that you can cook an egg on? Or the T-Bird that would just die if your cooler or fan failed for any reason cause it didn't have thermal protections?
AMD requested he take it down. Says he may bring it back since AMD initially gave him permission to put it up in the first place.WTF video is gone? Mirror???
Urban Terror (4.3.1?) is a standalone game now. What if I told you I was -=UrF=- Dave.